Reviews of Plasma Physics 24

Reviews of Plasma Physics

The series Reviews of Plasma Physics, founded in 1963, covers complex problems
of modern theoretical plasma physics. It contains reviews presented by the leading
Russian researchers in this field.

Editor-in-Chief

Professor Vitalii Dmitrievich Shafranov

Kurchatov Institute
Nuclear Fusion Institute
Kurchatov sq. 1
123182 Moscow, Russia

Associate Editor Managing Editor

Dr. Oleg G. Bakunin Dr. Maria Aksentieva

Kurchatov Institute Physics-Uspekhi
Nuclear Fusion Institute Leninsky Prospekt 15
Kurchatov sq. 1 117071 Moscow, Russia
123182 Moscow, Russia E-mail: [email protected]
E-mail: [email protected]
Reviews of Plasma Physics

Volume 24

Contributions to this Volume:

V. Rozhansky
Mechanisma of Transverse Conductivity and Generation
of Self-consistent Electric Fields
in Strongly Ionized Magnetized Plasma

O.G. Bakunin
Correlations and Anomalous Transport Models
Contributors:
Dr. V. Rozhansky Dr. Oleg G. Bakunin
St. Petersburg State Polytecnical University Kurchatov Institute
Polytechnicheskaya 29 Nuclear Fusion Institute
195251 St. Petersburg, Russia Kurchatov sq. 1
E-mail: [email protected] 123182 Moscow, Russia
E-mail: [email protected]

ISSN 0080-2050 e-ISSN 1574-8952

ISBN 978-0-387-71088-4 e-ISBN 978-3-540-74576-1

Library of Congress ControlNumber: 2008927236

© Springer-Verlag Berlin Heidelberg 2008
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specif ically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,
reproduction onmicrofilm or in any other way, and storage in data banks.Duplication of this publication or
parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965,
in its current version, and permission for use must always be obtained from Springer. Violations are liable
to prosecution under the German Copyright Law.
The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply,
even in the absence of a specif ic statement, that such names are exempt from the relevant protective laws
and regulations and therefore free for general use.
Typesetting by VTEX using a Springer LATEX macro
Cover design: eStudio Calamar Steinen
SPIN 12024315 57/3180/vtex
Printed on acid-free paper
9876543210

springer.com
Preface

The “Reviews of Plasma Physics”, Vol. 24, contains two reviews. This book, as well
as the previous volumes, presents reviews for specialists interested in plasma physics
theory. In the review by V.A. Rozhansky, the problem of currents and self-consistent
electric fields in the magnetized fully ionized plasma is considered. The situation
in fully ionized plasma is completely different from that in solid state, gases, or
partially ionized plasmas, where the current density is proportional to the electric
field. In contrast, in a fully ionized magnetized plasma, a homogeneous electric field
causes drift both for electrons and ions, while a current in the direction of the electric
field is absent. A perpendicular current arises when the electric field is temporary or
spatially dependent. Various mechanisms, when the perpendicular currents are driven
by inertia, collisions with neutrals and different components of viscosity tensor are
analyzed in the review. Current systems in a vicinity of a biased electrode are studied
in connection with such applications as probes, unipolar arc formation, pellets, where
two-dimensional or three-dimensional currents systems determine plasma dynamics,
shielding properties, and current–voltage characteristics.
Since the end of 80th it became clear that self-consistent electric fields play the
key role in the formation of improved confinement regimes (H-regimes) in tokamaks
and stellarators. The poloidal drifts suppress the turbulence and hence the turbulent
transport coefficients so that the transport barriers with steep density and temperature
gradients and reduced transport coefficients are created. It is demonstrated that the
electric field structure is consistent with the experimental observations.
More complicated issues when a radial current is generated in fusion devices are
also considered. In a group of experiments performed on several tokamaks, an elec-
trode was installed into the plasma (inside the separatrix or last closed flux surface),
and the voltage was applied between this electrode and limiter or divertor plates. The
current–voltage characteristic of such a system, the value of the effective transverse
conductivity, and related problem of toroidal rotation generation in the plasma are an-
alyzed in the review. In the last few years, new methods of radial current generation
are widely discussed. One of them is connected with the formation of a stochastic
magnetic layer in a separatrix vicinity by special coils (resonance magnetic pertur-
bations). The aim is to suppress the edge localized modes, and this issue is rather
critical for the successful operation of the International Thermonuclear Experimen-
tal Reactor (ITER). A radial current of electrons in a stochastic magnetic field should
be accompanied by the same radial current of ions similar to the biasing experiments.
As a result, the radial electric field, density, and temperature profiles are significantly
vi Preface

modified, and edge modes are stabilized. The effective electron perpendicular con-
ductivity in a stochastic magnetic field and the ion perpendicular conductivity, which
are the key players in this situation, are also discussed in the review.
In the second review “Correlations and anomalous transport models” by
O.G. Bakunin, numerous aspects of turbulent transport are considered. This review
is intended to summarize the recent results from the multidisciplinary field of anom-
alous diffusion in turbulent plasma. A description of turbulent transport in the pres-
ence of coherent structures, convective rolls, zonal flows, and stochastic magnetic
fields is a very complex problem.
From the methodological point of view, this review focuses on the general use
of correlation estimates, quasilinear equations, and continuous-time random walk ap-
proach. The structure of some derivations, when they may be useful for more general
purposes, is given in detail. Thus, the review provides a fairly informative treatment
of seed diffusion effects in the framework of the correlation description. The rela-
tionship between Lagrangian and Eulerian correlation functions is discussed, and
the problem of relations between stochastic instability and transport effects in a sto-
chastic magnetic field is analyzed.
The author reviews in details the percolation approach to turbulent transport.
Both the monoscale representation and multiscale approach are considered. The re-
lationships between the transport and correlation exponents are derived. Nonlocal
and memory effects in the framework of the continuous-time random walk model
are treated. The kinetic (phase-space) approach describing ballistic modes of anom-
alous transport in complex systems is studied.
The topics to be discussed include renormalized quasi-linear equations, the
Levy–Khintchine distributions, and intermittency effects. The author focuses on scal-
ing arguments that play an important role in obtaining estimates of transport effects.
A careful analysis of more important results obtained in this field is presented in this
review.

Moscow, V.D. Shafranov

June 2008
Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

1 Mechanisma of Transverse Conductivity and Generation of

Self-Consistent Electric Fields in Strongly Ionized Magnetized Plasma
V. Rozhansky . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Conductivity Tensor in Partially Ionized Plasma . . . . . . . . . . . . . . . . . . . . . 4
1.3 Main Mechanisms of Perpendicular Conductivity in Fully Ionized
Plasma: Currents Caused by Viscosity, Inertia, Collisions with Neutrals,
and ∇B, and Mass-Loading Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.1 Inertia Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.2 Currents Caused by Ion–Neutral Collisions . . . . . . . . . . . . . . . . . . . 7
1.3.3 Diamagnetic Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.4 Viscosity-Driven Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.5 Mass-Loading Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Inertial (Polarization) and ∇B Currents. Acceleration of Plasma Clouds
in an Inhomogeneous Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Alfven Conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6 Perpendicular Viscosity, Radial Current, and Radial Electric Field in an
Infinite Cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.7 Current Systems in Front of a Biased Electrode (Flush-Mounted Probe)
and Spot of Emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.7.1 Viscosity-Driven Perpendicular Currents . . . . . . . . . . . . . . . . . . . . . 18
1.7.2 Currents Driven by Ion–Neutral Collisions . . . . . . . . . . . . . . . . . . . 22
1.7.3 Inertia Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.7.4 General Situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.7.5 Spot of Emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.8 Currents in the Vicinity of a Biased Electrode That is Smaller Than the
Ion Gyroradius . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.9 Neoclassical Perpendicular Conductivity in a Tokamak . . . . . . . . . . . . . . . 30
1.9.1 Steady State Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.9.2 Time-Dependent Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
1.10 Transverse Conductivity in a Reversed Field Pinch . . . . . . . . . . . . . . . . . . . 38
1.11 Modeling of Electric Field and Currents in the Tokamak Edge Plasma . . . 39
viii Contents

1.12 Mechanisms of Anomalous Perpendicular Viscosity and Viscosity-

Driven Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.13 Transverse Conductivity in a Stochastic Magnetic Field . . . . . . . . . . . . . . . 43
1.13.1 Nonstochastic Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.13.2 Stochastic Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
1.14 Electric Fields Generated in the Shielding Layer between Hot Plasma
and a Solid State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2 Correlations and Anomalous Transport Models

O.G. Bakunin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.2 Turbulent Diffusion and Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.2.1 The Correlation Function and the Taylor Diffusivity . . . . . . . . . . . 57
2.2.2 The Richardson Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.2.3 The Davydov Model of Turbulent Diffusion . . . . . . . . . . . . . . . . . . 60
2.2.4 The Batchelor Approximation for the Diffusion Coefficient . . . . . 62
2.3 Nonlocal Effects and Diffusion Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.3.1 The Functional Equation for Random Walks . . . . . . . . . . . . . . . . . . 64
2.3.2 Nonlocality and the Levy Distribution . . . . . . . . . . . . . . . . . . . . . . . 65
2.3.3 The Monin Fractional Differential Equation . . . . . . . . . . . . . . . . . . 67
2.4 The Corrsin Conjecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.4.1 The Corrsin Independence Hypothesis . . . . . . . . . . . . . . . . . . . . . . . 69
2.4.2 The Simplified Corrsin Conjecture . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.4.3 The Correlation Function and Scalings . . . . . . . . . . . . . . . . . . . . . . 71
2.5 Effects of Seed Diffusivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.5.1 Seed Diffusivity and Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.5.2 “Returns” and Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.5.3 The Stochastic Magnetic Field and Scalings . . . . . . . . . . . . . . . . . . 75
2.5.4 The Howells Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.6 The Diffusive Tracer Equation and Averaging . . . . . . . . . . . . . . . . . . . . . . . 79
2.6.1 The Taylor Shear Flow Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.6.2 Generalization of the Taylor Model . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.6.3 The Zeldovich Flow and the Kubo Number . . . . . . . . . . . . . . . . . . . 83
2.6.4 Advection and Zeldovich Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.7 The System of Random Shear Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.7.1 The Dreizin–Dykhne Superdiffusion Regime . . . . . . . . . . . . . . . . . 87
2.7.2 The Matheron–de Marsily Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.7.3 The “Manhattan Grid” Flow and Transport . . . . . . . . . . . . . . . . . . . 91
2.8 The Quasi-Linear Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.8.1 Quasi-Linear Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.8.2 Short-Range and Long-Range Correlations . . . . . . . . . . . . . . . . . . . 95
2.8.3 The Telegraph Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.8.4 Magnetic Diffusivity and the Kubo Number . . . . . . . . . . . . . . . . . . 97
2.9 The Diffusive Renormalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
 Contents ix

2.9.1 The Dupree Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

2.9.2 The Dupree Theory Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
2.9.3 The Taylor–McNamara Correlation Function . . . . . . . . . . . . . . . . . 104
2.9.4 The Kadomtsev–Pogutse Renormalization and the Stochastic
Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
2.10 Anomalous Transport and Convective Cells . . . . . . . . . . . . . . . . . . . . . . . . . 108
2.10.1 Bohm Scaling and Electric Field Fluctuations . . . . . . . . . . . . . . . . . 108
2.10.2 The Bohm Regime and Correlations . . . . . . . . . . . . . . . . . . . . . . . . . 110
2.10.3 Convective Cells and Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
2.10.4 Complex Structures and Convective Transport . . . . . . . . . . . . . . . . 114
2.11 Stochastic Instability and Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
2.11.1 Stochastic Instability and Correlations . . . . . . . . . . . . . . . . . . . . . . . 115
2.11.2 The Rechester–Rosenbluth Model . . . . . . . . . . . . . . . . . . . . . . . . . . 116
2.11.3 Collisional Effects and the Stix Formula . . . . . . . . . . . . . . . . . . . . . 119
2.11.4 The Quasi-Isotropic Stochastic Magnetic Field and Transport . . . 121
2.11.5 Quasi-Linear Scaling for the Stochastic Instability Increment . . . . 123
2.12 Fractal Conceptions and Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
2.12.1 Fractality and Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
2.12.2 The Richardson Law and Fractality . . . . . . . . . . . . . . . . . . . . . . . . . 127
2.12.3 Intermittency and the Kolmogorov Law . . . . . . . . . . . . . . . . . . . . . . 129
2.13 Percolation and Scalings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
2.13.1 Continuum Percolation and Transport . . . . . . . . . . . . . . . . . . . . . . . 130
2.13.2 Renormalization and Percolation . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
2.13.3 Graded Percolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
2.14 Percolation and Turbulent Transport Scalings . . . . . . . . . . . . . . . . . . . . . . . 135
2.14.1 Random Steady Flows and Seed Diffusivity . . . . . . . . . . . . . . . . . . 136
2.14.2 The Spatial Hierarchy of Scales and Stochastic Instability . . . . . . 137
2.14.3 Low Frequency Regimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
2.15 The Temporal Hierarchy of Scales and Correlations . . . . . . . . . . . . . . . . . . 140
2.15.1 The Spatial and Temporal Hierarchy of Scales . . . . . . . . . . . . . . . . 140
2.15.2 The Isichenko Intermediate Regime . . . . . . . . . . . . . . . . . . . . . . . . . 142
2.15.3 Dissipation and Percolation Transport . . . . . . . . . . . . . . . . . . . . . . . 143
2.16 The Stochastic Magnetic Field and Percolation Transport . . . . . . . . . . . . . 146
2.16.1 The Stochastic Magnetic Field and Percolation Transport . . . . . . . 146
2.16.2 Percolation and the Kadomtsev–Pogutse Scaling . . . . . . . . . . . . . . 148
2.16.3 Percolation Renormalization and the Stochastic Instability
Increment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
2.17 Percolation in Drift Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
2.17.1 Graded Percolation and Drift Flows . . . . . . . . . . . . . . . . . . . . . . . . . 151
2.17.2 Low Frequency Regimes and Drift Effects . . . . . . . . . . . . . . . . . . . 153
2.17.3 Compressibility and Percolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
2.18 Multiscale Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
2.18.1 The Nested Hierarchy of Scales and Drift Effects . . . . . . . . . . . . . . 157
2.18.2 The Brownian Landscape and Percolation . . . . . . . . . . . . . . . . . . . . 159
2.18.3 Correlations and Transport Scalings . . . . . . . . . . . . . . . . . . . . . . . . . 161
x Contents

2.18.4 The Diffusive Approximation and the Multiscale Model . . . . . . . . 163

2.18.5 Stochastic Instability and Time Scales . . . . . . . . . . . . . . . . . . . . . . . 165
2.18.6 Isotropic and Anisotropic Turbulent Energy Spectra . . . . . . . . . . . 166
2.18.7 The Multiscale Model of Transport in a Tangled Magnetic Field . 168
2.19 Subdiffusion and Traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
2.19.1 The Balagurov and Vaks Model of Diffusion with Traps . . . . . . . . 170
2.19.2 Subdiffusion and Fractality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
2.19.3 Comb Structures and Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
2.20 Continuous Time Random Walks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
2.20.1 The Montroll and Weiss Approach and Memory Effects . . . . . . . . 174
2.20.2 Fractional Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
2.20.3 The Taylor Definition and Memory Effects . . . . . . . . . . . . . . . . . . . 177
2.21 Fractional Differential Equations and Scalings . . . . . . . . . . . . . . . . . . . . . . 180
2.21.1 The Klafter, Blumen, and Shlesinger Approximation . . . . . . . . . . . 180
2.21.2 The Stochastic Magnetic Field and Balescu Approach . . . . . . . . . . 181
2.21.3 Longitudinal Correlations and the Diffusive Approximation . . . . . 183
2.21.4 Vortex Structures and Trapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
2.21.5 Correlations and Trapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
2.22 Correlation and Phase-Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
2.22.1 The Corrsin Conjecture and Phase-Space . . . . . . . . . . . . . . . . . . . . 189
2.22.2 The Hamiltonian Nature of the Universal Hurst Exponent . . . . . . . 191
2.22.3 The One-Flight Model and Transport . . . . . . . . . . . . . . . . . . . . . . . . 193
2.22.4 Correlations and Nonlocal Velocity Distribution . . . . . . . . . . . . . . . 195
2.22.5 The Arrhenius Law and Phase-Space Distribution . . . . . . . . . . . . . 197
2.23 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
1
Mechanisma of Transverse Conductivity
and Generation of Self-Consistent Electric Fields
in Strongly Ionized Magnetized Plasma
V. Rozhansky

1.1 Introduction

In many areas of plasma physics a problem of currents flowing perpendicular to a

magnetic field in the presence of an electric field arises. This is also typical for fully
ionized plasmas. However, the situation in fully ionized plasma is completely dif-
ferent from that in a solid state, gases, or partially ionized plasmas, where current is
simply proportional to the applied electric field and the corresponding conductivity
is determined by the collisions between charged particles and neutrals. In fully ion-
ized plasmas, a homogeneous electric field causes E × B drift both for electrons and
ions perpendicular to the electric field and no current in the direction of the electric
field is generated.
On the other hand, perpendicular current is observed in many experiments.
Among the examples are the so-called biasing experiments in tokamaks and other
magnetic traps, where an electrode is installed into the plasma [inside the separatrix
or last closed flux surface (LCFS)] and the voltage is applied between this electrode
and the limiter or divertor plate. The whole flux surface is biased to the same po-
tential due to extremely large parallel conductivity and the radial current flows from
the core through the LCFS to another electrode (limiter). The current–voltage char-
acteristic, which is measured in these experiments, corresponds to effective plasma
resistance on the order of 1 ; such resistance was observed in experiments on CCT,
TEXTOR, Tuman-3M, and other machines. So effective perpendicular conductivity
exists and should be explained theoretically.
This issue is closely connected with the problem of the self-consistent radial
electric field in tokamaks and other fusion devices since the self-consistent electric
field corresponds to the particular point of the current–voltage characteristic, i.e.,
to the condition of zero net radial current flowing through the flux surfaces. And
2 V. Rozhansky

according to the modern conception the self-consistent electric field is a key element
in the formation of transport barriers in the core and edge tokamak plasmas and in
the transition to high confinement regimes.
At present, a theoretical understanding of the mechanisms responsible for the
formation of the radial electric field in a tokamak and the radial current in the bias-
ing experiments has been reached. The radial current in the presence of a poloidal
component of the magnetic field accelerates plasma in the toroidal direction. The ac-
celeration force is balanced by the radial transport of the toroidal momentum, which
to a large extent is determined by turbulent (anomalous) mechanisms. Fortunately,
from the parallel momentum balance the force associated with radial transport of
toroidal momentum may be expressed through the parallel viscosity, which is con-
nected, as known from neoclassical theory, with the radial electric field. As a result,
in some important cases the radial effective conductivity, as well as the radial elec-
tric field in the absence of current, may be expressed through the pure classical or
neoclassical viscosity coefficients.
The calculation of radial current and the self-consistent electric field in a cylinder
whose axis is parallel to the magnetic field is significantly simpler than for a tokamak.
Here, the first estimate of the self-consistent electric field has been done in the review
by Braginskii; however, the case of nonuniform temperature adds complexity and the
final result is not widely known in the thermonuclear community.
2D current systems in fully ionized plasma, where current flows both along
and across the magnetic field, are important in many applications. Among them are
probes in the magnetic field, which are used to measure local plasma parameters, in
particular flush mounted probes in the divertor region of a tokamak. Their current–
voltage characteristics, especially their slope, are determined by the mechanisms of
the perpendicular conductivity. Similar 2D current systems in the vicinity of unipo-
lar arcs are responsible for the ignition of the arcs and determine the dynamic of the
arc, the value of the return current to the electrode, and other parameters. This area of
plasmaphysics, in spite of its importance, is not very well developed and only several
publications on the subject exist.
The 2D problem of currents and self-consistent electric fields is typical for edge
tokamak plasma, including the divertor region, scrape-off layer, and vicinity of a
separatrix. In recent years, numerical 2D fluid codes have been developed, where
the self-consistent electric field is treated in detail and corresponding plasma fluxes
across and along the magnetic field, caused by this field, have been analyzed. The
structure of the electric field in the scrape-off layer is responsible for plasma flows to
the divertor plates, while the radial field near the separatrix determines the transition
to the regime of improved confinement. The theoretical progress in understanding
these important issues is also the subject of this review.
Situations where perpendicular currents in fully ionized plasmas determine the
whole dynamic are too diverse to consider in one review. Therefore, we tried to fo-
cus on the basic mechanisms and principles. In addition to those mentioned above,
considered are polarization currents in the Alfven wave, which might often act as
an effective wave conductivity, currents in a braided magnetic field, and few other
typical cases. The main characteristic and practically important situations in strongly
 1 Mechanisma of Transverse Conductivity 3

ionized plasma in the homogeneous magnetic field are considered, as well as more
complicated effects in the inhomogeneous magnetic field, which are typical for mag-
netic traps.
Conductivity in gases or solids is a tensor, which links electric field and current
density. In partially ionized plasmas this tensor can be easily calculated, and ex-
pressions for conductivity are widely used in many applications. In contrast, in fully
ionized magnetized plasmas a homogeneous electric field causes E × B drift of elec-
trons and ions with a common velocity. Such electric field does not create current
at all. Therefore, in fully ionized plasma in the steady state case the current arises
only along the inhomogeneous electric field (there is also polarization current, which
is proportional to the temporal derivative of the electric field). The inhomogeneous
E × B drifts produce different types of forces associated with inertia and viscosity,
which in turn cause the current in the direction of the electric field. As a result, the
current density in the steady state case is not directly proportional to the electric field
itself but to its spatial derivatives. Hence, generally speaking, in the fully ionized
plasma it is impossible to introduce the conductivity tensor as in the partially ionized
plasma. One can talk only about an “effective” conductivity, which can be introduced
analogously, using dimensional arguments.
In the presence of a strong turbulence (the situation typical for many plasma
devices), the transport coefficients (diffusion, heat conductivity, and viscosity) be-
come anomalous. There are many phenomenological approaches, in which an ad hoc
anomalous conductivity is also introduced to explain experimentally observed trans-
verse currents. However, due to the ambipolar character of plasma motion, which
is connected with momentum conservation in Coulomb collisions, such anomalous
conductivity does not exist. It can be shown rigorously that the only remaining cur-
rents are those controlled by inertia and viscosity with anomalous values of diffusion
and viscosity coefficients.
The transverse currents in fully ionized plasma are rather small, since they con-
tain a small parameter: the ratio of ion gyroradius to the plasma spatial scale. How-
ever, they are extremely important since the condition ∇ · j = 0 determines the dis-
tribution of self-consistent electric fields in the plasma and the corresponding E × B
drifts, which to a large extent control plasma dynamics. The problems of calculat-
ing these perpendicular currents in strongly ionized plasmas, closing currents in the
plasma, and generating self-consistent electric fields is analyzed in detail below.
In Sect. 1.2 a conductivity tensor in partially ionized plasma is derived and Ein-
stein relations between mobility and diffusion tensors are discussed. In Sect. 1.3
an expression for the perpendicular current is derived from the momentum balance
equations, and the contrast between the fully and partially ionized plasmas is em-
phasized. In Sect. 1.4 the effects of plasma acceleration, which are closely connected
with the combination of polarization and ∇B currents, are considered. Acceleration
of plasma clouds in space and in magnetic traps is discussed. The phenomenon of
Alfven conductivity, where the polarization currents in the Alfven wave close a cur-
rent circuit and compensate current inside the plasma inhomogeneity, is analyzed
in Sect. 1.5. In Sect. 1.6 the simplest 1D situation of a plasma cylinder in the ho-
mogeneous magnetic field where the perpendicular viscosity determines the radial
4 V. Rozhansky

current is considered. Expressions for effective conductivity and for the ambipolar
electric field, which corresponds to zero radial current, are derived. In Sect. 1.7 a
more complicated 2D problem of the current systems in the vicinity of a biased
electrode or spot of emission is studied. It is demonstrated that the current–voltage
characteristics and potential and current distribution in the plasma is controlled by
different mechanisms of effective transverse conductivity. Application to the prob-
lems of interpretation of the characteristics of flush-mounted probes and unipolar
arc formation is analyzed. The specific question of currents closing when the per-
pendicular scale of the electrode is smaller than the ion gyroradius is studied in
Sect. 1.8. Section 1.9 is devoted to the complicated problem of transverse currents
in a tokamak. It is demonstrated that the parallel component of a viscosity tensor
and the radial transport of toroidal momentum control transverse current. Interpre-
tation of tokamak biasing experiments is also given. A similar analysis for reversed
field pinch is presented in Sect. 1.10. In Sect. 1.11 the problem of electric fields
and currents in the tokamak edge plasma is studied numerically, and the results of
different transport codes are analyzed both for core and divertor plasmas. The key
parameters which are responsible for the transition into the improved confinement
regime (L–H transition) are investigated. A comparison with tokamak experiments
is also presented in this section. Section 1.12 is devoted to the analysis of the per-
pendicular anomalous viscosity and corresponding transverse currents, which are
generated by an electrostatic turbulence. The absence of ad hoc anomalous current is
demonstrated. A quasilinear expression for the anomalous viscosity coefficient and
effective perpendicular conductivity is analyzed for specific types of instabilities.
The transverse conductivity in a braided magnetic field is calculated in Sect. 1.13.
Here, the self-consistent electric field in systems with a stochastic magnetic field is
also analyzed. In Sect. 1.14, electric fields which are generated in a shielding layer
between a hot plasma and a solid state are considered. The plasma drifts there may
significantly change the shielding properties of the cold ablated plasma, and may
increase the ablation rate of the surface during such events as hard disruptions in a
tokamak.

1.2 Conductivity Tensor in Partially Ionized Plasma

Consider magnetized partially ionized plasma with a homogeneous magnetic field
B parallel to the z-axis and an electric field E parallel to the y-axis, Fig. 1.1. Let
us assume for simplicity that cyclotron frequencies ωce , ωci of electrons and ions
are much higher than the corresponding collision frequencies between the charged
particles and neutrals νeN , νiN . In the strong magnetic field both electrons and ions
drift in the crossed E × B fields in the x (Hall) direction with almost equal velocities
uex = uix = E/B and the current in the Hall direction is negligible. Collisions with
neutrals produce friction forces Rαx = −μαN nναN uαx in the reference frame where
neutrals are at rest [α corresponds to the ions and electrons, respectively, μαN =
mα mN /(mα +mN )]. These friction forces in their turn cause the drift of electrons and
ions in the y direction, i.e., along the electric field, with the velocity uαy = ±Rαx /eB
 1 Mechanisma of Transverse Conductivity 5

Fig. 1.1.

(the charge number of the ions is put to unity for simplicity). Finally,
eναN
uαy = ±bα⊥ E; bα⊥ = 2
. (1.1)
μαN ωcα
Here, ωcα = eB/μαN . The coefficient bα⊥ is the mobility of a particle perpendicular
to the magnetic field. Since the ion mobility is much larger than that of electrons, the
conductivity is determined by the ion mobility:
ne2 νiN

j = σiN E; σiN = 2
. (1.2)
μiN ωci
The general expression for the mobility tensor for an arbitrary ratio between cy-
clotron and collision frequencies can be found in [1]. In the so-called elementary ap-
proximation, when collision frequencies are supposed to be independent of charged
particle velocities, the mobility tensor has the form
⎛ ⎞

bα⊥ ±bα∧ 0
b α = ⎝ ∓bα∧ bα⊥ 0 ⎠, (1.3)
0 0 bα||
where
e eναN eωcα
bα|| = , bα⊥ = , bα∧ = .
mα ναN 2 + ν2 )
mα (ωcα 2 + ν2 )
mα (ωcα
αN αN
The conductivity tensor is
  
σ = ne(b i − b e ). (1.4)
In the partially ionized plasmas, the mobility tensor is linked to the diffusion tensor
by an Einstein relation [1]. In particular,
e
bα⊥ = Dα⊥ , (1.5)
Tα
where Dα⊥ is the diffusion coefficient perpendicular to the magnetic field.
6 V. Rozhansky

1.3 Main Mechanisms of Perpendicular Conductivity in Fully

Ionized Plasma: Currents Caused by Viscosity, Inertia,
Collisions with Neutrals, and ∇B, and Mass-Loading
Currents
If a constant homogeneous electric field is applied perpendicular to B in fully ionized
magnetized plasmas, the transverse conductivity in a steady state is zero, in contrast
to the situation discussed in the previous section. Indeed, since the drift velocities
of electrons and ions in the E × B direction coincide, the friction force between
electrons and ions is absent and hence no particle fluxes and no currents along the
electric field are generated. This result can easily be obtained in the reference frame
moving with the drift velocity E/B: in this reference frame the electric field is absent
due to the Lorentz transformation, and hence the current is also absent. Therefore,
to obtain a finite transverse current in fully ionized plasmas in a steady state it is
necessary to consider a nonuniform electric field or to take into account collisions
with neutrals.
There have been several publications, in which the authors introduced an ad
hoc anomalous conductivity for turbulent plasma in a manner similar to the case
of slightly ionized plasma,

j = σ⊥ E = neb⊥ E = ne2 (D/T )E,

where D is the anomalous diffusion coefficient. However, it is clear that this expres-
sion in the fully ionized plasma is erroneous, independent of the question of whether
anomalous transport is important or not. Indeed, in fully ionized plasmas, by chang-
ing the reference frame to that moving with E × B velocity, one can cancel the
homogeneous electric field. Consequently, perpendicular current, which is propor-
tional to the electric field itself, does not exist, and only current that is caused by the
temporal or spatial derivatives of the electric field remains.
To obtain the expression for the perpendicular current we start with the momen-
tum balance equation, which is the sum of the momentum balance equations for the
ions and electrons:
dn ui
= −∇p − ∇ · π + RiN + [j × B]
 + S(M) .
↔
mi (1.6)
dt

Here, p = pi + pe is the total pressure, RiN is the ion–neutral friction force, and
S(M) are other momentum sources and sinks. Resolving the perpendicular current
from this equation, we find
 
dn ui  2
j = B × mi B + [B × RiN ]/B 2 + [B × ∇p]/B 2
dt

+ [B × ∇ · π]/B 2 + B × S(M) /B 2 .

↔
(1.7)
 1 Mechanisma of Transverse Conductivity 7

1.3.1 Inertia Currents

The fourth term on the r.h.s. of (1.7) is a current caused by ion motion due to inertia
forces, which we shall denote as j(in) . The corresponding velocities are usually much
smaller than the velocities of the diamagnetic and E × B drifts, with small parameter
proportional to the ratio of the ion gyroradius to the plasma spatial scale, and the
ratio of inverse ion cyclotron frequency proportional to the time scale. Velocities as-
sociated with other currents in (1.7), i.e., with viscosity-driven currents and currents
driven by ion–neutral collisions, are also small. Therefore, in the d ui /dt term it is
possible to keep only the diamagnetic and E × B drifts and the parallel velocities.
The exception is the drift caused by gyroviscosity, which is approximately the same
as diamagnetic drift. In turbulent plasmas anomalous diffusive flux should also be
taken into account, since anomalous diffusion represents averaged E × B drifts in
stochastic electric fields. Thus, the expression for the ion velocity to be inserted into
the d ui /dt term is
(dia)
ui ≈ ui|| + ui + u(E) + u(D) + u(vis) , (1.8)
where
(dia) [B × ∇pi ] (E) [E × B]

ui = ; ui = ;
enB 2 B 2
(1.9)
[B × ∇ · π i ]
↔
(D) (vis)
ui = −D∇ ln n; ui =.
enB 2
Here, we use the simplest expression for anomalous diffusive flux. More detailed
expressions will be given below for special cases. In the homogeneous magnetic
field, the expression for the inertia current can thus be reduced to the form
 
 mi ∂(nE − ∇pi /e) (0) 
j (in)
= 2 + ui ∇(nE − ∇pi /e) , (1.10)
B ∂t
where
(0) (E) (D)
ui = ui + ui .
Here, the Braginskii expression for gyroviscosity [2] is used to cancel the contribu-
tion from the diamagnetic drift. For a more sophisticated expression in the inhomo-
geneous magnetic field see Sect. 1.11.

1.3.2 Currents Caused by Ion–Neutral Collisions

These currents might be important in strongly ionized plasmas in spite of the small
density of neutral particles because other currents are also rather small. If neutral
gas velocity may be neglected with respect to the ion velocity we have the same
expression for the perpendicular mobility as in partially ionized plasmas, (1.1). In
general, when the neutral gas velocity is taken into account, inserting expressions for
the perpendicular ion velocity, (1.8) and (1.10), into the perpendicular friction force,
we find
8 V. Rozhansky
 
j(iN) = σiN E + [  − σiN ∇pi /en
uN × B] (1.11)
for the second term on the r.h.s. of (1.7), where σiN is the ion–neutral perpendicular
conductivity, (1.2). For the important case of charge exchange collisions [1],
nmi ViN σex
nN
σiN = .
2B 2

1.3.3 Diamagnetic Currents

The third term on the r.h.s. of (1.7) is a diamagnetic current, which is almost di-
vergence free. Indeed, its divergence is proportional to ∇B and equals zero in a
homogeneous magnetic field. Sometimes, besides the diamagnetic current

j(dia) = [B × ∇p]/B 2 , (1.12)

it is convenient to introduce the divergent part of the diamagnetic current [2]

  
[B × ∇B] B × (B∇)(
 
j˜(dia) = p
B/B)
+ , (1.13)
B3 B3
which produces the same divergence as the diamagnetic current. This current corre-
sponds to the guiding center drifts of charged particles in inhomogeneous magnetic
fields and is responsible for plasma polarization. Other currents should balance the
divergence of the diamagnetic current to provide quasineutrality. The corresponding
example is discussed in the next section.

1.3.4 Viscosity-Driven Currents

The fourth term on the r.h.s. of (1.7) represents the viscosity-driven current:

j(vis) = [B × ∇ · π]/B 2 .

↔
(1.14)

In various situations this current can be caused by different components of ion vis-
cosity tensor: by perpendicular viscosity, by gyroviscosity, or by parallel viscosity.
Electron viscosity is much smaller and may usually be neglected.
The simplest case in which the perpendicular viscosity causes a current is illus-
trated by Fig. 1.2. Here, an inhomogeneous electric field is applied in the y direction
perpendicular to the magnetic field. This field generates inhomogeneous E × B drifts
in the x-direction u(E)
ix = Ey (y)/B. A similar contribution form ion pressure gradi-
ent comes from diamagnetic velocity uix = −(∂pi /∂y)(enB)−1 . Inhomogeneous
(dia)

velocities produce viscous force in the x-direction [2]:

↔ ∂ ∂uix
−(∇ · π )x = η1 , (1.15)
∂y ∂y
 1 Mechanisma of Transverse Conductivity 9

Fig. 1.2.

where η1 = 0.3nTi νi /ωci2 is the classical perpendicular viscosity coefficient. In tur-

bulent plasmas the perpendicular viscosity coefficient similar to the diffusion and
heat conductivity coefficients may be anomalous. The connection between this coef-
ficient and the spectrum of turbulent oscillations is analyzed in Sect. 1.12. However,
for applications one can simply use the anomalous value η for the viscosity coeffi-
cient instead of the classical value η1 , where roughly η = nmi D, with D being the
anomalous diffusion coefficient.
Combining (1.14) and (1.15), we obtain

1 ∂ ∂ ∂pi
(vis)
jy = − 2 η1 Ey − . (1.16)
B ∂y ∂y en∂y
(vis) (vis)
The corresponding particle flux Γy = jy /e is rather small. Indeed, assuming
Ey ∼ T /eL, where L is the spatial scale, for the case of anomalous transport we
(vis) (D)
have Γy /Γy ∼ ρci2 /L2 (ρci is the ion gyroradius), i.e., anomalous viscous flux
is much smaller than the anomalous diffusive flux. In the classical case the situation
is quite different: this ratio is (mi /me )1/2 times larger, since classical diffusive flux
is caused by the electron–ion collisions and the classical viscosity-driven current is
proportional to the ion–ion collision frequency.
In the classical case (1.16) is strictly valid only for constant ion temperature.
The reason consists in the fact that in general the ion temperature gradients and the
corresponding heat fluxes cause additional viscosity (thermostress terms) [3], which
in turn produce additional current proportional to the ion temperature gradient. We
shall not consider this effect in detail and only present some results for special cases,
see below.
Gyroviscosity terms in some cases may be combined with inertial terms, the
corresponding example considered in Sect. 1.11. In many situations gyroviscosity
produce divergence-free currents.
The contribution of the parallel viscosity to the perpendicular current is rather
important. It is well known that the parallel viscosity is responsible for the differ-
10 V. Rozhansky

ence between the perpendicular and parallel pressures. Physically, the perpendic-
ular pressure should contribute to the diamagnetic current in (1.12) instead of the
full pressure. Therefore, the viscosity-driven current j(vis) = [B × (∇ · π || )]/B 2
↔

should be added to the diamagnetic current in (1.12), and their sum represents the
current caused by the perpendicular pressure. Furthermore, similarly to the diamag-
netic current, current caused by the parallel viscosity is almost divergence-free. One
can check this using the Braginskii expression for the parallel viscosity in a homo-
geneous magnetic field parallel to the z-axis:

∂uix ∂uiy 2
π||xx = π||yy = −η0 + − ∇ · ui , (1.17)
∂x ∂y 3

where η0 = 0.96nTi /νi is the parallel viscosity coefficient. Inserting (1.17) into
(1.14) and taking the divergence, one obtains zero for the constant magnetic field. In
the more complicated geometry it is convenient to use the following expression for
the parallel viscosity tensor:

Bα Bβ 1
π||αβ = (p|| − p⊥ ) − δαβ . (1.18)
B B 3

This expression can be applied not only in the fluid case when the value (p|| −
p⊥ ) is known from the Braginskii expression, but also in low collisionality regimes
when the viscosity tensor must be calculated using a kinetic approach. Analogously
to the diamagnetic current, (1.13), in the inhomogeneous magnetic field it is con-
venient to introduce a divergent part of the current driven by the parallel viscos-
ity
π|| [B × ∇B] B × (B∇)(
 
j˜(vis||) = −
B/B)
3
+ π|| 3
, (1.19)
2 B B
where π|| = 2/3(p|| − p⊥ ).

1.3.5 Mass-Loading Current

This current, which corresponds to the last term on the r.h.s. of (1.7), arises in the
presence of a plasma source, for example in the ionization process, when the ionized
particles move due to E × B drift. Injected ions or ions which are created during
ionization do not have momentum in the E × B direction at first. They start to move
over a cycloid in the crossed fields, and their average shift in the E direction is given
by the cycloid radius E/(ωci B). Therefore, the mass-loading current is proportional
to the ion source intensity S [4, 5] (S is the number of ionized particles created per
second):
nmi S 
j (ml) = E. (1.20)
B2
This current is important in space physics and in some situations in laboratory plas-
mas.
 1 Mechanisma of Transverse Conductivity 11

1.4 Inertial (Polarization) and ∇B Currents. Acceleration

of Plasma Clouds in an Inhomogeneous Magnetic Field
Diamagnetic currents in an inhomogeneous magnetic field can produce strong polar-
ization of plasma. Let us start with the classical example of acceleration of a plasma
cloud by ∇B [6]. Consider a plasma cloud in the xy plane in a magnetic field parallel
to the z-axis, which decreases in the x direction: B = B0 (1 − x/R). The magnetic
pressure is assumed to be much larger than that of the plasma so that the magnetic
field is not perturbed by the plasma motion. Since the cloud is restricted in the z di-
rection, the Boltzmann potential on the order of Te /e should arise to prevent electrons
from running along B.  We shall neglect this potential with respect to that caused by
the ∇B current.
The physical picture of the acceleration is quite simple. The divergent part of the
diamagnetic current in (1.13), which is responsible for polarization of the cloud, is
directed vertically:
n(Te + Ti )
j˜(dia) = − . (1.21)
BR
This current creates an electric field, which can roughly be considered to be vertical
and homogeneous. The inertial current produced by this field is
nmi ∂Ey
j (in) = . (1.22)
B 2 ∂t
This current compensates the vertical ∇B current and from the condition j (in) =
−j˜(dia) it follows that the vertical electric field increases with time:

Bcs2
Ey = t, (1.23)
R
√
where cs = (Te + Ti )/mi is the speed of sound. The corresponding E × B velocity
in the x direction also grows linearly with time:

ux = gt; x = gt 2 /2; g = cs2 /R. (1.24)

Therefore, a simple estimate predicts acceleration of the plasma cloud towards the
low field side.
However, this picture is oversimplified. In reality, charged particles move along
equipotentials. Since there exist two potential extrema, plasma at the top and at the
bottom of the cloud move in the negative x direction opposite to the center of the
cloud and cloud acceleration is accompanied by its deformation. To describe this it
is necessary to solve current continuity equation div j = 0. This has been done in
[7]. For the inertial current, (1.10) is to be used, where the pressure gradient contri-
bution is neglected with respect to the electric field term. After integrating the current
continuity equation along the magnetic field we find [7]
 
∂ϕ ∂∇ϕ ∂ ln N
+ J (ϕ, ϕ) + + J (ϕ, ∇ϕ) ∇ ln N + = 0, (1.25)
∂t ∂t ∂y
12 V. Rozhansky

Fig. 1.3. a The isodensities for initially Gaussian cloud N ∼ exp[−(x 2 + y 2 )/a 2 ] at the
moment t = 4(aR)1/2 /cs . Coordinates are in the units of a; b The isodensities for the case
of a constant plasma source Q ∼ exp[−(x 2 + y 2 )/a 2 ], which was switched on at t = 0, at
t = 5(aR)1/2 /cs

∂β ∂β ∂α
where J (α, β) = ∂α ∂x∂y − ∂x ∂y and N is the cloud density integrated along the
∞
magnetic field, N = −∞ n dz. The continuity equation for the particles is reduced
to the form
∂N
+ J (ϕ, N ) = 0. (1.26)
∂t
Results of the numerical simulation of (1.25) and (1.26) with the boundary conditions
N → 0, ϕ → 0 at infinity are shown in Fig. 1.3a [7]. The initial cloud moves in the
x direction with an acceleration on the order of g, (1.24). The front of the cloud
becomes much steeper because the vertical electric field and corresponding drift in
the x direction are larger in the center of the cloud and smaller at its front. The
sides of the cloud are left behind since particles rotate around potential extrema. The
typical shape of the cloud resembles the cap of a mushroom.
The evolution of the plasma created by a constant plasma source is shown in
Fig. 1.3b. To obtain this result one has to add particle source Q to the r.h.s. of particle
continuity equation (1.26) and the term −N −1 ∇ · (Q∇ϕ) = 0 to the r.h.s. of current
continuity equation (1.25). The shape of the plasma cloud resembles a mushroom.
Plasma density in the vicinity of the source and in the stem remains approximately
 1 Mechanisma of Transverse Conductivity 13

constant. Here, the vertical ∇B current is balanced by the stationary part of the iner-
tial current [the second term on the l.h.s. of (1.10)]. Hence, the charged particles are
spatially accelerated in the inhomogeneous electric field and equipotentials converge
along x in the stem region. As a result, the stem becomes thinner for larger values
of x, while the vertical electric field becomes stronger. In the region of the cap, where
the electric field is the largest and increases with time, plasma evolution is similar
to that of a single cloud, so that the cap accelerates along x according to (1.24). At
later stages the cloud is split into separate striations [7]. Such processes (albeit more
complicated) are responsible for the acceleration of the evaporated plasma during
pellet injection into a tokamak. Note that in tokamak geometry, due to the curvature
of the magnetic field line, the nondivergent free part of the diamagnetic current is
2n(Te + Ti )
j˜(dia) = −
BR
instead of that given by (1.21). Hence, g = 2cs2 /R in (1.24). Plasma acceleration in
the tokamaks, predicted in [7], has been observed in several pellet experiments on
ASDEX-Upgrade, DIII–D, and others [8–11]. Density perturbations (blobs), which
have a similar mushroom shape, are also discussed as candidates for enhanced trans-
port in the edge plasma of tokamaks [12, 13].

1.5 Alfven Conductivity

In the previous case we neglected currents in the ambient plasma surrounding a
cloud. In reality, polarization of the cloud, which is caused by ∇B drifts or by cloud
motion across a magnetic field, propagates along magnetic field lines with Alfven
velocity. Inertial (polarization) currents of the Alfven wave flow perpendicular to the
magnetic field and, together with parallel currents, close the currents inside the cloud.
Parameters of the ambient plasma may be characterized by the so-called Alfven con-
ductivity, which determines the ability of the ambient plasma to reduce polarization
of the cloud and thus to reduce its E × B velocity.
This effect can be illustrated by considering deceleration of plasma inhomogene-
ity, which is infinite in the x direction and has initial velocity u0 in this direction with
respect to the ambient plasma [14], Fig. 1.4. Initial velocity u0 is caused by E × B
drift of the cloud in the x direction: u0 = [E0 × B]/B 2 . The kinetic energy of the

cloud is assumed to be small with respect to the magnetic energy so that the mag-
netic field is only slightly perturbed. Polarization of the cloud tends to spread along
the magnetic field lines, thus creating an electric field in places where the ambient
plasma was initially at rest. Hence, momentum is transferred from the cloud to the
ambient plasma. If the skin time τs = L2y σ μ0 , where σ is the Spitzer conductivity
and Ly is the cloud size in the y direction, is large with respect to the typical time-
scale of the problem, the polarization and the plasma drift velocity are spread along
the magnetic field as an Alfven wave:
∂ 2 ux 2
2 ∂ ux B2
= c A ; 2
cA = . (1.27)
∂t 2 ∂z2 μ0 mi n
14 V. Rozhansky

Fig. 1.4. Cloud (infinite in the x direction) in the zx (a) and zy (b) planes

The cloud itself also expands along B with a velocity on the order of the speed of
sound. Since the Alfven wave is assumed to propagate over a much larger distance
LA = cA t, it is possible to multiply (1.27) by mi n and integrate it along the magnetic
field to the distance ±δ, which is much larger than the cloud size but is much smaller
than LA . Considering ux as constant under the integral and taking into account the
symmetry over z, after neglecting δ with respect to LA we obtain
 
∂ 2 ux  B 2 ∂ux 
M 2  = , (1.28)
∂t z=0 μ0 ∂z z=0
∞
where M = −∞ nmi dz is the integrated cloud mass.
Equation (1.28) can be used as a boundary condition for (1.27). The solution for
z ≥0 has the form
 
2B 2
ux = u0 exp 2
(z − c A t) , z ≤ cA t;
μ0 McA (1.29)
ux = 0, z ≥ cA t,

where the Alfven velocity is calculated for the ambient plasma. The velocity of the
cloud at z = 0 decreases exponentially with the time scale τA = μ0 McA /2B 2 . On
the other hand, the distance over which the ambient plasma moves in the x direction
increases as cA t.
Since the electric field in the cloud and in the ambient plasma up to distances
z = ±cA t decreases with time, there is a negative inertial current flowing in the y
direction. The net negative current (for positive z) is obtained by integrating along
the magnetic field:
 cA t
nmi ∂Ey 2
I = Ic + I0 = 2 ∂t
dz = − Ey ,
0 B μ 0 cA
 1 Mechanisma of Transverse Conductivity 15

where Ey ≡ Ey (z = 0) = u0 B exp(−t/τA ). There are two equal contributions to

the current I . The first one, Ic , arises from integrating over the cloud itself, when Ey
is taken at z = 0 and the integral over n(z)mi gives M/2. This contribution corre-
sponds to the current flowing inside the cloud. The second contribution, I0 = Ic , is
obtained by integrating the inertial current in the ambient plasma with constant den-
sity and electric field, which is determined by (1.29). This part of the current flows
in the region 0 ≤ z ≤ cA t. From the current continuity equation it follows that the
net current perpendicular to the magnetic field should be zero. Consequently, there
should exist a third positive contribution to cancel current I . This current IF = −I
flows at the front of the Alfven wave, where the plasma is accelerated from zero
velocity up to u0 . Finally, we separate zero net transverse current into two different
parts: the negative current of the cloud Ic = −Ey /μ0 cA and the positive net current
of the ambient plasma IW = I0 + IF = Ey /μ0 cA . The latter can be rewritten in the
form
1
IW = ΣW E y ; ΣW = , (1.30)
μ0 cA
where the net conductivity ΣW is called the Alfven conductivity.
In other words, ambient plasma with a propagating Alfven wave may be replaced
by a simple resistance determined by (1.30) (one has to keep in mind that the same
current flows on the other side of the cloud at negative z). The Alfven conductivity
ΣW may be used in many other problems to determine electric fields and current
systems inside plasma clouds in the ionosphere and magnetosphere [14] and during
pellet injection in tokamaks [7, 15, 16].

1.6 Perpendicular Viscosity, Radial Current, and Radial Electric

Field in an Infinite Cylinder
In this section we demonstrate how the self-consistent electric field is formed in the
simplest geometry. Consider a cylinder of low β fully ionized plasma infinite in the
z direction, which is parallel to a homogeneous magnetic field. Neglecting viscosity,
one can obtain from the perpendicular components of the momentum balance equa-
tions for electrons and ions the well-known classical result [2, 3], radial diffusion of
a plasma column with the velocity

dn D 1 dTe dTi
ur = −D
(D)
+ − . (1.31)
n dr Te + Ti 2 dr dr
This radial diffusion and thermodiffusion are often known as “automatically am-
bipolar” processes, which means that the radial velocities of electrons and ions are
equal to each other and are given by (1.31) independently of the radial electric field.
This result can be understood if the radial velocities of the charged particles are con-
sidered as drifts produced by azimuth friction forces, which are caused by azimuth
particle diamagnetic fluxes and a radial temperature gradient (thermal force). Since
friction forces acting on ions and electrons have the same value and different signs,
they produce the same drift for electrons and ions in the radial direction.
16 V. Rozhansky

In other words, the ambipolar character of the classical diffusion in the absence
of viscosity is the result of momentum conservation during the collisions of charged
particles. However, the azimuth component of perpendicular viscosity, in accor-
dance with (1.6), produces radial current. Indeed, azimuth viscosity corresponds to
the transfer of azimuth momentum in the radial direction due to ion–ion collisions.
Hence, an inhomogeneous radial electric field and ion pressure gradient, which pro-
duce inhomogeneous azimuth E × B and ion diamagnetic drifts, cause the azimuth
force applied to ions only. This viscosity force creates radial current, which, in ac-
cordance with (1.7) [similar to (1.16) for the slab case], is
 
1 d 1 d dpi
jr = − 2 η1 r Er − . (1.32)
B dr r dr en dr
Here, the Braginskii expression for the perpendicular viscosity is used, which is valid
for constant ion temperature.
In the absence of an electrode in the center of the plasma column the net radial
current should be zero, as follows from the current continuity equation. Therefore,
the ambipolar radial electric field corresponds to the Boltzmann distribution for ions:
Ti d ln n
Er = . (1.33)
e dr
In other words, the plasma center is biased negatively with respect to its periphery.
This electric field causes such a rotation whereby the ion diamagnetic drift is can-
celed and thus the perpendicular viscosity force is turned to zero.
A more complicated situation with inhomogeneous ion temperature was analyzed
in [17]. In this case, the additional perpendicular viscosity caused by the azimuth
ion heat flux should be taken into account. It is proportional to the first and second
derivatives of the ion temperature and density. The ambipolar radial electric field,
which is calculated from the zero current condition, might be obtained using the
expression for the viscosity tensor from [17, 18]:
  r 
Ti d(nTi ) 19 r d 1 d ln Ti
Er = + Ti   dr 
e nTi dr 12 Ti 0 dr r dr 
 r   r 
5 r 1 dTi 2  11 r 1 dTi d ln n 
+ dr + dr . (1.34)
12 Ti 0 r  Ti dr  3 Ti 0 r  dr  dr 
The value and even the sign of the radial electric field generally depend on the ratio
of the density and the ion temperature gradients.

1.7 Current Systems in Front of a Biased Electrode

(Flush-Mounted Probe) and Spot of Emission
Perpendicular currents determine the structure of the current systems in magnetized
plasma. Among the simplest 2D problems is that of closing currents in front of a bi-
ased electrode. The current distribution determines, for example, the current–voltage
 1 Mechanisma of Transverse Conductivity 17

Fig. 1.5. Schematic of electric fields and currents generated by the viscosity mechanism in
front of a biased electrode. The radial electric field Er is caused by the potential rise in front
of a positively biased electrode, uϑ is the azimuth E × B drift velocity, jr is the viscosity
driven current, jz is the Spitzer current

characteristics of an electrostatic probe, which is a very simple diagnostic tool for

investigating fully ionized plasma. In particular, electrostatic probes (especially so-
called flush-mounted probes [19]) are widely used for the study of edge plasma in
fusion devices, where they are often considered to be a standard diagnostic. Simi-
lar current systems exist in front of a spot of emission, which can transform under
certain conditions into a unipolar arc [20].
We shall demonstrate the principles of current closing for the simplest geometry.
Consider non-uniform plasma with density n(z) restricted by a conductive surface.
Electrons and ions are created in pairs by a source S(z) and are driven towards the
wall by a pressure gradient. Electron and ion temperatures are assumed to be con-
stant. An infinite conductive surface is situated at z = L, while z = 0 is a plane
of symmetry where the z-component of plasma velocity equals zero, Fig. 1.5. The
magnetic field is parallel to the z-axis and is normal to the surface. An electrode
with radius a centered at r = 0 and at z = L is biased (for example, positively)
to potential V > 0 with respect to the conductive surface. The longitudinal size of
the electrode is unimportant because it is always smaller than the longitudinal scale
of the potential perturbation. We assume that for modest applied voltages the plasma
density in the vicinity of the electrode is only slightly perturbed due to the strong am-
18 V. Rozhansky

bipolar anomalous diffusion across the magnetic field. The transverse diffusion can
“wash out” the density perturbations caused by relatively small transverse currents.

1.7.1 Viscosity-Driven Perpendicular Currents

Let us start with the situation where the perpendicular viscosity determines the per-
pendicular current. In the absence of biasing (for V = 0) the plasma potential near
the surface is equal to the floating potential, which equalizes electron and ion fluxes
to the surface [21],

Te mi Te
ϕs = ϕf = ln . (1.35)
e 2πme (Te + γ Ti )

When the electrode is biased positively with respect to the surface, the potential in the
channel in front of it becomes larger than the floating potential, while at r → ∞ the
potential remains unperturbed. As a result, a radial electric field emerges, as shown
in Fig. 1.5. This electric field produces E × B plasma drift in the azimuth direc-
tion. The plasma rotation is differential, because the radial electric field is a function
of r. Therefore, the azimuth component of the ion perpendicular viscosity causes
the radial ion current from the channel to compensate the space charge of electrons
collecting by the probe. This mechanism was discussed in the previous section. The
potential perturbation is spread along the magnetic field over a distance l|| and the
potential perturbation also causes longitudinal currents to the biased electrode and to
the grounded surface. These currents are governed by the classical Spitzer conduc-
tivity. Due to the ion perpendicular current the zone of enhanced potential becomes
much larger than the channel with the radius r = a, and, as a result, a return current
to the surface is created to close the circuit. In this zone, a < r ≤ R plasma potential
at the sheath edge is larger than ϕf and the ion flux to the surface exceeds that of the
electrons.
The following boundary conditions should be imposed at the sheath edge:
  
Te e(ϕs − V )
j|| |z=L, r≤a = e(Γi|| − Γe|| ) = e ns cs − ns exp − ,
2πme Te
  
(1.36)
Te eϕs
j|| |z=L, r>a = e(Γi|| − Γe|| ) = e ns cs − ns exp − .
2πme Te

Here, ns and ϕs denote the density and the potential in plasma at the sheath edge, re-
spectively,
√ cs , according to the Bohm criterion, is the speed of sound, cs =
(Te + γ Ti )/mi , and γ = 1 or 5/3 for isothermal or adiabatic ions, respectively.
For a floating electrode V = 0 the unperturbed density n(z), the parallel velocity Vz ,
and the potential distribution in plasma should be calculated from the 1D momentum
balance equations. In particular, the potential corresponds to the Boltzmann distrib-
ution for electrons: ϕ = (Te /e) ln(n/ns ) + ϕf .
 1 Mechanisma of Transverse Conductivity 19

We shall now consider the transitional part of the I –V characteristic, where the
applied potential is modest so that the current is far from the saturation current
Iisat < I  Iesat . From the parallel component of the momentum balance equation
for electrons we have 
∂ϕ Te ∂ ln n
j|| = −σ − , (1.37)
∂z e ∂z
where σ|| = 1.96ne2 /me νei is the Spitzer conductivity. The perpendicular current
is given by (1.32). Neglecting the density perturbations, from the current continuity
equation it is easy to obtain [22, 23]

η1 1 ∂ ∗ ∂ψ ∂ 2ψ
rΔ = σ || , (1.38)
B 2 r ∂r ∂r ∂z2
where the perturbed potential ψ is defined as ψ = ϕ − (Te /e) ln(n/ns ) − ϕf , and

∗ 1 ∂ ∂ 1
Δ = r − 2.
r ∂r ∂r r
Equation (1.38) should be solved with the boundary conditions in plasma

∂ψ 
, ψ|r→∞ → 0, (1.39)
∂z z=0
and the boundary conditions at the sheath edge adjacent to the material surface,
which is obtained by combining (1.36) and (1.37),
   
∂ψ  Te e(ψs + ϕf − V )
−σ|| = e n c − n exp − ,
∂z z=L, r≤a
s s s
2πme Te
    (1.40)
∂ψ  Te e(ψs + ϕf )
−σ|| = e n s cs − n s exp − .
∂z z=L, r>a 2πme Te

When the applied voltage is small, |V | < Te /e, boundary condition (1.40) can
be linearized:

∂ψ  e(ψs − V )
−σ||  = ens cs ,
∂z z=L, r≤a Te
 (1.41)
∂ψ  eψs
−σ|| = ens cs .
∂z z=L, r>a Te

The longitudinal scale l|| of the potential perturbation can be estimated from this
condition. Estimating ∂ψ/∂z as ψ/ l|| , we find
√ √
σ Te Te /me mi /me
l|| = = √ . (1.42)
n s cs e 2 0.51νei 1 + Ti /Te
The longitudinal scale of potential perturbation is determined by the Spitzer conduc-
tivity and is independent of the viscosity coefficient.
20 V. Rozhansky

(vis)
Fig. 1.6. Equipotentials for normalized perturbed potential Φ = eψ/Te , a/R0 = 0.2,
L/ l|| = 5.0. The potential perturbation is normalized to unity at r = 0, z = L

(vis)
The perpendicular scale R0 of the potential perturbation can be obtained from
(1.38) and (1.42). For the Braginskii viscosity coefficient η1 in (1.38)
1/4 √ 
(vis) 3 Te /Ti Ti /mi mi 1/8
R0 = . (1.43)
5 1 + Ti /Te ωci me
If η1 is replaced by anomalous viscosity coefficient η, we have
1/4 √  
(vis) 3 Te /Ti Ti /mi mi 1/8 η1 1/4
R0 = . (1.44)
5 1 + γ Ti /Te ωci me η
For the small applied potentials, (1.38) was solved both analytically and numerically
in [22, 23]. The calculated normalized potential perturbation Φ = eψ/Te is shown
(vis)
in Figs. 1.6 and 1.7 for the case when the electrode radius a is smaller than R0 .
It can be clearly seen that the longitudinal scale of the potential perturbation is l|| ,
while the transverse scale is on the order of R0(vis) . The variation of the mean free path
does not change the transverse scale of the perturbation, which remains the same for
both Figs. 1.6 and 1.7. On the contrary, the longitudinal scale increases with λmfp
in the more collisionless regime. The scale R0(vis) determines the radius of the return
current to the conductive surface.
For the electrode, which is small with respect to the current collecting zone radius
(vis)
(a < R0 ), and for the small applied voltages V , the current I to the electrode
can be easily calculated. Indeed, since the net current to the electrode and to the
conductive surface is zero, the perturbed potential ψs is small with respect to V .
Hence, according to (1.41)
I = πens cs a 2 eV /Te = Iisat eV /Te . (1.45)
 1 Mechanisma of Transverse Conductivity 21

(vis)
Fig. 1.7. Equipotentials for normalized perturbed potential Φ = eψ/Te , a/R0 = 0.2,
L/ l|| = 0.2. The potential perturbation is normalized to unity at r = 0, z = L

This expression justifies the widely used method of determining the electron temper-
ature from the slope of current voltage characteristic (see, e.g., [24] and the literature
therein) 
dV 
Tes tan d = eIisat . (1.46)
dI V =0
(vis)
The situation when the electrode radius a is larger than R0 is quite different. In
this case, electron current to the electrode is collected not by the whole electrode area
(vis)
but by the band of width R0 at its periphery. Due to the finite transverse conduc-
tivity, the return current collecting zone at the surface and the band at the electrode
(vis)
have the same scale R0 . The potential perturbation is, therefore, on the order of V
to cancel the current to the center of the electrode [see (1.35)]. The remaining part
of the current to the electrode, as well as the current with the opposite sign to the
conductive surface, can be estimated as
σ|| V return eV return eV S return
I∼ S ∼ ens cs S = kIisat . (1.47)
l|| Te Te S electrode
Numerical simulations [22, 23] support this estimate and, moreover, give the exact
value of the numerical coefficient k = 0.5. We can conclude that the current to the
large electrode [see (1.47)] is significantly smaller than that given by (1.45). The
electron temperature calculated from the slope of the I –V characteristic is


sat dV  S return
Te = keIi . (1.48)
dI V =0 S electrode
It is significantly smaller than the standard expression (1.46). This fact can explain
some unrealistic values of electron temperature measured in several regimes in toka-
mak edge plasma using standard expression (1.46); for more details see [25].
22 V. Rozhansky

(vis)
Fig. 1.8. Calculated I –V characteristics for different electrode sizes (the ratio a/R0 is
specified in the rectangle for each plot). The broken line corresponds to (1.47) with k = 0.5

When the applied positive voltage V is on the order of or larger than Te /e, one
has to impose nonlinear boundary conditions, (1.40). The nonlinear regime was sim-
ulated in [22, 23]. The potential in front of the electrode remains on the order of V ,
since the potential difference between the plasma and the electrode should be on the
order of the floating potential, otherwise large electron thermal current would flow
(vis)
to the electrode. This potential decreases with the scale R0 and, therefore, the
return current and the current to the electrode are given by (1.47). In other words,
the transitional part of the current–voltage characteristic of the electrode is close to
the linear function of the applied potentials. An example of the simulation is shown
in Fig. 1.8.
For very large positive applied potentials, the perpendicular ion flux, which is
proportional to the viscosity driven current, becomes close to the particle diffusive
flux and the density in front of the electrode is depleted. This situation corresponds to
the electron saturation current. In fact, electron saturation current can be calculated
as diffusive flux to a depleted channel of length l|| in front of the electrode [26]. It
should be noted that an attempt to estimate electron saturation current as maximal
possible return current, which is proportional to the ion thermal flux to the area of
(vis)
radius R0 [27], is erroneous. The reason consists in the fact that the radius of
the current-collecting zone increases with applied voltage [22, 23], and the resulting
current is significantly larger.

1.7.2 Currents Driven by Ion–Neutral Collisions

When this mechanism dominates the perpendicular current is proportional to the

electric field itself. Neglecting density perturbations and assuming small neutral ve-
 1 Mechanisma of Transverse Conductivity 23

locity, according to (1.11) we have

j(iN) = −σiN ∇⊥ ϕ = −σiN ∇⊥ ψ. (1.49)
Substitution of this current into the current continuity equation yields the Laplace
equation
n0 μiN νiN ∂ 2ψ
Δ ⊥ ψ + σ || = 0. (1.50)
B2 ∂z2
Since the longitudinal scale of the potential perturbation is determined by the bound-
ary conditions (1.40), (1.41), it remains the same as in the previous case and is given
by (1.42). The perpendicular scale, which can be estimated from (1.50), is

 
n0 μiN νiN l||2 νiN mi 1/4
R0i−N = ∼ ρ ci . (1.51)
B 2 σ|| νii me
Equation (1.50) has been solved analytically in [23, 25] with the linearized boundary
conditions from (1.41), and with the nonlinear boundary conditions in [23]. It was
demonstrated that the results of the previous section remain valid after the substitu-
(vis)
tion R0i−N → R0 .

1.7.3 Inertia Currents

If a global convective flux across the magnetic field exists in the plasma, inertia
currents arise. Such a flux may be produced by E × B drift in a global electric field
or by low frequency large-scale turbulent electric fields. In [23, 25] a 2-D case with
a biased electrode being a band of width 2a infinite in the y direction is considered.
A constant plasma velocity ud in the x direction is responsible for the transverse
current
n0 mi ud ∂ 2 ϕ
jx = − . (1.52)
B 2 ∂x 2
Substitution of this expression into the current continuity equation yields
n0 mi ud ∂ 3 ψ ∂ 2ψ
2 3
+ σ|| 2 = 0. (1.53)
B ∂x ∂z
Again, as in the previous two cases, the longitudinal scale is given by (1.42), while
the perpendicular scale of the potential perturbation should be estimated from (1.53):
  
n0 mi ud l||2 1/3 ud λmfp mi 1/3
R0inertia = ∼ ρ ci . (1.54)
B 2 σ|| cs ρci me
Equation (1.53) has been solved analytically in [23, 25]. An example of the poten-
tial perturbation distribution in the plasma is shown in Fig. 1.9. In contrast to the
previous two cases, the potential perturbation is asymmetric in the direction of the
convection. However, the expression for the I –V characteristic again has the uni-
(vis)
versal character of (1.47) with the substitution R0inertia → R0 (for further details
see [23]).
24 V. Rozhansky

Fig. 1.9. Example of the equipotentials for normalized perturbed potential Φ = eψ/Te , when
(inertia)
the transverse current is caused by inertia for a/R0 = 0.2, L/ l|| = 5.0. The perturbed
potential is normalized to unity at the middle of the probe

1.7.4 General Situation

Under real conditions three mechanisms of the effective transverse conductivity may
act simultaneously. To choose the dominating mechanism it is necessary to compare
the three perpendicular scales of the current-collecting zone given by (1.43), (1.44),
(1.51), and (1.54). The largest value of R0 corresponds to the largest perpendicular
current in the system, which closes the current loop. The corresponding criteria can
be found in [23, 25].
The model described above can be easily generalized to the case of an inclined
magnetic field [23]; it is only necessary to replace the electrode size by its projection
across the magnetic field.
One more condition is connected with the assumption of the unperturbed plasma
density. This assumption remains valid if the perpendicular diffusive flux δnD/R0
(for a ∼ R0 ), which balances the perpendicular flux associated with current j⊥ /e in
the particle continuity equation, corresponds to small δn/n0 . In the case of a viscos-
ity dominated mechanism of the effective transverse conductivity, for example, this
condition is violated even for applied voltages on the order of Te /e if both perpendic-
ular viscosity and diffusion coefficients are classical. In contrast, for the anomalous
coefficients η ∼ mi nD and V ∼ Te /e the value δn/n0 ∼ ρci2 /R02 is small. However,
for sufficiently large positive applied voltage δn becomes close to n0 and it is nec-
essary to solve both current and particle continuity equations simultaneously. This
situation corresponds to the electron saturation current, which has been calculated
in [26].
 1 Mechanisma of Transverse Conductivity 25

1.7.5 Spot of Emission

Spots of enhanced emission in a magnetic field are often observed on the metal sur-
faces and walls of fusion devices (see, e.g., [28]). Under some conditions a unipolar
arc can be ignited in these places. The term unipolar arc was first introduced by
Robson and Thoneman [29] to describe an arc discharge with a metal surface as a
cathode and plasma as an anode, which is ignited without applying any external volt-
age. Many authors have reported arcing of the walls of fusion machines, limiters, or
divertor plates; see, for example, [30, 31]. The surface erosion produced by arcs is
one of the serious factors that restrict reactor operation, especially during disruptions
when dense and hot plasma is brought in contact with the surface.
It is well understood that unipolar arcs can exist only when there is a return
current flowing from the plasma to the surface to close the current circuit. Before
the ignition of an arc, the potential perturbation in the plasma in front of a spot of
enhanced emission and the corresponding current system are quite similar to those
discussed in the previous sections for the biased electrode. Indeed, electron current
from the spot of emission reduces the potential in front of it with respect to the
floating potential. Furthermore, the radial electric field arises and causes the radial
current to be dependent on the dominating mechanism of effective perpendicular
conductivity. The reduced potential also creates the return current to the electrode.
The potential perturbation is thus described by one of (1.38), (1.50), or (1.53). The
boundary condition at the surface should be replaced by
   
∂ψ  Te e(ψs + ϕf )
−σ = e n s cs − ns exp − + Γem . (1.55)
∂z z=L 2πme Te

Here, Γem is the flux of emitted electrons, which is a function of the radius. For
example, one can choose
2
Iem r
Γem (ρ) = exp − 2 .
eπa 2 a
In the linear case, when the potential perturbation is small, the solution is the same
as for the case of a biased electrode [20].
When the emission current is large and the unipolar arc is ignited, the solution
obtained is still valid outside the arc where the plasma is unperturbed and the return
current is collected. On the basis of the solution obtained a criterion of unipolar arc
formation has been put forward in [20]. The criterion is based on two experimental
parameters, which are usually known for each material: the cathode voltage drop of
the arc and the minimal current typical for the arc. When the reduced potential in
front of the emitting surface is larger than the cathode voltage drop of the arc and
the return current exceeds the minimal current typical for the arc, then ignition of
the arc is possible. The potential drop between the plasma and the spot of emission
as a function of emission current is calculated numerically as in the previous sec-
tions. In this criterion, in contrast to previous estimates, the return current, which is
determined by the perpendicular viscosity, is properly calculated.
26 V. Rozhansky

1.8 Currents in the Vicinity of a Biased Electrode That is Smaller

Than the Ion Gyroradius

When the transverse size of an electrode a is smaller than the ion gyroradius ρci
(but still larger than the electron gyroradius ρce ), the effective transverse current
is quite different from that considered in the previous section. Now ions can move
freely across the magnetic field and the parallel electron current, which should be
closed in the plasma, restricts the perpendicular current. The I –V characteristic of a
small electrode (probe) in fully ionized plasmas was calculated numerically by San-
martin [32]. However, his results were seldom used. One reason consists in the fact
that the perpendicular diffusivity in the fusion plasmas is anomalous while calcula-
tions in [32] were performed for the classical case.
The problem was analytically approached recently in [33]. It was demonstrated
that, on one hand, the I –V characteristic could be represented by a simple analytical
expression, which coincides with simulation [32], and, on the other hand, experi-
ments on the TdeV tokamak [34] with small probes can be described by the theory
with the classical diffusion coefficient. The approach of [33] is similar to that used
for weakly ionized plasma by Bohm [35] (saturation currents) and Rozhansky and
Tsendin [36] (whole characteristics), where the ion mobility across the magnetic field
is much larger than that of the electrons. Similar ideas were also discussed in [37],
however the erroneous attempt has been made to apply this approach to the case of
fully ionized plasma and a large electrode, which is analyzed in the previous section.
Let us consider a homogeneous plasma of density n0 . When the electrode is bi-
ased positively it collects electrons and the plasma density is reduced in an ellipsoidal
zone having characteristic dimensions a and l|| , Fig. 1.10. In contrast to the previous
case, we assume the classical type of diffusivity, since at scales a smaller than the ion
gyroradius it is unlikely to have intensive plasma fluctuations. Therefore, perpendic-
ular diffusion is unable to compensate plasma depletion in front of the electrode. The
ions can move freely across B substantially faster than the electrons. Thus, the ions
should be trapped in the electron ellipsoid, i.e., the electric field should be balanced
by the ion pressure gradient, and the Boltzmann distribution is established for the
ions:
Ti n
ϕ = − ln + ϕf . (1.56)
e n0
The electron fluxes can be obtained from the parallel and perpendicular projections
of the momentum balance equation for electrons:
∂n
∗
Γe|| = Γi|| − De|| , Γe⊥ = −De⊥ ∇⊥ n, (1.57)
∂z
∗ = (T + T )/(0.51m ν ) is the effective parallel diffusion coefficient of
where De|| e i e ei
electrons and De⊥ = (Te + Ti )νei /(me ωce 2 ) is the classical perpendicular diffusion

coefficient.
The ions are gathered to the electrode from the other region whose size is on
the order of the ion gyroradius ρci , Fig. 1.10, where (1.56) and (1.57) are violated.
 1 Mechanisma of Transverse Conductivity 27

Fig. 1.10. Structure of electron and ion current collection regions for a small probe: 1—
electrode, 2—conductive surface, 3—electron collection region, 4—ion collection region

Substituting (1.57) into the electron particle continuity equation yields

 2 
1 ∂ n ∂n ωce ∂ n0 ∂n
r + = 0. (1.58)
r ∂r n0 ∂r 0.51νei2 (n0 ) ∂z n ∂z

We neglected here the divergence of the parallel ion flux, which is directed out of the
zone of the ion collection. From (1.58) it follows that the characteristic longitudinal
dimension of the electron ellipsoid is

l|| = a(ωce /νei ). (1.59)

In contrast to the case of the large electrode, (1.42), this scale depends on the elec-
trode size a.
If the electron current to the electrode is smaller than the saturation electrode
current, the longitudinal potential profile in the plasma should be nonmonotonic (see
Fig. 1.11). This effect is known as potential overlap [33, 36]. In most parts of the
electron ellipsoid the potential corresponds to the Boltzmann distribution for the ions,
(1.56). The potential in this region increases from the plasma potential ϕf at infinity
to its maximum ϕ ∗ , which is related to the plasma density n∗ at this point according
to (1.56). Near the probe, however, there must be a region where the electrons are
trapped along B,  since their flux to the electrode must be less than the thermal flux.
Consequently, in this region the potential profile should correspond to the Boltzmann
distribution for the electrons and should therefore decrease on approaching the probe.
28 V. Rozhansky

Fig. 1.11. Potential distribution along a magnetic field: 1—electrode, 2—sheath, 3—electron
collection region

The electron flux to the probe can be expressed in terms of the potential difference
ϕ ∗ − V (V is the electrode potential)
 
∗ Te e(ϕ ∗ − V )
Γe|| = n exp − . (1.60)
2πme Te

Then, substituting the maximum of the potential ϕ ∗ from (1.56) into (1.60), we obtain
∗ ∗ 
Ti n Te n Te
− ln − ln = V − ϕf . (1.61)
e n0 e Γe|| 2πme

Most of the electron current Ie to the electrode is collected far from the electrode
where the plasma density is weakly perturbed. Thus, to obtain the analytical ex-
pression for the electron current, we replace the nonlinear equation (1.58) by the
corresponding linear Laplace equation
 ∗ (n ) 2
1 ∂ ∂n De|| 0 ∂ n
r + = 0. (1.62)
r ∂r ∂r De⊥ (n0 ) ∂z2
The problem may be solved analytically by imposing the boundary condition n = n∗
at the surface of the electrode (thus neglecting the longitudinal dimension of the ion
ellipsoid compared with the longitudinal dimension of the electron ellipsoid). In this
linearized case, the electron current Ie should be a linear function of the plasma
density n∗ ,
n∗ Ie − Ii (n∗ )
=1− . (1.63)
n0 Iesat
We shall assume the ion current to a positively charged electrode to be Ii = en∗ cs ×
S electrode = Ii0 n∗ /n0 . The electron current to the electrode is Ie = eΓe|| S electrode . The
electron saturation current Iesat is calculated by solving the Laplace equation (1.62)
with zero boundary conditions at the electrode surface, similar to the case of weakly
 1 Mechanisma of Transverse Conductivity 29

Fig. 1.12. Electron current to the electrode Ie versus applied potential V for various magnetic
fields, Te = Ti = T . Results of the numerical calculations of Sanmartin [32] (dashed curves)
are compared with those obtained by analytic expression (1.65) (solid curves)

ionized plasma [35]


∗ D C = k2.8πen ρ c C,
Iesat = k2πen0 De|| ρci = cs /ωci . (1.64)
e⊥ 0 ci s

The nonlinearity of the initial equation (1.58) is taken into account by introducing a
coefficient k on the order of unity. The function C is a geometric factor determined by
the size and shape of the electrode and corresponds to the capacity of a conductor of
the same shape as the electrode but whose longitudinal dimension is (De|| ∗ /D )1/2
e⊥
times shorter. For a disk electrode of radius a the coefficient is C = 2a/π. Other
cases are considered in [33].
An expression for the transition part of the I –V characteristic is obtained by
substituting (1.63) into (1.61). For example, if Te = Ti = T we find
   
Ie 2 en0 S electrode T Ii0 2 e(ϕf − V )
1 − sat = 1 − sat exp . (1.65)
Ie Ie 2πme Ie T

In Fig. 1.12 the I –V characteristics obtained from (1.65) are compared with the
numerical solution of the nonlinear problem of (1.58). Good agreement is achieved
over a wide range of magnetic fields for k = 0.7. For Te = Ti one should use the
more complicated expression which follows from (1.61), (1.63).
In [33] other geometries are considered, including the case of inclined magnetic
field lines. It is also demonstrated there that (1.65) can describe the I –V character-
istics of a small probe obtained on TdeV [34]. Therefore, in contrast to the case of a
30 V. Rozhansky

large electrode, the model with a classical diffusion coefficient gives results close to
the experimental ones.

1.9 Neoclassical Perpendicular Conductivity in a Tokamak

1.9.1 Steady State Current

The problem of a self-consistent radial electric field in an installation with an inho-

mogeneous magnetic field such as a tokamak is much more complicated than that
considered in Sect. 1.6 for the cylinder. As in the cylinder, the radial electric field is
determined by the condition I = 0 or I = const in the biasing experiments, where I
is the net current through the flux surface. However, the contributions from the other
radial currents, in particular from the diamagnetic current and current caused by the
parallel viscosity, are more important than those from the perpendicular viscosity
driven current. It is also necessary to take into account the possibility of toroidal
rotation caused by external forces or generated by internal torque.
We consider the general toroidal geometry where the x and y coordinates cor-
respond to the directions along and across flux surfaces, respectively, and z is the
toroidal direction. The metric coefficients are
1 1 1 √
hx = , hy = , hz = , g = hx hy hz .
∇x ∇y ∇z
One can also replace hz → 2πR, where R is the major radius. The physical compo-
nents of a vector are used. Subscript “⊥” denotes the direction perpendicular both to
the magnetic field B and to the y-axis, bx = Bx /B, bz = Bz /B.
The surface averaged radial current is
  
I hx hz dx = jy

= j˜y(dia) + jy(vis) + jy(in) + jy(iN) , (1.66)

 
where f

= f hx hz dx/ hx hz dx. The diamagnetic current here is taken in the

divergent form of (1.13). In the tokamak geometry

n(Te + Ti )Bz ∂ 1
j˜y(dia) = − . (1.67)
hx ∂x B 2

We separate the viscosity driven current into currents caused parallel, perpendicular,
and gyroviscosity. The divergent part of the current driven by the parallel viscosity,
(1.19), is 
(1/4)π|| Bz ∂ 1
j˜y(vis||) = − . (1.68)
hx ∂x B 2
In the fluid or Pfirsch–Schlueter regime, an expression for the parallel viscosity cur-
rent may be found in [38]. It is also demonstrated there that other currents in (1.66)
are significantly smaller in the core region of a tokamak.
 1 Mechanisma of Transverse Conductivity 31

Fig. 1.13. Divergent part of the diamagnetic and parallel viscosity-driven currents in a tokamak

The diamagnetic and parallel viscosity currents are shown in Fig. 1.13. Both
currents are directed vertically, as follows from (1.67) and (1.68). The diamagnetic
current is the largest in the system since in the fluid regime the parallel viscosity,
which depends on the parallel velocity and parallel ion heat flux distribution over the
flux surface, is a small correction to the pressure. However, the contribution from the
viscosity driven current to the average radial current equation (1.66) is on the same
order as the contribution from the diamagnetic current. The reason is connected with
the fact that the pressure is almost constant on the flux surface. Hence, the averaged
diamagnetic current is much smaller than its local value.
An arbitrary radial electric field causes poloidal E × B drift. In the inhomoge-
neous magnetic field the poloidal drifts are not divergence free and hence the parallel
flux is generated to satisfy the particle continuity equations. The ion parallel flux,
which closes the diamagnetic poloidal flux and poloidal E × B drift, is known as
Pfirsch–Schlueter flux; see, e.g., [39–41]. The net parallel velocity is
V|| B

V|| = V||P.S. + ;
 B    (1.69)
∂pi ∂ϕ Bz V|| B
B2
V|| =
P.S.
+ − 1− 2 ,
en∂y ∂y hy Bx B B B

 
where f
= f hx hy hz dx/ hx hy hz dx. This parallel flux is driven by the paral-
lel pressure gradient and hence the poloidal pressure is perturbed on the flux surface.
The parallel viscosity also arises due to this parallel flux and due to the similar ion
parallel heat flux [39–41]. The diamagnetic and the parallel viscosity-driven currents,
(1.67) and (1.68), which depend on the radial electric field, do not satisfy the condi-
tion I = 0 (or I equal to a given value in the biasing experiments) for an arbitrary
32 V. Rozhansky

radial electric field. Therefore, the self-consistent radial electric field is determined
by the ambipolarity condition I = 0 and the parallel momentum balance equation.
The parallel momentum balance equation in a simplified form contains the pres-
sure gradient, parallel viscosity, an external force in the case of unbalanced neutral
beam injection (NBI), and a term responsible for the radial transport of parallel mo-
mentum (for details see [38]):
∂p d(mi nV|| )
−bx − (∇ · π|| )|| + F (ex) = , (1.70)
hx ∂x dt
where  √ 
d(mi nV|| ) 1 ∂ hz g η ∂V||
= √ mi Γy V|| − . (1.71)
dt hz g ∂y hy hy ∂y
Here, Γy is the radial particle flux, Γy = −D h∂n y ∂y
+ V n with D and V being the
anomalous coefficients of diffusion and convection, and η is the anomalous viscos-
ity coefficient. We neglect here the ion–neutral friction with respect to the radial
transport of parallel momentum in accordance with the results of numerical simula-
tion [38]. The parallel momentum balance equation may be averaged over the flux
surface with the weight B to cancel the pressure gradient contribution:
 
  dV||
− B · ∇ · π ||
+ B · F (ex) = Bnmi
↔
. (1.72)
dt
The averaged parallel viscosity may be expressed through the radial electric field. In
the collision-dominated regime this can be done using the Braginskii viscosity and
additional viscosity, which is proportional to the parallel heat flux. In the collisionless
case one has to solve the drift kinetic equation. The general expression, which is valid
in all regimes, was calculated in the neoclassical theory (see reviews [39–41]):
 
 ↔ B ∂ϕ Ti ∂n ∂Ti
B · ∇ · π ||
= −ν (mp)
nmi + + kT − BV||
. (1.73)
hy Bx ∂y en ∂y e∂y
The magnetic pumping frequency is

B ∇B)
μi1
3 ( B 2
ν (mp) = , (1.74)
B 2
nmi
where the viscosity coefficient μi1 is calculated in [40]. In the collision dominated
regime viscosity coefficient μi1 coincides with Braginskii parallel viscosity coef-
ficient η0 = 0.96nTi /νi . For the tokamak with a circular cross section and small
ε = r/R, where r and R are the radius of flux surface and the major radius, respec-
tively, the magnetic pumping frequency is

3bx2 μi1
ν (mp) = . (1.75)
2R 2 nmi
The numerical coefficient kT is 2.7, 1.5, and −0.17 in the collision dominated, plateau,
and banana regimes, respectively.
 1 Mechanisma of Transverse Conductivity 33

The parallel momentum balance equation contains two unknown variables: the
radial electric field and the average toroidal rotation. As a second equation it is con-
venient to use the toroidal component of the momentum balance equation [41, 42]
dVz
−jy Bx − (∇ · π|| )z + F (ex) = nmi . (1.76)
dt
There is no pressure gradient in (1.76) due to the toroidal symmetry. The parallel
and toroidal velocities are almost the same, Vz ≈ V|| , and we shall neglect the small
correction cased by the projection of the poloidal velocity to the parallel direction.
The surface averaged radial current, as follows from (1.76), is
  
1 dVz
jy

= F (ex) − mi n . (1.77)
Bx dt
The parallel viscosity does not contribute to this equation. This fact may be shown
using the parallel viscosity in the form of (1.18).
Two equations, (1.72) and (1.77), determine the radial current as a function of
the radial electric field and also determine the average toroidal rotation for the given
radial current. Combining (1.72) and (1.77), we find
  
B · ∇ · π ||

↔
1 dVz
jy

= + F (ex) − mi n
BBx
Bx dt

B · [F (ex) − mi n dt|| ]

dV
− . (1.78)
BBx

Let us start with the simplest situation of the absence of biasing, when I =
jy

= 0. We introduce the parameter

ν (mp) L2
κ= , (1.79)
D
where L is the radial scale for the toroidal velocity. The mean toroidal velocity is
canceled in the last two terms on the r.h.s of (1.78) and only the Pfirsch–Schlueter
part contributes to the difference of the second and the third terms in the r.h.s. Insert-
ing (1.73) and (1.69), we see that the term with the parallel viscosity dominates in
(1.78) when
ν (mp) L2
κ= > ε2 . (1.80)
D
If this condition is satisfied, then the radial electric field is determined by the neo-
classical expression B · ∇ · π ||
= 0 [39–41]:
↔


Ti 1 d ln n 1 d ln Ti Bx
E (NEO)
= + kT − BV||
. (1.81)
e hy dy hy dy B
The profile of the average toroidal velocity may be obtained from (1.77). Note that in
the absence of NBI, when the toroidal rotation is small, mainly the density and tem-
perature radial profiles determine the radial electric field. If κ < ε 2 , the radial electric
34 V. Rozhansky

Fig. 1.14. The radial electric field in the equatorial midplane calculated by B2SOLPS5.0 fluid
code in the absence of NBI (a) and for NBI in the co-current direction (b)

field profile differs significantly from the neoclassical expression. In Fig. 1.14 the ra-
dial electric field in the separatrix vicinity calculated by means of the B2SOLPS5.0
fluid code [43] is shown and compared with the neoclassical electric field, (1.81).
With the exclusion of a small region at the separatrix vicinity on the order of 1 cm,
where the condition in (1.80) is violated, the radial electric field is close to the neo-
classical electric field and its dependence on the local parameters coincides with that
of the neoclassical electric field. Such an electric field has been observed in experi-
ments on tokamaks; see reviews [44–46] both for low (L) and high (H) confinement
regimes.
On several tokamaks (CCT [47, 48], TUMAN-3 [49], TEXTOR [50] and others)
the biasing experiments were performed where a biased electrode was inserted into
the core plasma. Thus, a potential on the order of a few hundred eV was applied to
the core flux surface with respect to the limiter, and the current–voltage characteristic
was measured. The typical I –V characteristic is shown in Fig. 1.15. To calculate the
I –V characteristic and to obtain local effective perpendicular conductivity of the
tokamak plasma it is necessary to keep the net radial current in (1.77) and (1.78)
as a prescribed value. According to the toroidal momentum balance of (1.77), the
radial current accelerates plasma in the toroidal direction due to the j × B force.
The toroidal velocity may be calculated from (1.77). Furthermore, if the condition in
(1.80) is satisfied, the radial current in (1.78) may be expressed through the parallel
velocity [42, 49]
B · ∇ · π ||

jy

= . (1.82)
BBx

The parallel velocity depends both on toroidal rotation and radial electric field. If

ν (mp) L2
1>κ= > ε2 , (1.83)
D
 1 Mechanisma of Transverse Conductivity 35

Fig. 1.15. Typical current–voltage characteristic during the biasing experiment in the
Tuman-3 [49]

then from the parallel momentum balance in (1.72) it follows that the toroidal rota-
tion velocity V||
∼ κEy /Bx is small and thus may be neglected in the expression
for the parallel viscosity (1.73). Equation (1.82) in this case yields [42, 49]
 
B ∂ϕ Ti ∂ ln n ∂ ln Ti
jy

= −ν (mp) nmi − + kT . (1.84)

BBx
Bx hy ∂y e ∂y ∂y

The average radial current in this regime is the linear function of the electric field.
In contrast, the situation in the case

κ>1 (1.85)

is quite different. The radial current still might be expressed through the parallel
viscosity according to (1.82), but the contribution of the toroidal rotation term to
the parallel viscosity of (1.73) is quite significant. Indeed, since κ > 1, the parallel
momentum balance, (1.72), may be satisfied only if the sum in the r.h.s. of (1.73)
is close to zero. In other words, the poloidal rotation in this regime is close to the
neoclassical poloidal rotation and the radial electric field is given by the neoclassical
expression in (1.81). To calculate the radial current one has to use (1.77) where the
radial current is expressed through the toroidal velocity. Neglecting the difference
in averaging of the toroidal velocity in (1.81) and (1.77), which is equivalent to ne-
glecting the Pfirsch–Schlueter velocities with respect to the main toroidal velocity,
and expressing the toroidal velocity in (1.81) through the radial electric field and
inserting it into (1.77), we find
   
1 d Ey Ti 1 d ln n 1 d ln Ti
jy

= mi n − + + kT . (1.86)
Bx dt Bx e hy dy hy dy
36 V. Rozhansky

Fig. 1.16. Radial profiles of the potential measured for different applied voltages in the biasing
experiment in TEXTOR

The profile of the radial electric field and the effective perpendicular conductivity in
this case is quite different from that given by (1.84). Finally, for κ < ε 2 the radial
current cannot be expressed through the parallel viscosity and the corresponding
equation is more complicated [51].
The analytical expressions in this section were checked by comparison with the
results of numerical simulation for ASDEX Upgrade performed by means of the
B2SOPS5.0 code [51]. The electric field observed on TEXTOR [52, 53], Fig. 1.16,
may also be understood from the analytical results (see [51]).
In some publications [50, 52, 53] it was suggested that the j × B force, which
accelerates plasma in the toroidal direction, is balanced by ion–neutral friction. How-
ever, the simulation results demonstrate that this force is an order of magnitude
smaller than the anomalous radial transport of toroidal momentum.
In experiments [47–49] it was observed that at large applied voltages the radial
current started to decrease. The current drop is usually accompanied by turbulence
suppression and transition to the regime of improved confinement (L–H transition).
In [42, 49, 54] this effect was associated with the decrease in the parallel viscosity
 1 Mechanisma of Transverse Conductivity 37

when the poloidal drift velocity becomes near the poloidal speed of sound bx cs .
Indeed, as is known (see, for example, review [41] and references therein), (1.73)
is valid only in the linear case if the poloidal velocity |Vx |  bx cs . In [54] it was
suggested that the radial profile of the electric field for large applied voltages may be
discontinuous. In [55] the radial current was taken in the form
B
jy

= −ν (mp) nmi
BBx
Bx hy
   
∂ϕ Ti ∂ ln n ∂ ln Ti  ∂ϕ 
× − + kT f   
∂y e ∂y ∂y bx cs hy ∂y 
  
1 ∂ ∂ ∂ϕ ∂pi
+ η1 + , (1.87)
B 2 hy ∂y hy ∂y hy ∂y enhy ∂y
where f is a decreasing function (e.g., exp(−|∂ϕ/∂y/(hy bx cs )| in the plateau re-
gime). It was shown that for large applied voltages, (1.87) has the soliton-like so-
lution, where a narrow region of a large electric field was localized somewhere be-
tween the biased electrode and the separatrix. Such a profile was also obtained in
the numerical simulations [56, 57] by the Monte-Carlo code. However, for large ap-
plied voltages the criterion of validity of (1.83) is rather severe. Indeed, the solitary
structure for an electric field also implies the solitary structure for Pfirsch–Schlueter
parallel flux, which results in strong radial transport of parallel momentum (the scale
L in (1.83) becomes small). Hence, the neoclassical part of the radial current, which
corresponds to the fist term in (1.87), cannot be expressed through the parallel vis-
cosity and is given by the more complicated equation (1.78). In the simulations with
full fluid code [51] the formation of a solitary structure was not observed.
The current drop for large applied voltages and the details of the radial electric
field profile observed in [58, 59] may be explained by the transition to the H-mode.
Due to the drop in the diffusion and the perpendicular viscosity coefficient, in the
H-mode the radial transport of toroidal momentum is strongly reduced and the con-
dition in (1.85) κ > 1 is more likely fulfilled even if it was not satisfied in the
L-regime. Therefore, assuming that radial current is given by (1.86), we see that the
current is proportional to the anomalous transport coefficients and hence should drop
in the H-mode.

1.9.2 Time-Dependent Current

The time-dependent radial current in a tokamak is quite different from the inertia
(polarization) current in the straight magnetic field of (1.10). The time-dependent
radial electric field creates Pfirsch–Schlueter time-dependent parallel flux and time-
dependent average toroidal rotation, which causes the parallel and toroidal inertia
terms. According to (1.78), these forces lead to an additional time-dependent current.
Combining (1.78) with (1.69) and (1.72) after adding the normal inertia current from
(1.10), we find the time-dependent part of the radial current,
  
 (t)  B 2
B2 1 1 ∂ 2 ϕy
jy = − 1− 2 + nmi . (1.88)
B
Bx
B/Bx

B
hy Bx2 hy B 2 ∂t∂y
38 V. Rozhansky

For circular cross section, (1.88) is reduced to

 
 (t)    1 ∂ 2 ϕy
jy = − 1 + 2q 2
nmi , (1.89)
B 2 hy ∂t∂y
where q is the safety factor. Thus, the polarization current is enhanced by a factor of
(1 + 2q 2 ). In the plateau and banana regimes the corresponding factor is larger;
√ see,
e.g., [41]. In the banana regime the factor (1 + 2q 2 ) should be replaced by ε/bx2 .

1.10 Transverse Conductivity in a Reversed Field Pinch

The radial current in the reversed field pinch (RFP) has much in common with that in
the tokamak, in spite of the fact that in the RFP the main component of the magnetic
field is the poloidal magnetic field. Therefore, in contrast to the tokamak situation,
the radial electric field and the radial pressure gradient generate the toroidal E × B
drift
∂pi
Ey = −Vz Bx + (1.90)
enhy ∂y
instead of the poloidal E × B drift.
The toroidal projection of the momentum balance equation then gives the current
voltage characteristic, which practically coincides with (1.86) (the coefficient kT is
replaced by 1). The radial profiles of the radial electric field and the toroidal veloc-
ities were measured in RFP [60] during the biasing experiments. The profiles are
in good agreement with those calculated according to (1.86) and (1.90), Fig. 1.17.

Fig. 1.17. Toroidal velocity in RFX calculated according to (1.86) and the measured density,
temperature, and radial electric field profiles (continuous line) and according to (1.90) and the
measured electric field profile
 1 Mechanisma of Transverse Conductivity 39

Hence we can conclude that in the absence of strong magnetic turbulence, which may
modify the perpendicular conductivity (see Sect. 1.13), the perpendicular current in
the RFP is determined by the expression in (1.86).

1.11 Modeling of Electric Field and Currents in the Tokamak

Edge Plasma

Such modeling was performed by two-dimensional transport fluid codes such as

B2SOLPS5.0 [38], UEDGE [61–64], EDGE2D [65], and TECXY [66]. Typical ra-
dial electric field profiles calculated by B2SOLPS5.0 [38] in the absence of bias-
ing are shown in Fig. 1.14. Similar radial profiles were obtained by UEDGE [67]
and TECXY [68]. In [43] the dependence of the radial electric field on the plasma
parameters was studied in detail. It was demonstrated that with the exception of
a small region in the separatrix vicinity the radial electric field depends on para-
meters like the neoclassical electric field. As an example, Fig. 1.18 illustrates the
weak dependence on the toroidal and poloidal magnetic fields in the absence of
NBI in accordance with (1.81). The deviation from the neoclassical expression in

Fig. 1.18. The radial electric field profiles at the outer midplane for discharges with different
toroidal magnetic fields B and different plasma currents I . No NBI, normal direction of the
magnetic field, n = 2 × 1019 m−3 , Ti = 80 eV in the reference point
40 V. Rozhansky

Fig. 1.19. Different components of the parallel momentum balance equation averaged over
the flux surface: 1—perpendicular anomalous viscosity and inertia [r.h.s. of (1.72)], 2—
↔(NEO)
parallel viscosity B · ∇ · π
[l.h.s. of (1.72)]. No NBI, reversed magnetic field, n =
19 −3
2 × 10 m , Ti = 98 eV in the reference point 1 cm inside the separatrix, I = 1 MA,
B =2T

the separatrix vicinity in Fig. 1.14 may be explained by the fact that the parame-
ter κ here becomes smaller than ε 2 due to the small scale L of the variation of the
plasma parameters. In Fig. 1.19, the parallel viscosity B · ∇ · π ||
averaged over
↔

the flux surface and the r.h.s. of (1.72), which corresponds to the radial transport of
toroidal momentum, are shown. One can see that the parallel viscosity decreases
towards the core with the scale δ which in the Pfirsch–Schlueter regime is esti-
mated [43] as δ ∼ (B 2 a 2 Dνii /Bx2 cs2 )1/2 , where a is the minor radius. Therefore,
the numerical modeling supports the conclusion that the radial electric field in the
core region in the absence of biasing is given by the neoclassical expression even
in the presence of anomalous transport. On the basis of the simulations a scaling
for the threshold of the transition to the high confinement regime (L–H transition)
has been put forward [43, 69]. A similar radial electric field and scaling has been
obtained by Monte-Carlo simulations using ASCOT code [56, 70]. The predicted
radial electric field and L–H transition threshold are consistent with experimental
observations [69, 71].
 1 Mechanisma of Transverse Conductivity 41

Fig. 1.20. The radial electric field profile at the outer midplane for κ = 0.1 (discharge para-
meters ϕ|electrode = 400 V, n|−1 cm = 2.7 × 1019 m−3 , Ti |−1 cm = 32 eV): 1—calculated
profile; 2—theoretical prediction

Outside the separatrix the radial electric field is quite different and is governed by
the parallel momentum balance equation [38]. A detailed study of the electric fields
and current system outside the separatrix was done in [72].
The simulations of biasing experiments were performed by means of B2SOL-
PS5.0 transport code in [51] for the real divertor geometry. It was demonstrated that
there indeed exist three regimes which correspond to the different values of the effec-
tive perpendicular conductivity and to different profiles of the radial electric field. As
an example, in Figs. 1.20 and 1.21 is shown the radial electric field calculated accord-
ing to equations (1.84) and (1.86). The case in Fig. 1.20 corresponds to inequality
(1.83) and in Fig. 1.21 to the case when κ is close to unity.

1.12 Mechanisms of Anomalous Perpendicular Viscosity and

Viscosity-Driven Currents

In turbulent plasma the current driven by the perpendicular viscosity (or by the cor-
responding component of the Reynolds stress tensor) can be obtained by direct av-
eraging over the fluctuations. For example, in accordance with (1.15) and (1.16), in
42 V. Rozhansky

Fig. 1.21. The radial electric field profile at the outer midplane for κ = 0.7 (discharge parame-
ters ϕ|electrode = 400 V, n|−1 cm = 5 × 1018 m−3 , Ti |−1 cm = 35 eV): 1—calculated profile;
2—theoretical prediction for the boundary condition at the separatrix V|| = 0; 3—theoretical
profile for the boundary condition for velocity taken from the code

slab geometry  
∂
jy(visan) = mi nṼx Ṽy B, (1.91)
∂y
where Ṽx,y are the fluctuating velocities. For the electrostatic turbulence
∂ 
jy(visan) = − mi nkx ky |ϕk |2 /B 3 . (1.92)
∂y
k
As was mentioned in [73], the radial current driven by the turbulent viscosity is not
automatically zero for waves propagating in the direction of the main inhomogeneity
(the y direction). In other words, the viscosity-driven current of (1.92) is nonzero
if the radial wave numbers have a finite real part. For drift waves in the cylinder,
where in the absence of the radial electric field the eigenfunctions are of the form
(0)
ϕk = ϕ  exp[−iμk (y −yr )2 /2], where yr is the coordinate of the resonance surface
k
for the given ky and kz and μk = (Ln /Ls )/ρci2 with Ln and Ls being the density and
the magnetic shear radial scales. The radial current is thus [73]
∂ 
jy(visan) = mi nkx μk (y − yr )|ϕk |2 /B 3 . (1.93)
∂y
k
 1 Mechanisma of Transverse Conductivity 43

Since modes from different resonance surfaces contribute to (1.92) and the spectrum
depends on the global plasma parameters, which vary slowly, it is convenient to
average equation (1.92) over the distance L larger than the mode radial width but
smaller than the radial scale of plasma parameters

∂  L  
jy(visan) = − mi nkx ky |ϕk |2 dy  / 2LB 3 . (1.94)
∂y −L
k

The integral in (1.94) remains nonzero only if the spectrum depends on y due to a
change in the density or temperature gradient, otherwise it is zero. Hence, the sum
is the linear function on density and temperature gradients and the viscosity-driven
current has the form of (1.16).
In the presence of the radial electric field the eigenfunctions become asymmetric
with respect to the resonance surface and the integral in (1.94) becomes a linear func-
tion of the radial electric field. For the case of the ITG mode the normalized sum in
(1.94) was calculated numerically in [74] in the quasilinear approach. Therefore, the
viscosity coefficient in (1.94) was calculated for this case. It may be shown that the
viscosity coefficient η is on the order of mi nD with D being the quasilinear diffusion
 L
coefficient D = k −L kx2 |ϕk |2 dy  Im(ω)/|ω|2 /(2LB 2 ). For strong turbulence the
calculation of the perpendicular viscosity still remains an open question. Note that
in the turbulent plasma, besides the viscosity-driven current, there exists the inertia
current of the form given by (1.10) with the anomalous radial particle flux. Both
currents are generally of the same order.

1.13 Transverse Conductivity in a Stochastic Magnetic Field

The perturbation of the magnetic field may change the average conductivity across
the main magnetic field and, therefore, may influence the value of the self-consistent
electric field. On some tokamaks (TM-4 [75], TEXT [76], Tore Supra [77]) electric
fields less negative than the neoclassical or even positive were observed for spe-
cial regimes. A tentative explanation is connected with an intrinsic magnetic field
stochasticity, which can create an additional current of electrons thus reducing the
negative neoclassical electric field. On TEXT and Tore Supra the extrinsic stochas-
ticity was created by an ergodic magnetic limiter. As a result, a strong reduction in
the negative electric field was observed [78]. This effect should be interpreted in
terms of a modified transverse conductivity associated with the perturbations of the
magnetic field.
The simplest approach to calculate the current in a braided magnetic field is to
calculate the current of test particles, as was done in [79–81] for the collisionless
case similar to the calculation of the heat conductivity [82]. However, in reality the
motion of electrons in the braided magnetic field is accompanied by the emergence
of the potential, density, and temperature perturbations, which arise to provide local
quasineutrality. These perturbations strongly affect the motion of charged particles,
44 V. Rozhansky

generate the local perturbed currents, and, as a result, significantly change the physi-
cal picture and the averaged cross-field current. The effective perpendicular conduc-
tivity taking into account these effects has been calculated in [83] for the case of an
externally perturbed magnetic field. More complicated situations, when the pertur-
bations of the magnetic field are generated self-consistently by plasma turbulence
[84, 85], are not considered here.
Let us analyze the case of uniform density and temperatures to focus attention
on the effects caused by the electric field perpendicular to the flux surface. Slab
geometry is chosen for simplicity with the plasma parameters depending on the
y-coordinate normal to the flux surfaces. The unperturbed magnetic field is given
by B = Bz ez + Bx (y)ey = B[ez + bx ey ]. The x-axis corresponds to the poloidal
direction in a tokamak, the z-axis to the toroidal direction, and the y-axis to the ra-
dial direction. The unperturbed potential is denoted as ϕ0 (y), and the corresponding
electric field is E0 . Toroidaleffects are neglected here. The perturbed magnetic field
is taken in the form B  = k B   exp(i k r ), where |kz |  |kx |, |ky |. The magnetic
0k
field perturbations are assumed to be sufficiently small, so that the following condi-
tions are fulfilled: |ky by /k|| | ∼ |kx bx /k|| |  1, ky L  1, where bx,y   /B,
= Bx,y
L = |d ln ϕ0 /dy|−1 , and k|| (y) = [kx Bx (y) + kz Bz ]/B is the parallel wave vec-
tor. Therefore, we can apply the quasilinear theory based on the small parameter
|ky by /k|| | ∼ |kx bx /k|| |  1. This condition means that the displacement of the
point at the magnetic field line in the y or x directions is smaller than the transverse
spatial scale of the perturbations provided the point passes a distance k||−1 along the
magnetic field.

1.13.1 Nonstochastic Magnetic Field

The main effect is more transparent in the simplest collisional case when the mean-
free path λmfp  k||−1 in the absence of the resonance where k|| = 0 [83]. The
analysis is based on the Braginskii fluid equations. If one neglects for a while the
perpendicular currents, then from the current continuity equation, which in this case
is reduced simply to ik|| j||k = 0, it follows that the parallel current should be zero.
Since j||k = −σ|| [b  (dϕ0 /dy) + ik|| ϕ 1 ] = 0, we find the potential perturbations:
yk k

b  dϕ
(1) yk 0
ϕ =− . (1.95)
k ik|| dy
In other words, the perturbed magnetic field line remains equipotential due to the
emergence of the perturbations of the potential. The potential perturbations change
the drift velocity and the inhomogeneous part of the drifts cause viscosity and inertia
forces and, consequently, perpendicular currents.
The potential perturbations, calculated taking into account perpendicular currents
from the full current continuity equation, are
2  ib
(1) σ⊥k k⊥ y k dϕ0
ϕ = 1 − , (1.96)
k σ⊥k k⊥ + σ|| k|| k|| dy
2 2
 1 Mechanisma of Transverse Conductivity 45

where
 2  2
σ⊥k = −∇ · j⊥k /Δ⊥ ϕk = σ⊥k,i
 + σ⊥k,v
 = imi V0 kx n/B + η1 k⊥ + 4kz /B .
2 2

Here, V0 = E0 /B is the velocity of unperturbed E × B drift in the x direction

and η1 is the Braginskii viscosity coefficient (or anomalous coefficient in the tur-
bulent plasma). The deviation from the equipotential magnetic field line, i.e., the
difference between (1.95) and (1.96) determines the parallel current
  
 1 σ|| E0
jy = j|| by Re j|| by k
2 2
k
 |σ⊥k,i
 k⊥ | + σ⊥k,v
2 2
 k⊥ (σ|| k|| + σ⊥k,v
2 2 2
 k⊥ )
× |by k |2 . (1.97)
|σ⊥k,i
 k⊥ | + (σ|| k|| + σ⊥k,v
2 2 2 2 2
 k⊥ )
k

We see that the radial current is proportional to the effective perpendicular conduc-
tivity and is quadratic with respect to magnetic field perturbations.
In the presence of resonance surfaces k|| → 0 the expression for the current has
singularity and the radial current increases. This situation is analyzed in [83].

1.13.2 Stochastic Magnetic Field

When the width of the magnetic island, which is formed near the resonance flux
surface where k|| → 0, exceeds the distance between the neighboring islands, the
magnetic field becomes stochastic. The stochasticity of magnetic field lines in this
case is characterized by the Kolmogorov length of exponential divergence Lk of two
neighboring magnetic lines being initially at distance δ0 : δ = δ0 exp(z/Lk ). In the
stochastic magnetic field the local transverse transport is strongly amplified because
the magnetic field lines approach each other very close at distances larger than Lk
due to the conservation of the magnetic flux in the magnetic flux tube. This fact
strongly effects the final expression for the perpendicular current. Let us consider
the flux tube of initial size δ0 . Due to the stochastic instability the average distance
between the neighboring lines increases exponentially. But, the cross-section of a
flux tube conserves. Therefore, since the distance between the neighboring magnetic
field lines exponentially diverges in one direction, in the other direction it decreases
exponentially: δ = δ0 exp(−z/Lk ). So, if the distance between the two magnetic
field lines at z = 0 is δ0 , it becomes equal to δ0 exp(−z/Lk ) at z > Lk . Due to rather
small cross-field conductivity, the equipotentials almost coincide with the magnetic
field lines. Hence, an initial potential drop δ0 E0 between the two magnetic field lines
is applied to the very small distance δ at z > Lk . As a result, a large electric field
on the order of E0 exp(z/Lk ) arises. At such places where the magnetic field lines
approach each other close, the parallel current is short-circuited by the ion transverse
current in spite of the low value of the cross-field conductivity (see Fig. 1.22).
46 V. Rozhansky

Fig. 1.22. Schematic of the currents for the stochastic magnetic field lines

In the collisionless case, where the mean free path is much larger than Lk , the
new longitudinal scale L may be introduced. This is the scale at which the diver-
gences of the perpendicular and the parallel currents become similar: σ|| /L2 ∼
2 exp(2L/L ). The corresponding scale is
σ⊥k k⊥ k

σ||
L = 2Lk ln 2 L2
. (1.98)
σ⊥k k⊥ k

The perpendicular current is thus reduced by a factor Lk /L with respect to the cur-
rent of the test particles because the potential perturbation that makes the magnetic
field line equipotential [see (1.95)] is “washed out” at a distance L larger than Lk .
According to [83], 
2 Lk
jx ≈ e2 Dst nE0 , (1.99)
πme Te L
where Dst is the stochastic diffusion coefficient of the magnetic field lines. This
model is consistent with experimental observations of the radial electric field in the
plasma edge of TEXTOR tokamak with ergodic magnetic limiter [86].

1.14 Electric Fields Generated in the Shielding Layer between

Hot Plasma and a Solid State

The self-consistent electric fields may be generated in plasmas by injection of a

beam, which causes the return plasma current. A typical example is the contact be-
tween solid state and hot plasma in an inclined magnetic field. Such contact takes
place, for example, during hard disruptions in tokamaks, when the hot plasma from
 1 Mechanisma of Transverse Conductivity 47

Fig. 1.23. The geometry considered

the core region is moving towards the divertor plates. During hard disruptions, an in-
tense vaporization of the divertor plates takes place. The process is accompanied by
the formation of a dense and partially ionized shielding layer in front of the plates,
which significantly reduces the particle and heat fluxes reaching the divertor tiles and
thus determines the evaporation rate of the surface.
The energetic plasma particles moving along the skewed magnetic field lines are
stopped inside the layer at different depths as in Fig. 1.23. Hence, in order to produce
a return current that compensates the current carried by the plasma particles, a self-
consistent electric field is generated (ambipolarity constraint). Since the temperature
of the vapor plasma is on the order of a few eV-s, the electrical conductivity is rather
low and the resulting electric field is of considerable strength. In a tokamak with
conducting divertor plates, the local electric field has to be normal to the plates, thus
creating plasma drift across the flux surfaces [87, 88].
The electric field and the resulting drift motion were modeled self-consistently
in [87, 88]. In the applied code, the deceleration of hot particles of different energy
groups is calculated, while the cold layer is described by means of fluid equations by
taking into account all important elementary processes taking place in the plasma.
The electric field is calculated from the condition of zero net current through the
plates.
The 1D geometry is considered in Fig. 1.23, ∂/∂z = ∂/∂x = 0, Bx assumed to
 and its pro-
be zero. The current of hot particles eΓi (y) − eΓe (y) is antiparallel to B,
jection in the y-direction is e(Γi − Γe ) sin α. It is calculated by tracing the depletion
of particle groups of different energies. The potential drop between the cold and the
hot plasmas is also taken into account. The current has to be balanced by the return
current j, created by the self-consistent electric field Ey . The return current jy is
the sum of the projections of the B-parallel current j|| sin α and the B-perpendicular
current j⊥ cos α (j⊥ belongs to the plane yz, and there is also another projection jx ).
The parallel current is primarily the electron current driven by the parallel projection
of electric field E|| = Ey sin α. The perpendicular current j⊥ is associated with E⊥ .
The condition of vanishing net current j in the y-direction throughout the vapor
48 V. Rozhansky

layer can be written as

j y = j|| sin α + j⊥ cos α + e(Γi − Γe ) sin α = 0. (1.100)

The component j|| is determined from the parallel projection of the electron momen-
tum balance equation

j|| = σ|| sin α(Ey + eTe ∂ ln ne /∂y + gT e∂Te /∂y), (1.101)

where σ|| is the electrical conductivity which depends both on Coulomb and electron-
neutral collisions [1], and the coefficient gT is larger than unity due to the thermal
force. From the perpendicular component of the electron momentum balance equa-
tion (neglecting the perpendicular thermal force), we have
  
B  + ∇pe ,
j + βe j × = σ⊥ E + [V × B] (1.102)
B ene

where the Hall parameter βe = ωce /νe , ωce is the electron cyclotron frequency, νe
is the electron collision frequency, and V is the ion velocity, σ⊥ = 2σ|| . Due to
the high collision rate, the ion and neutral gas velocities are assumed to be equal.
Another relation between the perpendicular current and plasma velocity is given by
the x-component of the momentum balance equation

dρVx ∂ρVx ∂(ρVx Vy )

≡ + = [j⊥ × B]
 x. (1.103)
dt ∂t ∂y
The velocity component Vy is calculated from the y-component of the momentum
equation by simultaneously solving the complete set of fluid equations, thus deter-
mining the evolution of the vapor layer. The electric field component Ex is assumed
to be zero.
A typical result for a scenario in which a carbon divertor plate is exposed to a
thermal disrupting plasma at t = 0. The following input parameters were assumed:
Te0 = 5 keV, ne0 = 5.5 × 1018 m−3 , B = 6 T, α = 5◦ . A cold and dense shielding
layer evolves at the surface. The potential distribution at t = 50 µs is shown in
Fig. 1.24. At this time instant, ionization degrees significantly less than unity only
exist in a narrow vapor layer adjacent to the surface (y less than 2 cm) in which
the neutral atom density is on the order of 1024 m−3 . The divertor plate, as well as
the vapor layer, is biased negatively with respect to the hot plasma. The potential
difference is on the order of the energy of the incident hot electrons. Large lateral
velocities may exist. Outside of the weakly ionized region the value of Vx practically
coincides with the drift velocity Vdrift = E⊥ /B ≈ Ey /B.
The perpendicular to magnetic field currents j⊥ and jx , which are generated
in the cold plasma, are determined by the general Ohm’s law, (1.102). On the other
hand, current j⊥ results in plasma acceleration in the x-direction, according
to (1.103), while current jx is connected with the V⊥ component of the plasma veloc-
 thus also reduc-
ity. Plasma motion reduces the r.h.s. of (1.102) with respect to σ⊥ E,
 1 Mechanisma of Transverse Conductivity 49

Fig. 1.24. The 5 keV disrupting plasma: potential distribution across the vapor layer, t = 50 µs

ing the current values. Therefore, a self-consistent analysis of (1.102) and (1.103) is
necessary. Evaluating the components j⊥ , jx from the two projections of (1.102) and
substituting them into the two projections of the momentum conservation equation,
we find
dρVx σ⊥ B 2

= (E⊥ /B − Vx ) + βe V⊥ ;
dt 1 + βe2
(1.104)
dρV⊥ σ⊥ B 2

=− −βe (E⊥ /B − Vx ) + V⊥ .
dt 1 + βe 2

The pressure gradient term is neglected here with respect to the electric force term.
Let us introduce the parameter

Vy 1 + βe2 n
qr = , (1.105)
ωci Ly βe ne

where Ly is the characteristic scale of the cold plasma, ωci is the ion cyclotron fre-
quency, n is the density of the heavy component (ions plus neutrals). We assume here
∂/∂t ∼ Vy ∂/∂y ∼ Vy /Ly . For qr  1, excluding E⊥ from (1.104) and (1.105) and
making ordering, we find
βe
V⊥ ∼ qr Vx  Vx . (1.106)
1 + βe2

Furthermore, neglecting βe V⊥ with respect to Vx in (1.104), we can conclude that

(1.104) can be satisfied only if

Vx ≈ E⊥ /B; (Vx − E⊥ /B) ∼ qr E⊥ /B  E⊥ /B. (1.107)

In the opposite case, qr  1, the result depends on the βe value. For βe  1 we

have V⊥ ∼ βe Vx , and hence V⊥ again can be neglected in (1.104). In this situation,
50 V. Rozhansky

as follows from (1.104), the lateral velocity is strongly reduced with respect to the
drift velocity E⊥ /B:
Vx ∼ qr−1 E⊥ /B. (1.108)
The results of the calculations are in agreement with this qualitative analysis. Close
to the surface in the low ionization high collision region we have βe < 1 and qr > 1;
here, the calculated lateral velocity Vx is much smaller than the drift velocity.

References
[1] V.A. Rozhansky, L.D. Tsendin, Transport Phenomena in Partially Ionized Plasma (Tay-
lor & Francis, London, 2001)
[2] S.I. Braginskii, in Reviews of Plasma Physics, vol. 1, ed. by M.A. Leontovich (Consul-
tants Bureau, New York, 1965), p. 205
[3] V.M. Zhdanov, Transport Processes in Multicomponent Plasma (Taylor & Francis, Lon-
don, 2002)
[4] C.K. Goertz, J. Geophys. Res. 85, 2949 (1980)
[5] V.A. Rozhasnkii, Sov. J. Plasma Phys. 16, 483 (1990)
[6] C.L. Longmire, in Elementary Plasma Physics (Interscience, New York, 1963), p. 341
[7] V. Rozhansky, I. Veselova, S. Voskoboynikov, Plasma Phys. Control. Fusion 37, 399
(1995)
[8] P.T. Lang et al., Phys. Rev. Lett. 79, 1487 (1997)
[9] J. de Kloe, E. Noordermeer, N.J. Lopes Cardozo, A.A.M. Oomens, Phys. Rev. Lett. 82,
2685 (1999)
[10] L.R. Baylor et al., in 26th EPS Conference on Controlled Fusion and Plasma Physics,
ed. by R.M. Pick. Maastricht, The Netherlands, 14–18 June 1999. Europhysics Confer-
ence Abstracts, vol. 23J (European Physical Society, Geneva, 1999), p. 297
[11] L.R. Baylor et al., Phys. Plasmas 7, 1878 (2000)
[12] S.I. Krasheninnikov, Phys. Lett. A 283, 368 (2001)
[13] Y. Sarazin, P. Ghendrih, G. Attuel et al., J. Nucl. Mater. 313–316, 796 (2003)
[14] M. Scholer, Planet. Space Sci. 18, 977 (1970)
[15] P.B. Parks, Nucl. Fusion 32, 2137 (1992)
[16] V. Rozhansky, I. Senichenkov, I. Veselova, R. Schneider, Plasma Phys. Control. Fusion
46, 575 (2004)
[17] S.M. Dikman, L.P. Pitaevskii, Sov. Phys. JETP 51, 879 (1980)
[18] A.B. Mikhailovskii, Electromagnetic Instabilities in an Inhomogeneous Plasma (IOP,
Bristol, 1992)
[19] M. Weinlich, A. Carlson, Contrib. Plasma Phys. 36S, 53 (1996)
[20] V.A. Rozhanskij, A.A. Ushakov, S.P. Voskobojnikov, Nucl. Fusion 36, 191 (1996)
[21] F.F. Chen, Introduction to Plasma Physics and Controlled Fusion, vol. 1, 2nd edn.
(Plenum, New York, 1984)
[22] V.A. Rozhanskii, A.A. Ushakov, S.P. Voskoboynikov, Plasma Phys. Rep. 24, 777 (1998)
[23] V.A. Rozhanskij, A.A. Ushakov, S.P. Voskoboynikov, Nucl. Fusion 39, 613 (1999)
[24] P.C. Stangeby, The Plasma Boundary of Magnetic Fusion Devices (IOP, Bristol, 2000)
[25] A. Carlson, Contrib. Plasma Phys. 38S, 38 (1998)
[26] V. Rozhansky, A. Ushakov, I. Veselova, in 26th EPS Conference on Controlled Fusion
and Plasma Physics, ed. by R.M. Pick. Maastricht, The Netherlands, 14–18 June 1999.
Europhysics Conference Abstracts, vol. 23J (European Physical Society, Geneva, 1999),
p. 841
 1 Mechanisma of Transverse Conductivity 51

[27] M. Weinlich, A. Carlson, Phys. Plasmas 4, 2151 (1997)

[28] M.Z. Tokar, A.V. Nedospasov, A.V. Yarochkin, Nucl. Fusion 32, 15 (1992)
[29] A.E. Robson, P.C. Thonemann, Proc. Phys. Soc. 73, 508 (1959)
[30] G.M. McCracken, P.E. Stott, Nucl. Fusion 19, 889 (1979)
[31] K. Erdl, B. Juttner, Nucl. Fusion 25, 1413 (1985)
[32] J.R. Sanmartin, Phys. Fluids 13, 103 (1970)
[33] V.A. Rozhanskii, A.A. Ushakov, Tech. Phys. Lett. 24, 869 (1998)
[34] P. Gunn, C. Boucher, B.L. Stansfield, C.S. MacLatchy, Contrib. Plasma Phys. 36, 45
(1996)
[35] D. Bohm, in The Characteristics of Electrical Discharges in Magnetic Fields, ed. by
A. Guthrie, R.K. Wakerling (McGraw-Hill, New York, 1949), Chaps. 1, 2, 9
[36] V.A. Rozhanskii, L.D. Tsendin, Sov. Phys. Tech. Phys. 23, 932 (1978)
[37] P.C. Stangeby, J. Phys. D: Appl. Phys. 15, 1007 (1982)
[38] V.A. Rozhansky, S.P. Voskoboynikov, E.G. Kaveeva, D.P. Coster, R. Schneider, Nucl.
Fusion 41, 387 (2001)
[39] F.L. Hinton, R.D. Hazeltine, Rev. Mod. Phys. 48, 239 (1976)
[40] S.P. Hirshman, D.J. Sigmar, Nucl. Fusion 21, 1079 (1981)
[41] V. Rozhansky, M. Tendler, in Reviews of Plasma Physics, vol. 19, ed. by B.B. Kadomtsev
(Consultants Bureau, New York, 1996), p. 147
[42] V. Rozhansky, M. Tendler, Phys. Fluids B 4, 1877 (1992)
[43] V. Rozhansky, E. Kaveeva, S. Voskoboynikov, D. Coster, X. Bonnin, R. Schneider, Nucl.
Fusion 42, 1110 (2002)
[44] R.J. Groebner, Phys. Fluids B 5, 2343 (1993)
[45] K.H. Burrel, Phys. Plasmas 4, 1418 (1997)
[46] V. Rozhansky, Contrib. Plasma Phys. 46, 103 (2006)
[47] R.J. Taylor, M.L. Brown, B.D. Fried et al., Phys. Rev. Lett. 63, 2365 (1989)
[48] R.R. Weynants, R.J. Taylor, Nucl. Fusion 30, 945 (1990)
[49] L.G. Askinazi, V.E. Golant, S.V. Lebedev, V.A. Rozhanskij, M. Tendler, Nucl. Fusion 32,
271 (1992)
[50] R.R. Weynants, G. van Oost, G. Bertschinger et al., Nucl. Fusion 32, 837 (1992)
[51] V. Rozhansky, E. Kaveeva, S. Voskoboynikov, D. Coster, X. Bonnin, R. Schneider, Phys.
Plasmas 9, 3385 (2002)
[52] J. Cornelis, R. Sporken, G. van Oost, R.R. Weynants, Nucl. Fusion 34, 171 (1994)
[53] M. Van Schoor, R. Weynants, Plasma Phys. Control. Fusion 40, 403 (1998)
[54] T.E. Stringer, Nucl. Fusion 33, 1249 (1993)
[55] K. Itoh, S.-I. Itoh, M. Yagi, A. Fukuyama, Phys. Plasmas 5, 4121 (1998)
[56] J.A. Heikkinen, T.P. Kiviniemi, A.G. Peeters, Phys. Rev. Lett. 84, 487 (2000)
[57] J.A. Heikkinen, S. Jachmich, T.P. Kiviniemi, T. Kurki-Suonio, A.G. Peeters, Phys. Plas-
mas 8, 2824 (2001)
[58] R.R. Weynants, S. Jachmich, G. Van Oost, Plasma Phys. Control. Fusion 40, 635 (1998)
[59] S. Jachmich, G. Van Oost, R.R. Weynants, J.A. Boedo, Plasma Phys. Control. Fusion 40,
1105 (1998)
[60] D. Desideri et al., Czech. J. Phys. 49(Suppl. S3), 119 (1999)
[61] T.D. Rognlien, D.D. Ryutov, Contrib. Plasma Phys. 38(1–2), 152 (1998)
[62] T.D. Rognlien, D.D. Ryutov, Plasma Phys. Rep. 25, 943 (1999)
[63] T.D. Rognlien, D.D. Ryutov, N. Mattor, Czech. J. Phys. 48(Suppl. S2), 201 (1998)
[64] T.D. Rognlien, D.D. Ryutov, N. Mattor, G.D. Porter, Phys. Plasmas 6, 1851 (1999)
[65] G.J. Radford et al., Contrib. Plasma Phys. 36(2–3), 187 (1996)
[66] R. Zagorski, H. Gerhauser, H.A. Claasen, Contrib. Plasma Phys. 38(1–2), 61 (1996)
[67] T.D. Rognlien, Plasma Phys. Control. Fusion 47, A283 (2005)
52 V. Rozhansky

[68] R. Zagorski et al., in 28th EPS Conference on Controlled Fusion and Plasma Physics,
ed. by C. Silva, C. Varandas, D. Campbell. Madeira, Portugal, 18–22 June 2001. Eu-
rophysics Conference Abstracts, vol. 25A (European Physical Society, Geneva, 2001),
p. 2069
[69] V. Rozhansky, Plasma Phys. Control. Fusion 46, A1 (2004)
[70] T.P. Kiviniemi et al., Phys. Plasmas 10, 2604 (2003)
[71] Y. Andrew et al., Plasma Phys. Control. Fusion 46, 337 (2004)
[72] V. Rozhansky, E. Kaveeva, S. Voskoboynikov, D. Coster, X. Bonnin, R. Schneider, Nucl.
Fusion 43, 614 (2003)
[73] P.H. Diamond, Y.B. Kim, Phys. Fluids B 3, 1626 (1991)
[74] G.M. Staebler, R.R. Dominguez, Nucl. Fusion 33, 77 (1993)
[75] V.I. Bugarya, A.V. Gorshkov, S.A. Grashin et al., Nucl. Fusion 25, 1707 (1985)
[76] D.L. Brower, W.A. Peebles, N.C. Luhman, Nucl. Fusion 27, 2055 (1987)
[77] P. Ghendrih, A. Grosman, H. Capes, Plasma Phys. Control. Fusion 38, 1653 (1996)
[78] X.Z. Yang, B.Z. Zhang, A.J. Wooton et al., Phys. Fluids B 3, 3448 (1991)
[79] R.W. Harvey, M.G. McCoy, J.Y. Hsu, A.A. Mirin, Phys. Rev. Lett. 47, 102 (1981)
[80] A. Gurevich, K. Zybin, Y.N. Istomin, Nucl. Fusion 27, 453 (1987)
[81] T. Kubota, S.-I. Itoh, Phys. Plasmas 2, 3368 (1995)
[82] A.B. Rechester, M.N. Rosenbluth, Phys. Rev. Lett. 40, 38 (1978)
[83] I. Kaganovich, V. Rozhansky, Phys. Plasmas 5, 3901 (1998)
[84] R.E. Waltz, Phys. Fluids 25, 1269 (1982)
[85] P.W. Terry, P.H. Diamond, T.S. Hahm, Phys. Rev. Lett. 57, 1899 (1986)
[86] C. Busch et al., in 33rd EPS Conference on Control Fusion and Plasma Physics, Rome.
ECA, vol. 30I (2006), p. 2.167
[87] L. Lengyel, V.A. Rozhansky, I. Veselova, P. Lalousis, Contrib. Plasma Phys. 38, 124
(1998)
[88] L.L. Lengyel, K. Lackner, P.J. Lalousis, P.N. Spathis, V.A. Rozhanskij, I.Y. Veselova,
P.B. Parks, Nucl. Fusion 38, 1435 (1998)
2
Correlations and Anomalous Transport Models
O.G. Bakunin

2.1 Introduction
At present, the major obstacle on the way to the realization of controlled thermonu-
clear fusion in closed magnetic configuration devices is commonly attributed to the
existence of anomalous energy losses due to particle and energy transport across a
confining magnetic field. The anomalous transport of particles is usually related to
the turbulent character of plasma behavior [1–5]. In spite of considerable effort, this
problem still remains to be solved [6–12].
The equations describing diffusion phenomena [13–16] are among the key tools
for investigating transport processes in plasmas. The ever-increasing complexity of
the problems requires the development of more and more elaborate and diverse
diffusion models [17–22]. The relation between heat conduction and random walk
processes was established as early as the beginning of the 20th century [13]. At the
first stage of research in this field, the main problem was that of calculating the dif-
fusion coefficient (thermal conductivity). The investigation of turbulent diffusion in
the atmosphere had led to the use of scalings, correlation functions, and new equa-
tions that differ substantially in structure from the conventional diffusion equation
[18–20]. The research on atmospheric turbulence alone allows one to understand the
hierarchic nature of turbulent scales and the importance of accounting for anisotropy.
The intensive investigation of processes in strongly magnetized plasma, which was
started in the middle of 20th century, essentially expands the notion about both trans-
port processes and the nature of turbulence. Thus, transport models in stochastic
magnetic fields and two-dimensional turbulent models were developed. It was re-
vealed that transport processes in turbulent plasmas are often nondiffusive in na-
ture. New forms of equations describing transport processes have constantly been
searched for since the first studies on quasi-linear theory [23, 24]. The description
of diffusion under strongly nonequilibrium conditions in highly turbulent plasma re-
quired the use of equations that take into account memory effects and the nonlocal
nature of transport processes [18–20].
54 O.G. Bakunin

The objective of this paper is to consider various methods for constructing such
equations, ranging from those in the quasi-linear approximation [23, 24] to those
with fractional derivatives [19–22]. The topics to be discussed include the telegraph
equation, the Levy–Khintchine distribution, and the Kohlrausch slow relaxation law
and continuous time random walks. Use will be made of some important notions
belonging to theoretical probabilistic analysis: the return probability, the self-inter-
section probability, and the probability of staying in a trap.
Another important aspect of the problem of describing turbulent transport is us-
ing the correlation methods of analysis [25–30]. Even from the common point of
view, one can see that the correlation function is a more relevant tool to investigate
a constant diffusion coefficient. Long-range correlations are responsible for anom-
alous transport. The methods of direct calculations and the diffusive approximation
of the correlation effects are represented. One can see that the analysis of correlation
effects and the interrelation between the diffusion coefficient and the autocorrelation
function have been of major importance. It would therefore be instructive to trace
the relation between Taylor’s paper [31] that introduced the autocorrelation function
and the papers on percolation diffusion [17], which are new trends of the turbulent
transport theory.
It is well known that scaling representation, which was initially developed by
Richardson and Kolmogorov, plays an important role in turbulence. A large number
of researchers have used the ideas of scaling laws and fractality [32–48] to describe
properties of turbulent transport. This is not surprising, because turbulent diffusion
models differ significantly from one-dimensional transport models. Thus, the pres-
ence of vortex structures in turbulent flows and plasma requires the consideration
of hierarchies of spatial and correlation scales. The system of convective cells is
one of the typical examples of quasi-regular vortex structures. On the one hand, the
space between convective rolls is responsible for convective transport. On the other
hand, in vortex structures trapping leads to subdiffusive transport. Therefore, ob-
taining the expression for the effective diffusion coefficient is a nontrivial problem.
Often, several different types of transport are present simultaneously in turbulent
diffusion [38, 39, 48], making it important to take into account initial diffusivity and
anisotropy. The anisotropy of the medium is thought to be due to the presence of a
strong magnetic field or shear convective flows.
The sealing approach to describe long-range correlation was essentially devel-
oped in papers on the theory of phase transitions and critical phenomena [49–57].
Thus, the power form dependence of correlation scale on the small parameter, which
describes the closeness of a system to the percolation threshold, is a fairly universal
model to describe anomalous transport. In such an approach, the existence of very
long (percolation) streamlines in two-dimensional random flows allows one to use
the well-developed mathematical methods of transport analysis in percolation sys-
tems [17]. A detailed analysis of the more important results obtained in this field is
presented in this chapter.
The review is structured as follows. It contains essentially seven parts. The first
part covers Sects. 2.1 and 2.2. Here, the diffusion equations for the description of
nonlocal effects are considered.
 2 Correlations and Anomalous Transport Models 55

Section 2.1 is devoted to the consideration of models using the conventional rep-
resentation of the diffusion equations and the definition of the correlation representa-
tion of the turbulent diffusion coefficient. Thus, the models by Richardson [58] and
Batchelor [59] are based on diffusion coefficient approximations that are in agree-
ment with the scaling for relative diffusion [58]. The Davydov idea [60] to describe
the turbulent transport of a passive scalar by a telegraph equation is treated in this
section.
In Sect. 2.2, the importance of using integral representation for the description of
nonlocal effects is pointed out. The Levy–Khintchine approach based on the power
representation of the Fourier-transformation of the nonlocal functional kernel is con-
sidered [61]. The Monin idea [62] of agreement between the transport equations in
the Levy–Khintchine form and the Kolmogorov law [63] for isotropic turbulence is
discussed in Sect. 2.2.
The second part consists of Sects. 2.3–2.8. Here, we deal with the initial diffu-
sivity effects and the quasi-linear approximation of the nonlinear equations.
The Corrsin conjecture to describe the relationship between Lagrangian and
Eulerian correlation functions is introduced in Sect. 2.3 [28]. It is an important aspect
of the problem that the Corrsin conjecture relates correlation effects to the seed diffu-
sion nature of transport. Later, this idea was developed in Dupree’s and Kadomtsev–
Pogutse’s papers [64–67]. Then in Sect. 2.4 we consider the effects of molecular
diffusion, which lead to the power approximation of the correlation function [68].
Another important aspect is anisotropy effects. In this connection, the double diffu-
sion regime in a stochastic magnetic field is considered [69]. The Howell represen-
tation [73] of the effective coefficient of diffusion is observed.
In Sect. 2.5, the heuristic Taylor method to obtain such an equation is considered.
The approximation suggested in [117] has become an important step in the develop-
ment of description methods of anomalous transport and complex correlation effects.
In Sect. 2.6, we consider anomalous transport in the system of random shear
flows [72] where the nontrivial character of diffusivity in an anisotropic system is
manifested. The quasi-linear equations are derived in Sect. 2.7 [23, 24]. We then
discuss the short-range and long-range effects in terms of the quasi-linear approach
[20]. The possibility of using the quasi-linear approximation for the description of
stochastic magnetic field diffusion is discussed [67, 70].
Section 2.8 treats the diffusive renormalization of correlation effects. We will
consider the direct calculations of the correlation function [71], the Corrsin conjec-
ture [72], and the renormalized quasi-linear equations [67]. The focus of Sect. 2.9 is
the derivation of the expression for the effective diffusion coefficient, which is based
on the balance of convective and diffusive fluxes, in the convective cell system.
The third part covers Sect. 2.10, where the problems of relations between sto-
chastic instability [74] and transport effects in the stochastic magnetic field are ana-
lyzed on the basis of Rechester and Rosenbluth’s models [75]. The relations between
the Rechester–Rosenbluth model and the Kadomtsev–Pogutse approach are treated
here.
The fourth part consists of Sects. 2.11 and 2.12. This part deals with the fractal
and percolation approaches to describe the transport effects.
56 O.G. Bakunin

Several important definitions from fractal theory [56, 57] are introduced in
Sect. 2.11. We then consider the fractal interpretation of Richardson’s [58] and Kol-
mogorov’s [63] laws by using the notion of fractal dimensionality [76–78].
Section 2.12 is devoted to the consideration of percolation methods for describing
transport effects. Here we discuss the fractal representations of important formulas of
transport theory, the percolation renormalization technique [79], and the convective
cells problem [33, 80, 81] as the simplest examples to describe transport effects in
the presence of the structure.
The fifth part covers Sects. 2.13–2.17. Here, percolation methods for describ-
ing the anomalous transport in random two-dimensional flows are observed on the
grounds of both the monoscale and multiscale approaches [17]. We point out the
importance and universality of renormalization methods to describe turbulent trans-
port in terms of percolation theory. The model of steady flow percolation [82], time-
dependent percolation [83], and the influence of drift effects [84] are considered.
Section 2.17 of this review deals with the multiscale approach [85, 86] that is
applied to the description of percolation effects. The relationships between the expo-
nents (the hull exponent, the correlation exponent, the Hurst exponent) are consid-
ered [17].
The sixth part consists of Sects. 2.18–2.20. Here, the memory and trap effects
are represented on the basis of both fractal and continuous time random walk ap-
proaches.
Section 2.18 treats the problem of subdiffusive regimes and the trap approxi-
mation to describe anomalous transport effects [18–20]. The Balagurov–Vaks trap
model is considered [87]. The simplest fractal representation of subdiffusive be-
havior is explored, as are comb structures. Section 2.19 describes the continuous
time random walk approach for the description of nonlocal and memory effects [18–
20]. Using the relaxation function in the power form leads to the consideration of
transport fractional differential equations. Fractional differential equations with the
correlation exponent as a parameter are derived and the relationships between the
correlation exponent and the Hurst exponent are obtained in Sect. 2.20.
The last part consists of Sect. 2.21. Here, we discuss the relationship between the
conventional space approach to transport and the phase-space approach. The Hamil-
tonian approach gives the advantage of using degrees of freedom to treat nonlocality
and memory effects in the framework of phase-space. The kinetic model provides the
possibility of describing ballistic modes and establishing the relationship between
different exponents and distributions [96]. We consider the phase-space modification
of the Corrsin conjecture, sticky island exponent, and nonlocal velocity distribution
function.

2.2 Turbulent Diffusion and Transport

In spite of considerable progress in the understanding of anomalous transport, many

aspects of the first papers in this region remain present-day. Thus, at the first stages
 2 Correlations and Anomalous Transport Models 57

of research on turbulent diffusion processes it was proposed using correlation func-

tions, modifying the conventional diffusion equation, and searching scaling laws that
describe nondiffusive transport. In this section we will discuss the aforementioned
ideas using the classical papers by Taylor [31], Richardson [58], Davydov [60] and
Batchelor [59].

2.2.1 The Correlation Function and the Taylor Diffusivity

In this section we briefly consider defining the turbulent diffusion coefficient. Tay-
lor published a paper [31] (1921) in which he suggested a formula showing a direct
relationship between the diffusion coefficient and the autocorrelation function of ve-
locity. Actually, a new “tool” was suggested for the analysis of diffusion processes.
Following ideas in Langevin’s and Einstein’s papers [88, 89], Taylor wrote a
stochastic equation of motion of a test Lagrangian particle in a random field,
 t
x(t) = V (x0 , τ ) dτ, (2.1)
0

where x(t) is the coordinate of the particle at time t, V (x0 , t) is the random function
of Lagrangian velocity, and x0 is the initial coordinate of the Lagrangian particle.
The purpose of his calculation was the mean square of a random displacement of the
particle  t  t 
 2  
x = x(t)x(t) = V (t1 ) dt1 V (t2 ) dt2 . (2.2)
0 0
The brackets  indicate an average over the ensemble of Lagrangian trajectory. Here,
we omit the calculations described in detail in [27–29]. The final result of the calcu-
lations can be written in the form
 t  t1
 2
x =2 dt1 C(τ ) dτ , (2.3)
0 0

where C(τ ) is the Lagrange correlation function,

 
C(τ ) = V (x0 , t)V (x0 , t + τ ) . (2.4)

A somewhat different form of this formula was suggested by Kampe de Feriet in [90]:
 t
 2
x =2 (t − τ )C(τ ) dτ . (2.5)
0

Here, the symmetry of integral expression (2.3) is used to simplify the representation
of the formula. Estimates of the turbulent diffusion coefficient in the Taylor approach
lead to the expression
 t  t
1 d  2 d
DT = x = (t − t  )C(t  ) dt  = C(τ ) dτ ≈ V02 τ. (2.6)
2 dt dt o 0
58 O.G. Bakunin

Here, V0 is the characteristic velocity and τ is the correlation time. The specific
form of the expression for the turbulent diffusion coefficient DT (t) depends on the
behavior of the correlation function C(t). This differs significantly from the “graded”
representation of the familiar Fick’s Law [7–9] with D0 ∝ Δ2COR /τ . Here, ΔCOR is
the spatial correlation scale. Usually, the exponential form is used:
 
|t|
C(t) = V0 exp −
2
, (2.7)
τ

where V0 is the characteristic velocity and τ is the characteristic correlation time.

Such a representation of the correlation function for t  τ is in agreement with the
rigorous results from the stochastic equation theory [13, 14, 16].
There are two asymptotic cases of significance. In the first, when t  τ , upon
substitution of (2.7) into (2.5) and simple transformations one can easily obtain
 ∞
 2
x = 2V02 tτ − 2 τ C(τ ) dτ ≈ 2V02 τ t. (2.8)
0

This expression coincides with the well-known Einstein law for the root-mean-square
displacement, R 2 ∝ t.
In the second case, when t → 0, we can use the simplest approximation of the
correlation function in the form
 
t2
C(t) ≈ V0 1 − 2 .
2
(2.9)
τ

This is an important correction of representation (2.7), from a formal point of view,

because we have to take into account the rigorous condition of applicability of cor-
relation approximations [27],

d
C(t) → 0. (2.10)
dt t→0

Upon substitution of (2.9) into formula (2.5), one obtains the ballistic motion law in
the form  2
x = 2V02 t 2 . (2.11)
Another important relationship, which will be used in the subsequent discussions, is
the expression
d 2  2 
x = 2C(τ ). (2.12)
dt 2 t=τ
This formula can be obtained by differentiation of expression (2.6).
Even from general considerations, it is clear that the correlation function is a
more relevant tool of investigation than the constant diffusion coefficient. In the next
sections, we will show that the development of correlation ideas had an essential
influence on the form of diffusion equations.
 2 Correlations and Anomalous Transport Models 59

2.2.2 The Richardson Law

The problem formulated by Taylor [31] appears to be especially actual in relation to

the Richardson investigations of turbulent diffusion that were carried out in 1926 [58].
The author of [58] discovered that the laws of atmospheric diffusion essentially dif-
fer from conventional expression (2.8). An analysis of experimental results has led
to the expression for the mean square separation of a pair of marked particles,
1 d 2   2/3
Y (t) ≈ const · Y 2 (t) . (2.13)
2 dt
Here, Y (t) is the separation between two particles that are situated at the points x1 (t)
and x2 (t),
Y (t) = x2 (t) − x1 (t). (2.14)
Dependence (2.13) shows the accelerating character of particle relative motion. The
expression can be represented in the scaling form:
 2 
Y (t) ∝ t 3 . (2.15)

Result (2.15) is not trivial because it differs significantly even from the ballistic scal-
ing (2.11). Indeed, from the formal standpoint we can expect that
 2   2     
Y (t) = x1 (t) − 2 x1 (t)x2 (t) + x22 (t) . (2.16)

Destroying correlations in time leads to the result that is in accord with the following
estimates:  2 
Y (t) ≈ 2(2DT )t. (2.17)
However, from the point of view of the scaling law (2.13) we deal with the depen-
dence  2/3
DR ≈ Y 2 (t) . (2.18)
This expression, in fact, mirrors the nonlocal character of transport effects in the con-
ditions of atmospheric turbulence, since the separation between the diffusing parti-
cles significantly changes only under the influence of eddies comparable in size to
interparticle separation.
Richardson suggested using the diffusion equation for the description of the prob-
ability density evolution F to find two initially close particles at a distance l from one
another at the moment t:
∂F (l, t) ∂ ∂F (l, t)
= DR . (2.19)
∂t ∂l ∂l
In the framework of the offered scaling law (2.13), the expression for DR (l) takes
the form
DR (l) ≈ l 4/3 . (2.20)
60 O.G. Bakunin

This result was later confirmed in the framework of the theory of uniform isotropic
turbulence [63]. Kolmogorov and Obuchov showed in their articles [63, 91] that the
rate of energy dissipation εK is the only dimensional characteristic in a wide interval
of scales l. Then it is possible to compose the scaling laws based on the dimensional
character of the value εK = [L2 /T 3 ] and the variable k that characterizes the spatial
scale k ≈ 1/ l(k) = [1/L]. Simple calculations then yield the dimensional estimate
for the Richardson coefficient:

2 2/3
L 1/3 k 1/3 1 1/3
DR (l) = ≈ εK ≈ εK 4/3 ≈ εK l 4/3 . (2.21)
T k2 k
Thus, the idea of describing turbulence by the hierarchy of eddies of different scales
[63] has obtained its first experimental confirmation.
It is important to note that expression (2.13) suggested by Richardson is in ac-
cord with the experimental data in a wide spectrum of parameters. This quite jus-
tifies the usage of the expression “the Richardson law”. The papers by Taylor and
Richardson undoubtedly opened up a fundamentally new avenue for research and
had a profound effect on the subsequent development of the theory of transport
processes.

2.2.3 The Davydov Model of Turbulent Diffusion

The nonlocal character of transport that has been investigated by Richardson is man-
ifested not only for the relative diffusion of particles. The problem of the diffusion
description of a single test particle in the field of turbulence also leads to the necessity
to take into account the interaction between different scales l(k). Such an approach
naturally requires considerable modification of Fick’s diffusion equation:

∂n ∂ 2n
= D 2. (2.22)
∂t ∂x
Here, n is the particle density and D is the conventional coefficient of diffusion. One
of the first models to describe turbulent diffusion is the Davydov model [60], which
is based on the telegraph equation

1 ∂n ∂ 2 n ∂ 2n
+ 2 = V2 2, (2.23)
τ ∂t ∂t ∂x
where V is the velocity scale and τ is the correlation time. From the dimensional
standpoint, the use of this expression permits one to obtain scaling laws for the mean
square displacement of a particle in the ballistic form
 2
x ∝ t 2. (2.24)

Note that Maxwell [18] was the first to suggest the hyperbolic model of heat-conduc-
tivity for the description of the finite velocity of perturbation spreading. This corre-
sponds fairly well to his investigations of electromagnetic theory.
 2 Correlations and Anomalous Transport Models 61

Davydov used the phenomenological set of equations for the particle density
n(x, t):
∂n(x, t) ∂q(x, t)
+ = 0; (2.25)
∂t ∂x
∂q q0 − q
= . (2.26)
∂t τ
Here, q is the particle flux. It is natural to use the classical expression for the particle
initial flux:
∂n(x, t)
q0 (x, t) = −D . (2.27)
∂x
Formal manipulations with this set of equations yield telegraph equation (2.23). The
author of [60] suggested using (2.23) to take into account the finite particle velocity
V during the molecular diffusion. The classical parabolic diffusion equation follows
from telegraph equation (2.23) in the limit

τ → 0; D ≈ V 2 τ → const. (2.28)

The physical meaning of the representation proposed by Davydov for the particle
flux q can be easily clarified by writing the formal solution
 t  t
 dt  ∂n  dt 
q(x, t) = q0 (x, t  ) exp −(t − t  ) =− D exp −(t − t  ) . (2.29)
0 τ 0 ∂x τ
Obviously, such an expression for the particle flux contains memory effects. After
Davydov, this formula was generalized in many studies in such a way as to replace
the exponential function by an arbitrary memory function M(t − t  ),
 t
dt 
q(x, t) = q0 (x, t  )M(t − t  ) . (2.30)
0 τ
In this general case one obtains the diffusion equation in the form

∂n(x, t) t ∂ 2 n(x, t  )  dt

= D M(t − t ) . (2.31)
∂t 0 ∂x 2 τ
Later, the telegraph equation in the form (2.23) was often applied to describe turbu-
lent diffusion [27, 92–95].
From the modern point of view such an approach looks fairly naive. However,
in essence, the idea of using the additional derivative in the equations describing the
anomalous character of turbulent diffusion was clearly formulated by Davydov as
early as 1934 [60]. At present, not only are conventional partial derivatives used, but
fractional derivatives are also used, better mirroring the essence of the nonlocality
and memory effects because they have the integral character of the operator [18–22]

∂ξ n ∂ζ n
, , .... (2.32)
∂x ξ ∂t ς
62 O.G. Bakunin

Here, ξ and ζ are the fractional parameters of the problem. Moreover, this approxi-
mation method is also applied to the description of strong nonequilibrium processes
in the framework of kinetic equation [96, 97] for the distribution function f (V , x, t).
Here, V is the velocity. Thus, it is suggested replacing the Fokker–Planck diffusive
operator ∂ 2 f/∂V 2 by the fractional derivatives ∂ ξ f/∂V ξ that reflect the nonlocal
character of relaxation in the phase space. In the next sections of the paper we will
discuss these problems in detail.

2.2.4 The Batchelor Approximation for the Diffusion Coefficient

The approximation suggested by Richardson (2.20) corresponds to his ideas about

the hierarchical and nonlocal character of turbulent transport. Thus, he related nonlo-
cality to the increasing scale of eddies taking part in transport processes. Therefore,
in his approach the diffusion coefficient DR is the function of the interparticle dis-
tance l. However, there exist alternative possibilities. Batchelor [59] considered the
problem from a different point of view. In his model the diffusion coefficient DR
is the result of statistical averaging over the ensemble of different scales. Hence, he
proposed using the temporal dependence for the definition of DR . In the framework
of this approach the dimensional consideration yields the expression

Y 2 (t)  2 2/3
DR (t) ≈ ≈ l (t) ∝ t 2. (2.33)
t
The equation for the probability density then takes the following form, which is
similar to the conventional one (2.22) but with the time-dependent coefficient of
diffusion:
∂F (l, t) ∂ ∂F
= DR (t) . (2.34)
∂t ∂l ∂l
After the simplest analysis it becomes clear that the Richardson model and the Batch-
elor model lead to different results in spite of the underlying law (2.13). Thus, in the
conventional diffusion equation (2.22) the law of temporal relaxation of the function
F in the Fourier form corresponds to

F̃k (t) ∝ exp(−t), (2.35)

whereas in the Batchelor model we deal with stronger damping:


F̃k (t) ∝ exp −t 3 . (2.36)

Here, F̃k (t) is the Fourier transformation of the function F (x, t) over the variable x.
It is obvious that the characters of the solutions suggested by Richardson and Batch-
elor describing the probability density evolution are also different. Considering the
model with a point-source of particles, one can obtain for the Richardson model [58]
 9/2  
8 9 9l 2/3
F (l, t) = exp − . (2.37)
315π 8/2 4t 4t
 2 Correlations and Anomalous Transport Models 63

Under analogous conditions (the model with a point-source) for the Batchelor model
one obtains [59]
 3/2  
1 l2
F (l, t) = exp − . (2.38)
2πl 2 (t) 2l 2 (t)

Note that the arguments in favor of one type or another of the diffusion coefficient
have a qualitative character in both these cases. Moreover, the “combination” of both
these approaches is possible, if one supposes that DR can depend on both the inter-
particle distance l and time t:

DR (t, l) ≈ t φ l ϕ . (2.39)

To save the Richardson law we need to take into account the relationship between
exponents φ and ϕ:
2φ + 3ϕ = 4. (2.40)
Then, the case φ = 0, ϕ = 4/3 corresponds to the Richardson law and the case
φ = 2, ϕ = 0 corresponds to the Batchelor supposition. Thus, Okubo [98] suggested
a mixed algebraic representation for the diffusion coefficient:

DR (t, l) ≈ tl 2/3 . (2.41)

The three-dimensional solution for the point-source at t = 0 is given by [98]

 
− 3(1+φ) l 2−ϕ
F (l, t) = const · t 2−ϕ exp −const · 1+φ . (2.42)
t

Unfortunately, it is impossible to decide what is a correct equation, if one looks at

this problem from the conventional diffusion point of view, because the physical
arguments from Kolmogorov and Obukhov lead to an explanation in terms of the
hierarchy of scales, whereas Richardson and Batchelor deal with the local diffusive
equation with partial differentials. However, these classical papers [58–60] formu-
lated problems that allow us to develop theoretical methods of anomalous transport
description that are based on the analysis of correlation effects and scaling laws.

2.3 Nonlocal Effects and Diffusion Equations

The nonlocal nature of relative diffusion has stimulated the search for equations that
differ significantly from conventional diffusion equations. An elegant integral equa-
tion corresponding to this problem was suggested by Einstein [89]. The use of this
equation in combination with the scaling ideas has led to the necessity to consider a
distribution function that differs essentially from the Gauss function. A new type of
distribution, called the Levy–Khintchine distribution, is now one of the basic tools
for researching anomalous transport.
64 O.G. Bakunin

2.3.1 The Functional Equation for Random Walks

For the Richardson law Kolmogorov and Obuchov [63, 91] obtained dimensional
estimates, which give qualitative explanations of the nonlocality transport effects of
turbulent diffusion in terms the interaction of different scales. However, the nonlocal
effects can also be described by means of the random walk model. Thus, besides
the different phenomenological methods of improvement of the diffusion equation,
there exists a possibility to use the integral equation to describe the random walk
processes.
As early as 1905, Albert Einstein obtained a functional equation for the particle
density solely on the basis of the general ideas about the process of random walk
[89]:  +∞
n(x, t + τ ) = W (y)n(x − y, t) dy, (2.43)
−∞
where W (y) is the probability density of undergoing a jump y. This fundamentally
nonlocal equation can be made local by reducing it to a diffusion equation. Assuming
that the time scale τ is short and the jump y is small, Einstein arrived at the classical
diffusion equation. In this way, he used the expansions
∂n
n(x, t + τ ) = n(x, t) + τ + ···, (2.44)
∂t
∂n y2 ∂ 2n
n(x + y, t) = n(x, t) + y+ + ···. (2.45)
∂x 2 ∂x 2
Simple calculations yield
 ∞  
∂n ∂n ∞ ∂ 2n ∞ y2
n+ τ =n W (y) dy + W (y)y dy + 2 W (y) + · · · .
∂t −∞ ∂x −∞ ∂x −∞ 2
(2.46)
Assuming that the function W is symmetric, W (y) = W (−y), and specifying the
normalization condition  +∞
W (y) dy = 1, (2.47)
−∞
one obtains the conventional diffusion equation
 +∞
∂n ∂ 2n 1 y2
= D 2, where D = W (y) dy. (2.48)
∂t ∂x τ −∞ 2
Note that the number of terms in expansions (2.44), (2.45) was chosen in a physi-
cally meaningful way. Based on the relationship characterizing the average behavior
of Brownian particles, x 2  ≈ R 2 ∝ t. We can estimate the orders of the terms
for t → ∞ in the expansions as follows: n ∝ 1/R. This corresponds to the one-
dimensional case. Then one can obtain
∂n n ∂ 2n n
n ∝ t −1/2 , ∝ ∝ t −3/2 , 2
∝ 2 ∝ t −5/2 ,
∂t t ∂t t
(2.49)
∂n n ∂ 2n
∝ ∝ t −1 , ∝ t −3/2 .
∂x R ∂x 2
 2 Correlations and Anomalous Transport Models 65

Retaining only two terms in expansions (2.44) and (2.45) each results in a telegraph
equation. However, this does not indicate that the telegraph equation is invalid. The
reason is that, in this case, the effects of the finite propagation velocity of the pertur-
bations come into play, which are absent in the classical diffusion model.
The integral approach was further developed in the papers by Smoluchowski,
Chapman, and Kolmogorov [99–101]. A key element in their approach is Markov’s
postulate [13, 14] that the length of the jump y is independent of the prehistory of
motion. To describe the nonlocal effects, just the integral form of the equations is
important.
Using expansion (2.44) of functional (2.43), we can readily obtain the Smolu-
chowski equation [99]
 +∞
∂n(x, t)  
= K(x  , x)n(x  , t) − K(x, x  )n(x, t) dx  . (2.50)
∂t −∞

Here, K(x, x  ) dx dx  is the probability for a particle at position x at time t to pass

over to the interval x  + dx  during the time interval dt. We introduce the functional
 +∞
  
G(x , x) = K(x , x) − δ(x − x ) K(x, x  ) dx  . (2.51)
−∞

For a uniform isotropic medium, we have G(x  − x) = G(|x − x  |). In the simplest
case under consideration, this functional has the form
 +∞
∂n
= G(x − x  )n(t, x  ) dx  . (2.52)
∂t −∞

This representation reflects the nonlocal character of transport and at the same time
it has a close relation to the conventional diffusion equation (2.22). It is possible to
consider several analytical functions G(x) to find some solution of this equation [15].
As an example, one can form such an approximation on the basis of Poisson’s proba-
bilistic law [13, 14]. But there is another good way that leads to new and very fruitful
research trends, which are especially relevant for turbulent diffusion problems.

2.3.2 Nonlocality and the Levy Distribution

Functional (2.52) is linear and it is more convenient to switch to the Fourier repre-
sentation for n(x, t) with respect to the variable x. Formal manipulations yield
∂ ñk (t)
= G̃k ñk (t), (2.53)
∂t
which indicates the absence of memory effects for the Fourier harmonics. Here, G̃k
and ñk (t) are the Fourier transformations of the functions G(x) and n(x, t) with
respect to the variable x. The expression

G̃k ñk = −Dk 2 ñk (2.54)

66 O.G. Bakunin

corresponds to the classical diffusion equation, where D is the conventional diffusion

coefficient. In the case of telegraph equation (2.23), the memory effects were taken
into account (see (2.31)):
 t
∂ ñk (t) dt 
= −k 2
ñk (t)M(t − t  ) = −k 2 M(t) ∗ ñk (t), (2.55)
∂t 0 τ
where the asterisk indicates the convolution operation.
Applying the Laplace transformation in time, we obtain the following expression
for the telegraph equation with memory:

˜ Dk 2
G̃(k, s) = − . (2.56)
1 − isτ
˜
Hereafter, G̃(k, s) will denote both the Fourier and Laplace transformations of the
function G(x, t) with respect to the variables x and t. It is an easy matter to combine
the memory and nonlocality effects into a common expression containing a convolu-
tion:
 t
∂ ñk (t) dt 
= −k 2
ñk (t)D̃k (k, t − t  ) = −k 2 D̃k (k, t) ∗ ñk (t). (2.57)
∂t 0 τ
Performing the Laplace transformation in time gives the transition from the conven-
tional result to the new one:

−Dk 2 → −k 2 D̃˜ k,s (k, s). (2.58)

In the theoretical probabilistic approach, however, this heuristic method is unsatis-

factory. Below, we will consider this point in more detail.
The approach based on (2.53) was developed by Levy and Khintchine [61], who
used the approximate equation of the form

∂ ñk (t)
= −k αL ñk (t), 0 < αL ≤ 2. (2.59)
∂t
It is easy to see that, for αL = 2, we are dealing with a Gaussian distribution (corre-
sponding to a conventional diffusion equation). Some other analytic distributions are
also known.
For the case αL = 1, we obtain the Cauchy distribution [102].
For the case αL = 3/2, one arrives at the familiar Holtsmark distribution [13].
For the case αL = 1/2, we have the Levy–Smirnov distribution [18].
For the case αL = 2/3, we obtain the Smirnov distribution [18]. In this context,
it is important to note that all the probability densities with αL < 2 have power-law
tails. Another important feature is that the second and higher order of moments of the
distributions with 1 ≤ αL < 2 and all moments of the distributions with 0 < αL < 1
diverge.
 2 Correlations and Anomalous Transport Models 67

Representation (2.59) is sufficient to consider the important models of anomalous

diffusion, which are often described by the scaling law,
 2 1/2
x ≈ R ∝ tH . (2.60)

Here, H is the Hurst exponent [18–22]. In the case of classical diffusive behavior we
find H = 1/2. The cases 0 < H < 1/2 correspond to the subdiffusive behavior. The
cases 1/2 < H < 1 correspond to the superdiffusive behavior. There is a relationship
between the Hurst exponent H and the Levy–Khintchine exponent aL that is the
parameter of the power approximation (2.59):
1
H = . (2.61)
αL
There is also a very interesting result which follows from the Fourier representation
of density n(x, t):  
 2 1/2 ∂ ∂
x =− ñk (t) . (2.62)
∂k ∂k t=0
This expression is very useful for relating scaling laws to probabilistic approxima-
tions in the Levy–Khintchine term.

2.3.3 The Monin Fractional Differential Equation

Monin [62] used the Einstein–Smoluchowski functional given in (2.52) and (2.59)
to describe turbulent diffusion in the atmosphere. That paper anticipated the devel-
opment of modern ideas of using additional fractional partial derivatives in diffusion
equations.
Monin was guided by Kolmogorov’s ideas about the universal properties of well-
developed isotropic turbulence [63]. In the corresponding formulation of the prob-
lem, all statistical parameters are determined exclusively by the scale length lk ≈
1/k ≈ [L] and the mean energy dissipation rate εK = [L2 /T 3 ]. Here, L and T char-
acterize the physical dimensionality of space and time. In the framework of Fourier’s
representation (2.59) there is a single “uncertain parameter” aL , which defines the
power form of the kernel of the nonlocal functional. Based on dimensional consid-
erations, it is possible to compose the approximation for function G̃(k), which has
[1/T ] order that is in agreement with relaxation law (2.35). Monin obtained the
following expression for the kernel of the nonlocal functional describing turbulent
diffusion:
1/3
G̃(k) ∝ G̃(εK , k) = εK k 2/3 . (2.63)
The expression used by Monin has a dimensionality that is inversely proportional to
time. This fact reflects the essential difference of such a model from the Batchelor
approach [59]. This representation is consistent with the results derived in 1926 by
Richardson [58] under the assumption

G̃(k) = −D(k)k 2 , (2.64)

68 O.G. Bakunin

l2
where D(k) ≈ t ∝ l 4/3 ∝ k −4/3 . Also, in modern terminology [20–22], the equa-
tion
∂ ñk (t) 1 2
= −εK3 k 3 ñk (t) (2.65)
∂t
is the one with the fractional derivative with respect to x,

∂ αL n n
α
∝ ∝ k αL n, (2.66)
∂x L (x)αL
where αL = 2/3 [see formula (2.59)]. Monin was the first to obtain this equation
for the probability density on the basis of physical considerations. He solved this
equation and wrote the solution in terms of the Whittaker functions. The solution
behaves asymptotically as n(x → ∞) ∝ x −11/13 . The problem of the relaxation in a
self-similar regime was discussed in detail in [27, 62].
However, Monin was unsatisfied with the above form of the equation. In fact, he
derived the following equation with fractional derivatives:

∂n 2/3 n
1/3 ∂
= εK . (2.67)
∂t ∂x 2/3
It is only recently that the idea of using fractional derivatives has come to be recog-
nized [18–22]. In an effort to derive an equation that would be as clear as the tele-
graph equation, Monin differentiated his equation twice with respect to time and
obtained the expression
∂ 3n ∂ 2n
= εK . (2.68)
∂t 3 ∂x 2
Note that Monin suggested his equation to describe the diffusing particle probabil-
ity density evolution n. However his idea can be used to describe the probability
density evolution F , which describes relative diffusion. Such a version was con-
sidered in [103, 104]. But in those papers [103, 104] use was made of the modern
terminology and the fractional differential is represented as nonlocal integral opera-
tor √ 
 t)
∂F (l, 3  t) 3 
F (l,
= (2/3) 2 L d l . (2.69)
∂t 4π |l − l |5/3
Here,  is the Gamma function and L is the Laplace operator.
It is natural that the use of nonlocal operator (2.69) leads to the distribution func-
tion, which differs significantly from the Richardson and Batchelor models. Never-
theless, convincing arguments in favor of the choice of the specific type of equation
describing the behavior of the distribution function are absent and the search for ad-
equate models and experimental proofs has been continued. Note that in spite of the
assumptions of isotropy and the relative simplicity of experiments, these problems
remain acute. From this standpoint, the absence of a detailed pattern of anomalous
transport in high-temperature plasma does not look so catastrophic.
 2 Correlations and Anomalous Transport Models 69

2.4 The Corrsin Conjecture

The classical correlation definition of Taylor’s coefficient of turbulent diffusion does
not contain any information on molecular diffusion. It is obvious that a serious prob-
lem arises when we analyze the passive tracer transport. In this section we will con-
sider several important models in which the effects of molecular (seed) diffusion and
correlation effects play a significant role. The scaling arguments are presented.

2.4.1 The Corrsin Independence Hypothesis

The definition of the correlation function suggested by Taylor [31] is based on us-
ing Lagrangian velocities V (x0 , y), but their experimental determination is a serious
problem. That is why use is made of the Eulerian representation for the correlation
function, which takes into account the velocity correlation at points separated by a
distance λ:  
CE (λ, t) = u(x0 , T )u(x0 + λ, T + t) . (2.70)
This form of the correlation function is more convenient for experimenters. We can
also express the Lagrangian correlation function through the Eulerian velocity,
  
C(t) = u(x0 ; T )u x(x0 , T + t); T + t . (2.71)

Here, U (x0 , T ) is the Eulerian velocity at point x0 and time T . However, there is no
simple relation between the Lagrangian correlation function and the Eulerian one.
Actually, there is no Lagrangian relation between the points x0 and x0 + λ in expres-
sion (2.70). Here, λ is merely some arbitrary displacement.
Corrsin [28] suggested an approximation formula in terms of the randomization
of the Lagrange correlation function with the probability density ρ(x, t),
 ∞
C(t) = ρ(λ, t)CE (λ, t) dλ, (2.72)
−∞

in which he expressed the Lagrangian correlation function through the Eulerian one
[25–30]. However, a more important point is the idea of the diffusion nature of the
displacement λ, because for ρ(λ, t) Corrsin used the classical solution of the diffu-
sion equation in a three-dimensional space,
 
1 λ2
ρ(λ, t) = exp − . (2.73)
(4πD0 t)3/2 4D0 t

This formula also includes the molecular diffusion coefficient D0 . Hence, one can
consider both the turbulent transport and the molecular diffusion. Finally, Corrsin
obtained the integral expression
 ∞  
CE (λ, t) λ2
C(t) = 3/2
exp − dλ. (2.74)
−∞ (4πD0 t) 4D0 t
70 O.G. Bakunin

From this point of view, one can note that λ is the distance and the diffusive displace-
ment at the same time. In fact, instead of formal averaging in the form
 ∞
      
V x(0) V x(t) = V (0)V (y)δ y − x(t) dy, (2.75)
−∞

the factorization approach was used (the “independence hypothesis”):

      
V (0)V (y)δ y − x(t) = V (0)V (x) δ y − x(t) . (2.76)

Moreover, Corrsin used Gaussian distribution (2.73) to describe trajectory correla-

tions:   
δ y − x(t) ≈ ρ(y, t). (2.77)
Using rigorous analysis, Weinstock [105] and Kraichnan [106] showed that the
Corrsin conjecture is equivalent to a first-order truncation of the renormalization
expansion, which can be considered systematically.
The Corrsin conjecture has been tested against kinetic simulations of two- and
three-dimensional flows with an energy spectrum sharply peaked about one well-
determined length scale [105, 107, 108] with the conclusion that it is valid for all
times (not only for large times) provided that there is no helicity and that the flow is
not frozen in time.

2.4.2 The Simplified Corrsin Conjecture

The interesting paper by Hay and Pasquill [109] was written almost simultaneously
with the Corrsin paper [28]. They taken into account that Eulerian and Lagrangian
correlation functions have similar shapes; but at the same time, their characteristic
scales are different. Thus, the characteristic temporal scale, which corresponds to the
Lagrangian correlation function, is defined by the expression
 ∞
1
τL = CL (t) dt. (2.78)
V 2  0
The characteristic Eulerian temporal and spatial scales are defined analogously
 ∞
1
τE = CE (Δ, t) dt, (2.79)
U 2  0
 ∞
1
lE = CE (Δ, t) dΔ. (2.80)
U 2  0
The authors of [109] supposed that there exists a certain universal constant βC that
allows us to relate Lagrangian and Eulerian scales,

τL = βC τE , (2.81)

lL (βC t) = βC lE (t). (2.82)

 2 Correlations and Anomalous Transport Models 71

The variety of turbulence types leads to the fact that the values of βC defined exper-
imentally lie in a wide interval:

1 ≤ βC < 8.5. (2.83)

In spite of the obvious simplicity of the approach suggested by Hay and Pasquill, in
recent papers it was shown that in the framework of the consideration of one-particle
vertical diffusion in strongly stratified turbulence, the Eulerian and Lagrangian ve-
locity correlation functions are almost the same:
   
Vi (0)Vj (t) = Ui (x, 0)Uj (x, t) . (2.84)

Thus, Kaneda and Ishida [110] considered the Fourier transformation of the Corrsin
conjecture (2.76) in the form

     )−x(t)) 
Vi (t)Vj (t  ) = d 3 k R̃ij (k,
 t, t  ) e−i k(x(t , (2.85)

where 
1  r 3 
 t, t  ) =
R̃ij (k, U i (
x + 
r , t)Uj (
x , t  −i k
)e d r . (2.86)
(2π)3
From the physical point of view the Eulerian velocity correlations must be dominated
by large eddies (see [107]). This corresponds to small values of k.  Therefore, for

k ≈ 0 one can expect that
 −i k(x(t
  )−x(t)) 
e ≈ 1. (2.87)
This simple estimate gives the simplified Corrsin conjecture
   
Vi (t)Vj (t  ) = Ui ( x, t ) .
x , t)Uj ( (2.88)

This new representation

CL (τ ) ≈ CE (λ, τ )|λ=0 (2.89)
was first discussed by the author of [107]. In the framework of the Boussinesq ap-
proximation of strongly stratified flows, the validity of the simplified Corrsin con-
jecture (2.88) was checked by direct numerical simulations [110, 112]. It was shown
that the simulation results agree well with the hypothesis (2.88).

2.4.3 The Correlation Function and Scalings

The Corrsin conjecture looks fairly formal; however, it allows us to visualize corre-
lation effects and to take into account the effects of molecular diffusion. Note that
the definition of Taylor’s coefficient of turbulent diffusion does not contain any in-
formation on molecular diffusion. It is obvious that a serious problem arises when
analyzing the passive tracer transport [27, 28]. Moreover, the Corrsin representation
offers an additional possibility of developing the scaling approximation of transport
by the power approximations of the Eulerian correlation function and different kinds
of probability density [18–22].
72 O.G. Bakunin

The effective use of scaling laws will be considered here. Thus, the authors of
[113] made a rigorous analysis of equations for a random noncompressible flow
where the mean velocity is zero and the spatial correlation function decays as
1
CE (λ) ∝ . (2.90)
λαC
Koch and Brady used a continuum nonlocal advection–diffusion theory (the direct-
interaction approximation) and obtained an expression which connects the Hurst ex-
ponent H that describes the transport character with the correlation exponent αC :
2
H = . (2.91)
2 + αC
Here, αC describes the power behavior of the spatial correlation function of velocity
 αC
  λ0
C(λ) = V (x)V (x + λ) ∝ V02 . (2.92)
λ
Here, V0 and λ0 are the dimensional parameters of the model. Note that this re-
lationship can be obtained by simple calculations based on both the dimensional
consideration of the correlation function
λ2
C ≈ V2 ≈ (2.93)
t2
and the power dependence (2.92). Then, the comparison of (2.93) with (2.92) yields
 α
λ2 2 λ0
≈ V 0 . (2.94)
t2 λ
If we suppose that the spatial scale λ is the correlation scale and the diffusive dis-
placement at the same time (as in the Corrsin conjecture), then it is possible to treat
(2.94) as the transport scaling. Now, one can obtain the diffusive estimate:
 1 2
λ ∝ V02 λα0 C 2+αC t 2+αC (2.95)

and
2
H = , (2.96)
2 + αC
where 0 < αC < 2 since the result (2.95) was obtained for incompressible flow,
where the subdiffusive transport is absent [17, 114]. Later, relationship (2.96) was
repeatedly discussed in connection with the analysis of more complex models of
turbulent transport [17, 85, 86].

2.5 Effects of Seed Diffusivity

Corrsin was one of the first to understand the importance of accounting for seed
diffusivity effects to describe correlations. Further investigation of turbulent transport
 2 Correlations and Anomalous Transport Models 73

led the appearance of numerous estimates and scalings based on diffusive estimates.
Thus, in the framework of diffusive approximations it is possible to consider not only
transport of a passive tracer but also anomalous diffusion of particles in a braided
magnetic field.

2.5.1 Seed Diffusivity and Correlations

It is well known that interactions both create and destroy correlations. There is a use-
ful estimate which illustrates this in terms of the correlation function. It was assumed
[16, 68] that the number of interactions NI is proportional to the number of particles
that are located in the correlation region WD :

NI ≈ nWD ≈ nRD
d
. (2.97)

Here, n is the density of particles in this region, RD is the spatial scale of this region,
and d is the space dimensionality. Then, from the dimensional point of view, the
correlation effects can be expressed in the form

  V0 V2
C(t) = V (0)V (t) ≈ V0 ≈ 0 , (2.98)
NI nWD
where V (t) is the velocity at the moment t and V0 is the characteristic scale of the
velocity. The estimate of WD can be obtained from the conventional Gaussian distri-
bution  
1 x2
ρ(x, t) = exp − . (2.99)
(4πD0 t)d/2 4D0 t
Here, D0 is the molecular coefficient of diffusion. To derive the estimate it was sup-
posed that correlation scale RD has the diffusion nature

RD ∝ (D0 t)1/2 for t → ∞. (2.100)

This corresponds to the Corrsin assumptions. Simple calculations then yield

  V02 1
C(t) = V (0)V (t) ≈ ∝ d/2 . (2.101)
n(D0 t)d/2 t
In spite of the difference between this result and the exponential form (2.7), the ob-
tained power approximation of the correlation function is not senseless. The “long
tails” of correlation functions C(t) ∝ t −3/2 are being investigated in molecular dy-
namics and are related to “the collective” (hydrodynamic) nature of the evolution of
a system [16, 68]. The correlation function is related to diffusion coefficient (2.6),
which in our case leads to the estimate
d R2 1
C(t) ∝ DT ∝ 2 ∝ d/2 . (2.102)
dt t t
This yields the transport scaling, which differs significantly from the classical diffu-
sive one.
74 O.G. Bakunin

2.5.2 “Returns” and Correlations

In the previous sections we touched on some questions concerning the relationship

between the diffusion coefficient and correlations. Now, we analyze the effect under-
lying the notion of correlations—return of a randomly moving particle to the initial
point. This is best illustrated by considering the problem of one-dimensional ran-
dom walks at the very beginning of the process. In the problem as formulated, the
particle can definitely return to its initial position, thereby providing a clear realis-
tic interpretation of the abstract notion of correlations. Rigorous analysis of returns
on complicated spatial grids is necessarily based on the chain functional equation for
the return probability P0 (t) [18, 19]. Recall that most of the fundamental problems in
the theory of random-walk processes can be formulated in terms of chain functional
equations [18, 19]. However, we restrict ourselves here to considering the effects of
returns.
Simple estimates for these effects can be obtained from the classical solution to
the equation for the probability density function describing the random walks of a
particle. For a space of dimensionality d, one obtains the distribution
 
(δx)d x2
P (x, t) = ρ(x, t)(δx) =d
exp − . (2.103)
(4πDt)d/2 4Dt

Here, (δx)d is the small area around the point x. The probability of a particle return-
ing to point x = 0 at time t has the form

(δx)d
P0 (t) ∝ . (2.104)
(4πDt)d/2
Generally, this simple (although rather efficient) formula serves merely to obtain es-
timates [18, 19]. It has the same drawbacks as the simple diffusion model, namely,
those associated with the infinite propagation velocity of perturbations and the pres-
ence of a point-source term. However, for our purposes here, this solution is impor-
tant because it provides evidence that the dimensionality of the space, d, which was
used above as a formal parameter, plays a significant role. It turns out [18, 19] that,
for grids of dimensionality d ≤ 2, the particle will inevitably return to its initial
position. For grids with d > 2, the particle can execute random walks without re-
turning. We thus see that the case d = 2 is intermediate and, as such, attracts much
attention of mathematicians. Note that the correct dependence for P0 for d = 2 and
d = 3 is
1 1
P0 (N ) ∝ 2  d/2 . (2.105)
N N
Along with the return probability P0 , use is made of the number of returns and the
number of visited grid points. Usually, the task is to express these numbers as certain
scaling laws and to establish their relationships to other scaling laws.
Using approximation (2.104), we derive an important scaling relation for par-
ticles executing random motion with no self-intersections (self-avoiding random
 2 Correlations and Anomalous Transport Models 75

walk). We introduce the probability p(N) of self-intersection after N random walks,

N
p(N) ≈ , (2.106)
Rd
where R 2 (N ) is the root-mean-square displacement, d is the dimension of the space,
and N = t/τ is the number of random walks. Here, t is the time and τ is the cor-
relation time. In fact, we are assuming that the probability of the particle trajectory
intersecting itself is proportional to the number density of visited grid points within
the region of random particle motion. Then, the probability for a particle to execute
N self-avoiding random walks can be estimated as
 
N
N2
PS (N) ≈ (1 − p) N →∞ ≈ exp(−pN) ≈ exp − d . (2.107)
R

Taking into account the fact that the relationship between the quantities R and N is of
a diffusive nature, we can estimate the effective probability of self-avoiding random
walks by averaging the probability PS (N ) with the Gauss distribution:
 ∞      
1 t 2 1 R2
PS (t) = exp − d exp − (dR)d . (2.108)
−∞ R τ (4πDt)d/2 4Dt

We assume that the main contribution to the integral comes from the extremum of
the integrand,
  2 
1 t R2
min + , (2.109)
Rd τ 4Dt
and perform simple manipulations to obtain the scaling law:
3
R(t) ∝ t 3/(2+d)  t 1/2 , H = (2.110)
2+d
for d ≤ 3. Here, we must take into account the fact that, in a space of dimensionality
d = 1, nonself-intersecting random walks can occur only for the particles moving in
one direction, which indicates that R ∝ t. We see that estimate (2.110) satisfies this
condition automatically.
This scaling, which was first obtained in the theory of polymers by Flory [18], is
very important in describing the properties of different systems with nonconventional
correlation properties and complex topology.

2.5.3 The Stochastic Magnetic Field and Scalings

The description of the interaction of different scales in turbulent transport can also
be considered in another aspect. Thus, the analysis of transport in an anisotropic
medium leads to the necessity of considering the interplay of both the longitudi-
nal and transverse correlation mechanisms. This problem is particularly relevant in
connection with investigations of the processes of turbulent diffusion of plasma in a
76 O.G. Bakunin

Fig. 2.1. Force lines random walk

stochastic magnetic field. Many papers are devoted to this problem, but its complete
solution is still a long way off. In this section, we consider one of the first models
of particle transport in a “braided” magnetic field. This problem was formulated in
connection with the description of cosmic ray diffusion in a galactic magnetic field
[69, 70]. But this is a relevant model for some mechanisms of anomalous diffusion
in magnetized plasma [5–8]. The concept of random walks of magnetic force lines
in the transverse direction was the basis of this consideration (see Fig. 2.1). In this
case, it is convenient to introduce a magnetic diffusion coefficient of streamlines

Δ2⊥
Dm ∝ . (2.111)
L

Here, Δ⊥ is the displacement of the perturbed force line in the transverse direction
under the condition of displacement along the streamline on the length L . If we
assume that particles in their motion strictly follow the displacement of the force
line, which was initially chosen by them (particles are skewed on the streamline
similar to beads), then it is easy to obtain the expression for the coefficient of the
transverse diffusion of particles,

Δ2⊥ L L
D⊥ ∝ ≈ Dm . (2.112)
L t t

In the case of ballistic motion of particles along the streamline, the estimate of the
transverse diffusion coefficient has the form

D⊥ ≈ Dm V . (2.113)

Here, V is the particle velocity under the condition of motion along the force line.
 2 Correlations and Anomalous Transport Models 77

A nonstandard situation arises when we consider collisions between particles

which are located in the “braided” field under consideration. Then it is natural to
suppose that the motion in the longitudinal direction has a diffusive character (ran-
dom walks along the force line):

L2cor L2
D ≈ ≈ . (2.114)
2τ 2t
Here, Lcor is the particle longitudinal correlation length and τ is the correlation time.
Then the estimate of longitudinal displacement is the value

L ≈ 2D t. (2.115)

The substitution of (2.115) into (2.112) yields


2D t  Dm
D⊥ ≈ Dm ≈ 2D √ . (2.116)
t t
This result demonstrates an essential deviation of the transverse transport from the
classical diffusion transport, since
 √
Δ2⊥ ≈ Dm 2D t ∝ t 1/2  t. (2.117)

This corresponds to the subdiffusive character of transport with the Hurst exponent
H = 1/4. Note that in this model particles never leave the force line of the magnetic
field in which they were initially situated. The result obtained shows the nontrivial
character of the relation between longitudinal and transverse correlation effects in
the case of the description of transport in the stochastic magnetic field. Another im-
portant aspect of this problem is the necessity to consider the essentially nondiffusive
character of transport.
Note that double diffusion is one of the simplest models of anisotropic anomalous
transport in the stochastic magnetic field. This mechanism is destroyed under the
influence of time-dependent perturbations and the effect of stochastic instability of
nearby streamlines [3, 5, 17]. These effects will be considered in the next section of
this paper.

2.5.4 The Howells Result

The “seed diffusivity” concept is very effective and allows the consideration of fairly
complex correlation effects. The natural generalization of this approach is the ap-
plication of the “self-consistent” diffusion coefficient that describes both turbulent
transport and molecular diffusion effects. Here, we consider very briefly the How-
ells model [73] that will be repeatedly discussed in the context of scaling arguments.
A more detailed consideration of this problem can be found in the well-known Mof-
fat reviews [32, 33].
78 O.G. Bakunin

The Howells model of turbulent transport is based on the spectral energy function
E(k) that plays a significant role in the framework of the Kolmogorov theory of
isotropic turbulence [27, 29, 30]:
 ∞
V 2 
= E(k) dk. (2.118)
2 0

Here, V is the velocity scale and k is the wave number. From this point of view the
Kolmogorov–Obukhov law has the form

Vk2 1
E(k) ∝ ∝ 5/3 . (2.119)
k k
Howells [73] obtained an important relationship between the turbulent diffusion co-
efficient DT and the energy spectrum E(k). He considered a “local” diffusion coef-
ficient δD(k) related to the specific scale length lk ≈ 1/k of eddies with the charac-
teristic velocity Vk , δD(k) ≈ Vk2 τ0 , where Vk2 ≈ E(k)δk. Here, δk is a small interval
of wave numbers and τ0 is the characteristic correlation time. We will consider that
τ0 is related to the molecular diffusion Do , τ0 ≈ 1/k 2 D0 . We obtain the expression
that is differential in form:
δD(k) E(k)
= 2 . (2.120)
δk k D0
Note that the value of D(k) should be taken into account along with molecular diffu-
sion D0 . This is the simplest example of renormalization by “seed” diffusivity. One
then obtains
dD(k) E(k)
= 2 . (2.121)
dk k (D0 + D(k))
Upon solving this equation, we obtain the expression for the turbulent diffusion co-
efficient. This expression describes the influence of different scales,
 ∞
 2 E(k)
D(k) + D0 = 2
dk + D02 . (2.122)
k k
Here, it is assumed that D(∞) = 0. In this expression the integral term plays the
main role for scales that are larger than the characteristic turbulent scale lT , which
enters into the expression for the Reynolds number: Re = V0 lT /η. Here, V0 is the
characteristic velocity and η is the coefficient of viscosity. Therefore, neglecting the
molecular diffusion effects (on the right-hand side), the Howells expression can be
derived:  ∞
E(k)
DH =
2
dk. (2.123)
k k2
From the standpoint of dimensional estimates, the expression obtained differs signif-
icantly from the Taylor definition for DT (2.6). The classical dimensional estimate
of the diffusion coefficient is a formula closely associated with the model of random
walk, D0 ≈ Δ2 /τ . Here, Δ is the characteristic spatial correlation scale length and
τ is the characteristic correlation time. An analogous estimate in the Taylor formula
is DT ≈ V02 τ . The formula suggested by Howells yields a different (kinetic) type of
 2 Correlations and Anomalous Transport Models 79

estimates for the diffusion coefficient,

DH ≈ V0 Δ. (2.124)

It is possible to determine a relationship between these expressions. Let us consider

the Peclet number [17, 32, 33],
V0 Δ
Pe = . (2.125)
D0
This dimensionless quantity reminiscent of the Reynolds number has the same signif-
icance. The Peclet number allows the estimation of the fraction of convective trans-
port and its comparison with the diffusion one. In terms of the Peclet number we
obtain

D0 = D0 Pe0 ≡ D0 , DT = D0 Pe2 , DH = D0 Pe. (2.126)

The result presented in this form with different exponents ϑ is now in wide use
[17, 32, 33]:
Deff = D0 Peϑ . (2.127)
Here, Deff is the effective diffusion coefficient. It permits us to describe transport in
terms of the exponent ϑ, and such a representation reflects the nature of the depen-
dence on the velocity amplitude. Of course, there is no unique recipe to obtain the
estimates of turbulent transport. Thus there is an important example that was obtained
by Dykhne, Isichenko, and Horton [116] for anomalous transport enhancement due
to the hexagonal cells controlled by the boundary tubes:

Deff ≈ D0 ln Pe, where Pe  1. (2.128)

The detailed analysis and scaling estimates for this nontrivial model can be found in
[17, 116].

2.6 The Diffusive Tracer Equation and Averaging

In spite of the effectiveness of diffusive estimates to describe correlation effects and

transport, the equations describing the evolution of tracer distribution density play
an important role in the description of anomalous diffusion effects. In this section,
the heuristic Taylor method of obtaining such an equation is considered. The ap-
proximation suggested in [117] has become an important step in the development of
description methods of anomalous transport and complex correlation effects.

2.6.1 The Taylor Shear Flow Model

In this section, we consider the description of the effective diffusion of a passive

scalar in a shear flow in the presence of seed diffusivity. In his paper [117] Taylor
80 O.G. Bakunin

suggested a new method to obtain the effective diffusion coefficient, which is based
on averaging the convection–diffusion equation
∂n ∂n
+ V (y, z) = D0 ∇ 2 n. (2.129)
∂t ∂x
Here, n is the scalar density, V is the longitudinal (along the x-axis) velocity, and
D0 is the seed diffusivity. Actually, in the Taylor model the influence of transverse
molecular diffusion on longitudinal convective transport was considered.
The paper [117] deals with Poiseuille (parabolic) flow in a cylindrical tube, but
in order to simplify calculations we will analyze a flat model. Considering the profile
of the longitudinal flow in the form
V0  2
V (y) = L − y2 , (2.130)
L2
we can learn the scalar transport problem in the framework of the decomposition
method
n = n + n1 (x, y, t) = n0 + n1 (x, y, t), (2.131)
V = V  + V1 (y) ≡ V0 + V1 , (2.132)
where V0 is the characteristic velocity and L is the characteristic spatial scale. Here,
use is made of the expression for mean values
 L
1
n ≡ n(x, y, t) dy ≡ n0 (x, t), (2.133)
2L −L
 L
1 2
V  = V (y) dy = V0 . (2.134)
2L −L 3
Hence, one obtains

 2 
1 y
V1 = V0 − . (2.135)
3 L
The substitution of the expression for n and V into the initial equation (2.139) yields
∂ ∂ ∂
n0 + n1 + (V0 + V1 ) (n0 + n1 ) = D0 ∇ 2 [n0 + n1 ]. (2.136)
∂t ∂t ∂x
Taking the average of this equation, it is easy to obtain an expression for the mean
density evolution,
 
∂ ∂ ∂ ∂ 2 n0
n0 + V0 n0 + V1 n1 = D0 2 . (2.137)
∂t ∂x ∂x ∂x

Actually, in order to obtain a closed-equation for the mean density of the scalar, it is
necessary to find an expression for n1 (x, y, t). Subtracting (2.137) from (2.136), one
 2 Correlations and Anomalous Transport Models 81

obtains the equation for the evolution of density perturbation n1 :

   2 
∂n1 ∂n0 ∂n1 ∂n1 ∂n1 ∂ n1 ∂ 2 n1
+ V1 + V0 + V1 − V1 =D + . (2.138)
∂t ∂x ∂x ∂x ∂x ∂x 2 ∂y 2
The expression obtained is too complex; therefore, Taylor suggested the heuristic
method to obtain the estimates of effective diffusion, which is based on some hy-
potheses:
∂n1
• quasi-steadiness of n1 , i.e., ∂t ≈ 0;
∂n1 ∂ 2 n1 ∂n0 ∂ 2 n1
• smallness of ∂x and ∂x 2
in comparison with ∂x and ∂y 2
.

Taylor kept the term ∂ 2 n1 /∂y 2 , basing this on the necessity of taking into account
the density gradient in the direction of the walls, which has to be greater to satisfy
the no flux condition ∂n1 /∂y = 0 at y = L and y = −L. Then, solving the equation
obtained from (2.138),
∂ 2 n1
D0 2 = Q(x, y, t), (2.139)
∂y
where Q = V1 (y)∂n0 (x, t)/∂x, it is easy to find


∂n0 V0 y 2 y4
n1 = − + const(x, t). (2.140)
∂x 3D0 2 4L2
Applying the condition n1  = 0 we arrive at const = (∂n0 /∂x) × (V0 /3D0 ) ×
(−7L2 /60),

∂n1 ∂ 2 n0 V0 y 2 y4 7 2
= − − L . (2.141)
∂x ∂x 2 3D0 2 4L2 60
The expression V1 ∂n1 /∂x, which defines an additional contribution in longitudinal
diffusive transport, can be rewritten in the form
 
∂n1 ∂ 2 n0 8 (V0 L)2 ∂ 2 n0
V1 = −D1 2 = − . (2.142)
∂x ∂x 945 D0 ∂x 2
The result obtained is not trivial, because the additional diffusive contribution de-
pends inversely on seed diffusivity D0 . The physical interpretation of this result is
the limitation of the influence of nonuniformity of the longitudinal velocity profile
V (y) by transverse diffusion. Hence, nonuniform longitudinal convection in combi-
nation with transverse diffusion leads to longitudinal diffusion. Naturally, the new
diffusive mechanism manifests itself at a large distance downstream only, since the
equation obtained is correct only for t  τD ≈ L2 /D. On the other hand, use was
made of the condition of the smallness of the transverse spatial scale in comparison
with the longitudinal one l: L  l. The effective diffusion coefficient obtained can
be rewritten in terms of the Peclet number
   
8 V02 L2 8
Deff = D0 + = D0 + D0 Pe2 . (2.143)
945 D0 945
Note that the author of [117, 118] did not use any restriction on the Peclet number.
82 O.G. Bakunin

2.6.2 Generalization of the Taylor Model

The flow model considered in the previous section is fairly simple, however, the
method suggested by Taylor allows one to apply an analogous approach for the de-
scription of different shear flows [118, 119]. Thus, from the formal standpoint, the
equation describing density perturbation

∂ 2 n1 ∂n0
D0 =V (2.144)
∂y 2 ∂x
can easily be supplemented with the terms omitted before in order to describe more
complex situations. Thus, including the term describing the time dependent character
of density perturbations ∂n1 /∂t or using the Laplace operator, which characterizes
diffusion on the plane of a transverse cross-section of a canal in the form
 2 
∂ n1 ∂ 2 n1
D0 + (2.145)
∂y 2 ∂z2

keeps the advantages of the algorithm of the solution suggested by Taylor; since the
equation for density perturbation n1 has, as before, a linear form,
∂n1 ∂n0
= D0 ∇ 2 n1 − V1 . (2.146)
∂t ∂x
It is necessary to note that there is also some arbitrariness in the choice of conditions
for solving the partial differential equation. The solutions founded in the framework
of a more precise formulation of the problem allow us to see that the nontrivial de-
pendence (2.143) obtained by Taylor of the effective diffusion Deff on the flow para-
meters D0 and L is correct. The nontrivial character of effective diffusivity behavior
in the Taylor model impels us to use the heuristic methods of including nonlocal
effects and memory effects into the initial local equation. In the framework of the
heuristic approach, it is easy to include memory effects in the equation under analy-
sis (2.129). This allows us to analyze trapping effects, which play an important role
in tracer transport. Thus, one can represent the total concentration of tracer n(x, t)
as two parts,
n(x, t) = p(x, t) + q(t), (2.147)
where p(x, t) corresponds to actively transporting particles and q(x, t) describes
trapping. In the simplest case the relationship between p and q is given by
∂q
= αT (βT p − q). (2.148)
∂t
Here, αT and βT are the parameters of the problem. If all the particles are released
in the untrapped region at t = 0, one obtains
 t
q(x, t) = αT βT p(x, τ )e−αT (t−τ ) dτ . (2.149)
0
 2 Correlations and Anomalous Transport Models 83

Here, αT βT exp[−αT (t − τ )] is the memory function. In general, we can rewrite the

expression in the form with an arbitrary memory function M:
 t
q(x, t) = p(x, t)M(t − τ ) dτ . (2.150)
0

Then, based on tracer conservation, it is possible to describe transport by

∂p ∂q ∂p ∂ 2p
+ +V = D 2. (2.151)
∂t ∂t ∂x ∂x
Using the Taylor method, the modified equation for the mean density can be rewritten
in the form  t
∂n0 ∂ 2 n0 (x, τ )
= M(t − τ ) dτ . (2.152)
∂t 0 ∂x 2
The expression for the effective diffusion coefficient, which takes into account mem-
ory effects, is then given by
 ∞
Deff = M(τ ) dτ. (2.153)
0

Here, we consider the behavior over a long time period.

Another aspect of the problem considered in [117] is the possibility of analyzing
more complex profiles of shear flow V . Thus, in the next parts we will consider the
model profiles Vx = Vx (y, t), which allow us to understand time periodic velocity
fields and tracer diffusivity in the turbulent Poiseuille flow [35]. On the other hand,
the use of randomly distributed stream functions Ψ (z) makes it possible to describe
the system of random shear flows [72] Vx = −[∇z × Ψ (z)] based on their correlation
properties.

2.6.3 The Zeldovich Flow and the Kubo Number

The time-dependence of a flow is an important factor that has an influence on trans-
port processes. This leads to reconstruction of the streamline topology. To describe
this situation we need a new dimensionless parameter, which includes the character-
istic time T0 ≈ 1/ω. In the case of high frequencies ω, the path of a test particle can
be estimated using the ballistic method as lω ≈ V0 /ω. Then the convective fraction
of transport can be characterized by the dimensionless Kubo number
lω V0
Ku = ≈ , (2.154)
λ ωλ
where λ is the spatial scale of the flow under consideration. In the conventional sit-
uation (the quasi-linear limit), we obtain the estimate of the diffusivity in the form
DT ≈ V02 τc ≈ V02 /ω. However, in the case of high frequencies it is necessary to use
the path lω = V0 /ω as the correlation length:

lω2 V2 1
D≈ ≈ 02 . (2.155)
τc ω τc
84 O.G. Bakunin

For τc ≈ 1/ω one then obtains the quasi-linear estimate. There is also another pos-
sible method. We can relate the correlation time to the diffusive mechanism of the
particle escape from streamlines 1/τc ≈ D0 /λ2 ≈ 1/τD , where D0 is the seed dif-
fusion coefficient. Then the new estimate of the effective diffusion coefficient takes
the form
D0 V 2
D ≈ 2 20 ≈ D0 Ku2 . (2.156)
λ ω
These simple estimates can be validated by considering Zeldovich’s model of two-
dimensional flow. Zeldovich considered [35] turbulent diffusion in a system of regu-
lar but time-dependent flows (compare with the Taylor model). The solved equation
has the form
∂n ∂n
+ VX (z, t) = D0 n. (2.157)
∂t ∂x
The expression for the velocity of flows is given by

VX (z, t) = 2V0 cos(kz) cos(ωt). (2.158)

Zeldovich suggested finding the solution as follows:

n(z, t) = n0 + n1 = n0 + sin(kz)(nS sin ωt + nC cos ωt + · · ·). (2.159)

The amplitudes of the harmonics nS , nC can be defined as a result of the solution of

(2.157). It was suggested that

n0 = n = a + bx. (2.160)

The values V0 , a, b are the flow characteristics. Substitution of (2.159) into (2.157)
with allowance for the assumption n1  n0 yields the equation for the average
density n0 ,  
∂n0 ∂n1 ∂ 2 n0
= − VX + D0 2 . (2.161)
∂t ∂x ∂x
The equations for the amplitudes of the harmonics nS , nC have the form

2V0 ∂n0 D0 k 2 D0 k 2
nS + =− nC , nC = nS . (2.162)
w ∂x ω ω
Simple calculations allow us to obtain

 2
∂n0 V02 k 2 ∂ n0 ∂ 2 n0
= D0 + D 0 . (2.163)
∂t ω2 + D02 k 4 ∂x 2 ∂x 2

In terms of the Kubo number we get the expression for the effective diffusion coeffi-
cient,

Ku2
Deff (k, ω) = D0 + 1 , (2.164)
1 + (ωτD )−1
 2 Correlations and Anomalous Transport Models 85

where τD = 1/(D0 k 2 ). For high frequencies ω > 1/τD we arrive at the formula

2 2
V k
D ≈ D0 0 2 ≈ D0 Ku2 . (2.165)
ω

In the case of linear dispersion ω2 ≈ βV02 k 2 , Zeldovich obtained the expression for
β → 0: D ≈ V02 /D0 k 2 . The effective coefficient of diffusion takes the form of
the Howells diffusion coefficient [72] DD0 ≈ Deff 2 ≈ V 2 /k 2 . However, in contrast
0
to Howells’ paper, where an isotropic model was considered, the Zeldovich case is
essentially anisotropic.

2.6.4 Advection and Zeldovich Scaling

Based on the Taylor renormalization method, it is appropriate to raise a question

about the estimation of effective transport in a turbulent flow. Zeldovich [35, 36]
suggested the phenomenological method to estimate scalar transport in a turbulent
flow in the presence of seed diffusion based on
∂n
= ∇(D0 ∇n) − V (t)∇n. (2.166)
∂t
Here, V (t) is the velocity field. The terms describing effects that are important for
us here enter into the left-hand side of this equation; therefore, multiplying the equa-
tion by n and using the Gaussian theorem, we can omit the term describing density
evolution,
  
1 ∂
0= n2 dW = nD0 (∇n)N dS − D0 (∇n)2 dW , (2.167)
2 ∂t W S W

since in the framework of quasi-steady turbulence it is natural to omit the term on the
left-hand side of (2.167). The expression for D0 (∇n)N characterizes the contribution
of external sources inside the volume W , which is bounded by the surface S, whereas
the term D0 (∇n)2 is related to the scalar redistribution inside the considered volume
W . It is convenient to introduce here the effective diffusive coefficient in the form

1
Deff = 2 D0 (∇n)2 dW, (2.168)
n L0 W
where L0 is the system characteristic size. Then the minimum condition for Deff (the
condition of minimizing functional) comes to a purely diffusive equation

∇(D0 ∇n) = 0, where min Deff = D0 . (2.169)

Applying a method analogous to the above-considered Taylor approach, it is easy

to obtain the upper estimate of the effective diffusion coefficient in a quasi-steady
turbulent flow. Consider the steady scalar density equation

D0 n − V ∇n = 0 (2.170)
86 O.G. Bakunin

using the perturbation method:

n = n + n1 = n0 + n1 , (2.171)
V = V  + v1 = v1 , (2.172)
where V  = 0, n1  n0 , and D0 n0 = 0. Simple calculations lead to the equation
for density perturbation n1 which actually coincides with the Taylor “renormaliza-
tion” formulated now for a turbulent velocity field:
∂ 2 n1 ∂n0
D0 2
= v1 . (2.173)
∂x ∂x
For the sake of simplicity, this equation is presented in the one-dimensional form. In
the framework of the dimensional estimate, we obtain
n1 ≈ ν1 Ln0 /D0 ≈ Pen0 , (2.174)
where Pe  1, which corresponds to weak turbulence regimes. In obtaining (2.174)
we use the condition of smallness of the term v1 ∇n1 in comparison with ν1 ∇n0 . The
definition of the effective diffusion coefficient in terms of (∇n)2 yields

1  
Deff ≈ 2 D0 (∇n0 )2 1 + A · Pe2 dW ≈ D0 1 + A · Pe2 , (2.175)
n0 L W
where A is the dimensionless constant. Note that the term ∇n0 ∇n1 is illuminated
because of the extremal properties of the distribution n0 . This upper estimate of
transport Deff in the steady turbulent flow coincides with the quasilinear scaling
Deff ≈ V02 τ when the correlation time τ has the diffusive meaning

τ ≈ τD ≈ L20 /D0 . (2.176)

At the same time, there is certainly a strong relation between the Taylor approach
to transport in a laminar shear flow and the Zeldovich general estimate of turbu-
lent diffusivity [40]. In the next sections we consider how different modifications
of the diffusive term in the equation for density perturbations n1 make it possible to
move from the quasi-linear description of transport to nonlocal models with complex
topology.

2.7 The System of Random Shear Flows

The problem of calculating the turbulent diffusion coefficient is closely related to the
character of the correlation function. In “common language”, the correlation means
a relation between events. The return of a walking particle to the initial point [18, 19]
is a good example of this relationship. Corrsin [34] was probably the first to suggest
using the return probability to describe turbulent diffusion. He formulated several
other probabilistic problems in the framework of turbulent diffusion in [34]. In this
section, we consider anomalous transport in the system of random shear flows where
the nontrivial character of diffusivity in anisotropic systems is manifested.
 2 Correlations and Anomalous Transport Models 87

2.7.1 The Dreizin–Dykhne Superdiffusion Regime

A physically transparent model was suggested and considered in Dreizin and
Dykhne’s paper [72] related to the incorporation of the influence of returns. The ac-
tion of “seed” diffusion in the longitudinal direction (which relates to the magnetic
field direction) with the diffusion coefficient D0 was considered. Random fluctua-
tions creating narrow convective flows of width a and velocity V0 act in the transverse
direction on the diffusing test particle (see Fig. 2.2). A simple model was proposed
[72] for the calculation of the transverse diffusion coefficient D⊥ :
λ2⊥ δN
D⊥ ≈ , where λ⊥ ≈ V0 tP∞ , P∞ = . (2.177)
t N
Here λ⊥ is the transverse displacement during time t and P∞ is the relative √number of
the small fraction of “noncompensated” fluctuations δN. The value N ≈ 2D0 t/a is
the number of shear flows intersected by √ the particle during its longitudinal motion.
Using the Gauss representation δN ≈ N, the following formula was obtained
in [72]: 
t
D⊥ ∝ V0 a2
. (2.178)
D0
In the superdiffusive case under consideration, it was found that H = 3/4 > 1/2.
To explain this result, the correlation function in the following form was consid-
ered in [72]:  ∞
 
C(t1 , t2 ) = Vx (0)Vx (z) ρ(z, t2 − t1 ) dz. (2.179)
−∞
The probability density has the Gaussian form:
 
1 z2
ρ= exp − . (2.180)
(4πD0 (t2 − t1 ))1/2 4D0 (t2 − t1 )

Fig. 2.2. Shear flows model

88 O.G. Bakunin

Here, Vx (z) is the velocity of the flow at the point z. This approach corresponds
exactly to the Corrsin idea about the diffusive nature of decorrelation [28] with al-
lowance for the anisotropy of the model. However, using the conjecture about the
significant role of returns has become the main step in the description of anomalous
diffusion, since the condition z → 0 for (2.179) corresponds to the return to the
initial point. Thus, we have the expression

V 2a
C(t1 , t2 ) = C(τ ) ≈ √ 0 , τ = t2 − t1 . (2.181)
4πD0 τ
Using the classical expression from the turbulent diffusion theory (2.6), estimate
(2.181) was obtained as
 t t
 2 V02 a dt1 dt2 V 2 a 3/2
λ⊥ ≈ √ √ ≈√ 0 t . (2.182)
4πD0 0 0 t1 − t2 4πD0
Here   denotes the averaging symbol.
It is important that representation (2.182) be valid only if the perpendicular spa-
tial displacement is no larger than the perpendicular correlation length Δ⊥ ≈ a:
λ⊥ < a. This restriction has the form
 
V 2 a 3/2 D0 a 2 1/3
λ2⊥ ≈ √0 τC ≤ a 2 , τC ≈ . (2.183)
D0 V04

If time scales t > τC , then we are dealing with classical diffusion, D⊥ ∝ a 2 /2τC .
We see that Dreizin and Dykhne suggested a fairly simple and, at the same time,
nontrivial model of anisotropic transport. Later, the analogous problem was consid-
ered by Matheron and De Marsily [120] in the context of transport effects in porous
media. Avellaneda and Majda [41, 42] treated this model in detail in the framework
of the renormalization theory. Analyses carried out in [41, 42, 72, 120] show that if
the value V  is not strictly parallel to the “layers” and if the sample is finite in the
longitudinal directions then the superdiffusion regime (2.182) is destroyed. Never-
theless, the anomalous regime exists for intermediate times [see (2.183)].

2.7.2 The Matheron–de Marsily Model

In spite of the detailed physical analysis of the shear flows model, which was carried
out by Dreizin and Dykhne, this result became well known after the paper [120] by
Matheron and de Marsily in connection with learning transport in a porous media.
Their analysis is based on the consideration of stochastic equations of motion in
longitudinal and transverse directions:
dx 
= VX z(t) , (2.184)
dt
dz
= η(t). (2.185)
dt
 2 Correlations and Anomalous Transport Models 89

Here, Vx  and the shear velocity distribution are taken to be white noise:
 
VX (z)VX (z ) = σV δ(z − z ). (2.186)

At the same time, the motion along the z-axis is the conventional Brownian motion:
 
η(t)η(t  ) = 2D0 δ(t − t  ). (2.187)

Note that in the framework of the consideration of the Dreisin–Dykhne superdiffu-

sive regime, the Brownian motion in the transverse direction can be neglected. Then,
longitudinal and transverse displacements can be rewritten in the form
 t
 
z(t) = η(t  ) dt  or z2 = 2D0 t, (2.188)

 t 
x(t) = VX z(t  ) dt  . (2.189)

To obtain the scaling describing transport in the transverse direction, let us introduce
a distribution function J (z, t) of the number of visits that were made by a test particle
into different “jets” by the moment t,
 t

J (z, t) = δ z − z(t  ) dt. (2.190)
0

The formal expressions for the displacement x(t) can then be represented in the form
 ∞
x(t) = Vx (z)J (z, t) dz. (2.191)
−∞

Since in the transport analysis the mean values play the main role, consider the sim-
plest Gaussian approximation of the mean probability density J ,
 t  
  |z| 1 z2
J (z, t) = ρ(z, t ) dt = √  − ; , (2.192)
0 4 πD0 2 4D0 t

where (− 12 ; x) is the incomplete gamma function. Then the mean displacement is
given by the expression
 ∞  
  1 1 z2
x(t) = √ dz |z| − ; Vx (z). (2.193)
4 πD0 −∞ 2 4D0 t

Using the averaging over environments we obtain the expression for the mean-
squared displacement that takes into account the statistical properties of the field
Vx (z),
 
 2    
x(t) = dz dz VX (z)VX (z ) J (z, t)J (z , t) . (2.194)
90 O.G. Bakunin

Applying the statistical properties of the velocity field V (z)V (z ) = σV δ(z − z ) it
is easy to find a final solution:
  ∞  2
 2  σV t 3/2 1
x (t) = σV dz J (z, t)2 = √ s 2  − ; s 2 ds
D0 π 0 2
4 √ σV 3/2
= √ ( 2 − 1) √ t . (2.195)
3 π D0
The averaging method that was used here is not traditional, however, the scaling
obtained is correct, R 2 ∝ t 3/2 .
The authors of [120] repeatedly pointed out the nontrivial character of the differ-
ent possible ensemble averages. Thus, the alternative possibility is the definition of
x m (t) by the expression
 t  t1  tm−1
 m     
x (t) = m! dt1 dt2 dtm VX z(t1 ) · · · VX z(tm ) , (2.196)
0 0 0

where
   
VX z(t1 ) · · · VX z(tm )
 ∞
 
= dz1 · · · dzm−1 dzm VX (z1 ) · · · VX (zm−1 )VX (zm )
−∞
× ρ(zm , tm )ρ(zm−1 − zm , tm−1 − tm ) · · · ρ(z1 − z2 , t1 − t2 ). (2.197)

To calculate x m  it is convenient to use the Laplace transformation in time:


 m  m! ∞  
x (s) = dz1 · · · dzm VX (z1 ) · · · VX (zm )
s −∞
× ρ(zm , s)ρ(zm−1 − zm , s) · · · ρ(z1 − z2 , s). (2.198)

Using the Gaussian representation for ρ(z, t) allows us to carry out the necessary
calculations. Thus, for the mean-squared displacement one obtains [120]
 
 2  2σV ∞ ∞
x (s) = dz1 dz2 δ(z1 − z2 )ρ(z2 , s)ρ(z1 − z2 , s)
s −∞ −∞
 ∞  
2σV σV 1
= ρ(0, s) dz ρ(z, s) = √ (2.199)
s −∞ D0 s 5/2
which by the Laplace inversion yields
  4σV 3/2
x 2 (t) = √ t . (2.200)
3 πD0
Actually, in such an approach we consider the averages, which are first taken over
the walks · · ·W and then over the shear flow configurations · · ·C , i.e., · · · ≡
· · ·W C .
 2 Correlations and Anomalous Transport Models 91

2.7.3 The “Manhattan Grid” Flow and Transport

The Dreizin–Dykhne renormalization (6.1) shows that even a small fraction of non-
compensated flows
δN 
P∞ (t) = ∝ 1/ N (t) (2.201)
N
leads to a considerable deviation in transport from the diffusive behavior. Moreover,
several other important models of anomalous and percolation transport can be de-
scribed in the framework of a similar approach. Thus, the formula for correlation
function (2.181) includes the number of returns NB or the number of intersected
flows NI for the particle walking along a straight-line trajectory for time t as
√
2D0 t
N (t) ∝ . (2.202)
a
Indeed, the number of returns NB in the interval δz can be estimated after we assume
that the return probability ρ(0, t)δz ≈ ρ(0, t)a. Thus, we obtain the estimate
√
(t) t D0 t
NB ∝ ρ(0, t)a ∝ . (2.203)
τ a

Here, τ ∝ D0 /a 2 is the longitudinal correlation time. In the case of one-dimensional

longitudinal diffusion (d = 1), which we are considering, we have NB (t) ∝ NI (t).
There exists an opportunity to generalize the Dreizin–Dykhne result for the case of
a more complex “topology” of flows with the value d > 1 as follows:

V02 V02
C(t) ∝ or C(t) ∝ . (2.204)
NI (t) NB (t)
A similar formula is used in analyzing correlation effects in statistical physics [16]
as the simplest correlation estimate (2.98).
In the isotropic case, for a system of random flows, it is possible to use the
Alexander–Orbach conjecture for the number of visited sites, NI (t) ∝ t 2/3 , for
2 ≤ d ≤ 6 [17, 18]. Hence, we obtain the following scaling law for the displacement
R of the particle:

R2
∝ C(t) dt, R ∝ t 2/3 . (2.205)
t
Here, the Hurst exponent is denoted by H = 2/3. Indeed, Redner [122] and Bou-
chaud et al. [123] obtained such a superdiffusion regime for the “Manhattan-grid”
flow, which is a generalization of the shear flows model [72, 120, 124]. Thus, from
the formal standpoint a shear flow system is described by the velocity field
  
V = − ∇z × Ψ (z) = u(z), 0 , (2.206)

where u(z) is a random function. A flow with the stream function Ψ = Ψ (x) + Ψ (z)
is then a two-dimensional generalization of the shear flows model (see Fig. 2.3). The
authors of [22] assumed that in the case under consideration transport is described
92 O.G. Bakunin

Fig. 2.3. “Manhattan-grid” flow

by the Hurst scaling R ≈ t H . On the other hand, scaling arguments yield

R ≈ V t. (2.207)
The expression for V  for the case of Gaussian statistics has the form

1 
N
V0 δN V0
V  = Vi = ≈ 1/2 . (2.208)
N N N
i=1

Here, the value N corresponds to the number of “layers” intersected by the test par-
ticle. The authors of [122] suggested the simplest approximation (analogous to the
one-dimensional case) N ∝ R ∝ t H , which leads to the scaling
R ≈ V0 t 1−H /2 . (2.209)

A comparison between the last expressions yields the Hurst exponent H = 2/3.
Note that expression (2.209) implies that the auto-correlation function of veloc-
ity has the power tail C(t) ≈ 1/t H , which is in accordance with estimate (2.204).
Redner [122] supposed a “hyper scaling” for models with the dimensionality d ≤ 3,
2
H = , (2.210)
d +1
which implies the accuracy of the result obtained for both the one-dimensional
(d = 1) and the isotropic three-dimensional (d = 3) cases. The “Manhattan-grid”
flow could be a simple model for two-dimensional turbulent transport. Thus, the
analysis of the experimental data of turbulent transport in a tokamak (where turbu-
lence can be considered two-dimensional due to the presence of a strong magnetic
field) yields the following estimate of the Hurst exponent: 0.6 < H < 0.75, which
approximately corresponds to H = 2/3.
 2 Correlations and Anomalous Transport Models 93

2.8 The Quasi-Linear Approximation

The quasi-linear approximation has become widely popular due to its exceptional
efficiency and close relation to hydrodynamic models. However, an analysis shows
that quasi-linear equations are also closely related to correlation concepts. Thus, the
dimensional estimate of the Taylor coefficient of turbulent diffusion DT is usually
called the quasi-linear expression for the diffusivity. In this section, we consider a
derivation of the quasi-linear equation and a simple quasi-linear model of the sto-
chastic magnetic field diffusivity.

2.8.1 Quasi-Linear Equations

From the formal standpoint, the construction of perturbation theory based on the
continuity equation is more correct than the consideration of equations, where seed
diffusion is incorporated. However, historically the paper by Taylor [117], which
shows the nontrivial character of calculations of mean magnitudes, has played an
important role, since the heuristic arguments used in [117] are the basis of many
papers where use is made of the simplification of equations by the elimination of
“fast modes”. Note that quasi-linear equations were first considered in [23, 24] in
connection with the description of the phase space consideration of the interaction
between waves and particles. For our purposes here, it is sufficient to consider only
some of the ideas advanced in the cited papers, namely, those associated with an
averaging of the quasi-linear equations [123–125].
We consider the continuity equation for the density of a passive scalar in an
incompressible flow:
∂n ∂n
+V = 0, (2.211)
∂t ∂x
where n(x, t) is the spatial density of the passive scalar and V (t) is the random
velocity field. We use the method of averaging over the ensemble of realizations for
(2.211), assuming that the density field can be represented as a sum of the mean
density n0 and the fluctuation component n1 = n − n,

n = n0 + n 1 . (2.212)

We also set n1  = 0 and V = v0 + v1 , where v0 = const and v1  = 0. As a result,

after simple manipulations (which are frequently used in the literature [123–125]),
we arrive at the following two equations:
 
∂n0 ∂n0 ∂n1
+ v0 + v1 = 0; (2.213)
∂t ∂x ∂x
 
∂n1 ∂n1 ∂n0 ∂n1 ∂n1
+ v0 + v1 + v1 − v1 = 0. (2.214)
∂t ∂x ∂x ∂x ∂x
We assume that the fluctuations n1 and v1 are as small as ε in comparison with
the mean density n0 . The quasi-linear character of the approximation indicates that,
94 O.G. Bakunin

in the equation for n0 , we keep the nonlinear term on the order of ε 2 but, in the
equation for n1 , we keep only the terms that are of the first order in ε. As a result,
the transformations put the equation for n1 into the form
∂n1 ∂n1 ∂n0
+ v0 = −v1 . (2.215)
∂t ∂x ∂x
We solve this equation by the method of Green functions. We consider (2.215) to be
a first-order linear hyperbolic equation with the source term I (x, t) = −v1 ∂n0 /∂x,
where the derivative ∂n0 /∂x is the parameter of the equation. We also supplement
the equation with the uniform initial condition n1 (x, 0) = 0. We then consider the
equation for the Green function G:
∂G ∂G
+ v0 = δ(x − x1 )δ(t − t1 ). (2.216)
∂t ∂x
It is easy to solve this equation by applying the Laplace transformation in time t and
the Fourier transformation in the spatial coordinate x:

G̃˜ k,s =
exp(−t1 s)
exp(ikx). (2.217)
s + ikv0
Here and below, the tildes mark the Fourier or Laplace-transformed quantities. The
solution has a simple physical meaning: it describes a perturbation propagating along
the characteristic z = x − v0 (t − t1 ):

G(x, t, x1 , t1 ) = δ x − x1 − v0 (t − t1 ) Θ(t − t1 ), (2.218)

where we have used Θ(t) to denote the Heaviside function. The solution for n1 (x, t)
has the form  t
∂ n (z, t)
n1 (x, t) = − v1 (t1 ) 0 dt1 . (2.219)
0 ∂z
We substitute this expression for n1 into (2.213) and perform simple manipulations
[123, 124] to obtain
 t
∂n0 ∂n0   ∂ 2 n0 (z, t1 )
+ v0 = v1 (t)v1 (t1 ) dt1 . (2.220)
∂t ∂x 0 ∂z∂x
The integral nature of this equation reflects the Lagrangian character of the relation-
ships between the derivatives of n0 (x, t). In this respect, the continuity equation at
hand is quite different from the fundamentally local continuity equation. The char-
acteristic that appeared in our analysis relates the derivatives at different times. The
left-hand side of (2.220) contains the partial derivatives with respect to x and t. On
the right-hand side, we sum the values of the derivative ∂ 2 n0 /∂x 2 calculated along
the characteristic with a weighting factor, which is the autocorrelation function of
velocity, C(t, t1 ) = v1 (t)v1 (t1 ). Thus, in the case of a steady random process, the
function C(t, t1 ) ≈ C(t − t1 ) in the equation under analysis plays the role of the
memory function. The final form of the transport equation in the framework of the
quasi-linear approach depends on the correlation function approximation.
 2 Correlations and Anomalous Transport Models 95

2.8.2 Short-Range and Long-Range Correlations

The particular form of the equation is governed by the choice of the correlation
function C(τ ). In the simplest physically meaningful case, (2.220) reduces to the
classical diffusion equation

∂n0 ∂n0 ∂ 2 n0 (x, t)

+ v0 =D . (2.221)
∂t ∂x ∂x 2
This is possible only if the main contribution to the integral on the right-hand side
of (2.220) comes from a short interval (t − t0 ; t) such that t0  t. If the second
derivative changes insignificantly over the short interval, we obtain
 t 
 ∂ 2 n0 (z, t) ∂ 2 n0 (x, t) t
v(t)v(t1 ) dt1 ≈ C(t − t1 ) dt1 . (2.222)
0 ∂z∂x ∂x 2 t−t0

In fact, we are assuming that the correlations are short-range. Thus, in this approx-
imation, we arrive at the familiar Kubo–Green formula for the diffusion coefficient
[16, 20]:  ∞
D= C(τ ) dτ . (2.223)
0
In terms of the δ-correlations [18–20], C(t − t1 ) ≈ C0 τ δ(t − t1 ), the quasi-linear
equation (2.220) takes the conventional form with the Taylor estimate of diffusivity
D ≈ V02 τ ≈ C0 τ :
∂n0 ∂n0 ∂ 2 n0 (x, t)
+ v0 = C0 τ . (2.224)
∂t ∂x ∂x 2
In the case of long-range correlations, we could assume that C ≈ const for t1  0,
and (2.220) reduces to
 t 2
∂n0 ∂n0 ∂ n0 (z, t1 )
+ v0 = C0 dt1 . (2.225)
∂t ∂x 0 ∂z∂x
This equation can be further simplified by using the properties of the characteristic z.
Differentiating (2.225) with respect to x gives

∂ 2 n0 ∂ 2 n0 t ∂ 3 n0 (z, t1 )
+ v 0 2 = C0 dt1 . (2.226)
∂t∂x ∂x 0 ∂x 3
Differentiating (2.225) with respect to t gives

∂ 2 n0 ∂ 2 n0 ∂ 2 n0 t ∂ 3 n0 (z, t1 )
+ v 0 = C 0 − v0 dt1 . (2.227)
∂t 2 ∂x∂t ∂x 2 0 ∂x 3
Eliminating the integral in (2.226) and (2.227) yields

∂ 2 n0 ∂ 2 n0  2 ∂ 2 n0
2
+ 2v0 + v 0 − C0 = 0. (2.228)
∂t ∂x∂t ∂x 2
96 O.G. Bakunin

This equation differs markedly from the classical diffusion equation. For C0 > 0, it
is a hyperbolic equation, possessing the corresponding properties. Thus, a complete
solution to this equation can be represented as a superposition of two initial distrib-
utions n0 (x, 0) moving at different velocities. As is known, the fact that hyperbolic
equations have characteristics opens new possibilities for describing nonlocal effects.
It should be noted, however, that, from the rigorous point of view, the above passage
from a parabolic to a hyperbolic equation is incorrect. The Cauchy problems for
these two types of equations are radically different. The well-known example of the
transport equation, which has a hyperbolic form, is the telegraph equation:

∂n ∂ 2n ∂ 2n
+ τ 2 = D 2. (2.229)
∂t ∂t ∂x
This equation was one of the first so-called nondiffusion equations to describe turbu-
lent transport [60, 92–95].

2.8.3 The Telegraph Equation

In the theory of random processes [18–20], one of the most widely used correlation
functions is an exponential one,

C(t) = C0 exp −|t|/τ . (2.230)

Here, τ is the characteristic time. This choice is quite natural because it is in this
form that the correlation function is used in the Langevin model of random-walk
processes. By means of this exponential function, we can transform integral equation
(2.220) into a partial differential equation. To do this, we set

C(t, t1 ) = C(t − t1 ). (2.231)

Differentiating (2.220) with respect to x gives


∂ 2 n0 ∂ 2 n0 t ∂ 3 n0 (z, t1 )
+ v0 2 = C(t − t1 ) dt1 . (2.232)
∂t∂x ∂x 0 ∂x 3
Differentiating (2.220) with respect to t gives

∂ 2 n0 ∂ 2 n0 ∂ 2 n0 1 t ∂ 2 n(z, t1 )
+ v 0 = C 0 − C(t − t 1 ) dt1
∂t 2 ∂t∂x ∂x 2 τ0 0 ∂x 2
 t
∂ 3 n0 (z, t1 )
− v0 C(t − t1 ) dt1 . (2.233)
0 ∂x 3
Eliminating the integral in these two equations yields
 2 
∂n0 ∂n0 ∂ n0 ∂ 2 n0  2 ∂ 2 n0
+ v0 + τ0 + 2v 0 + v 0 − C 0 = 0. (2.234)
∂t ∂x ∂t 2 ∂x∂t ∂x 2
 2 Correlations and Anomalous Transport Models 97

In accordance with the hyperbolic nature of the problem, we introduce the new set
of variables
ξ = t; (2.235)
η = x − v0 t (2.236)
to obtain the telegraph equation form:

∂n0 ∂ 2n ∂ 2 n0
+ τ 0 2 = C0 τ 0 2 , (2.237)
∂ξ ∂ξ ∂η
√
where C0 is the propagation velocity of the perturbations. This is actually telegraph
equation (2.23) in a frame of reference related to coordinates ξ, η.
One can see that the quasi-linear approach offers the possibility to bring into
accord the hydrodynamic equation for the particle density and correlation meaning
of the problem.

2.8.4 Magnetic Diffusivity and the Kubo Number

Quasi-linear ideas [23, 24] were used in the first papers [70, 126] devoted to the
description of the influence of the stochastic magnetic field on transport processes in
a plasma. This allows consideration of the magnetic diffusion coefficient Dm (2.111)
from the correlation point of view. The analysis of diffusion in the quasi-linear limit
is based on the stochastic equation for force lines

d r⊥  r⊥ ), B 
= b(z, b = ≈ b0 . (2.238)
dz B0

Here, a weak random field B  (Bx , BY , 0) is superimposed on a strong constant field

 0, B0 ) aligned with the z-axis and b0 is the characteristic relative scale of pertur-
B(0,
bations. This representation is analogous to the continuity equation of a test particle
motion in a flow (2.211). Jokipii and Parker [70] suggested a quasi-linear expression
for the transverse diffusion coefficient of the force lines of a magnetic field

1 ∞ 
 0)b(0,

 0) ∝ b02 LZ .
Dm = dz b(z, (2.239)
4 −∞
∞
Here, LZ is the longitudinal correlation length LZ = (1/b02 ) −∞ dz b(z, 0)b(0,
 0).
The determination of the interrelation between the magnetic diffusion coefficient Dm
and the diffusion coefficient of particles in the braided magnetic field is a complex
problem. However, there exists a simple model of “double” diffusion (2.116) sug-
gested in [69].
The simple quasi-linear estimate (2.238) that we considered earlier demonstrates
the necessity of a careful analysis of longitudinal and transverse correlation effects.
Therefore, the neglect of the transverse displacement Δ⊥ in expression (2.238)
 Δ⊥ ) ≈ b(z,
b(z,  0) is a serious drawback. A similar situation also arises in the theory
of the turbulent diffusion of a passive scalar.
98 O.G. Bakunin

The quasi-linear representation of the magnetic diffusion coefficient permits us to

obtain an estimate of particle diffusion in the stochastic magnetic field, which differs
significantly from the double diffusion. Thus, from the standpoint of the dimensional
analysis we can consider the expression
Lcor Lcor
D⊥ ≈ Dm ≈ b02 Lz . (2.240)
τ τ
Here, Lcor and is the particle correlation length and τ is the time that characterizes
the particle transport. However, if the longitudinal correlation length Lz that char-
acterizes the stochastic magnetic field is comparable with the longitudinal correla-
tion length Lcor , then using the expression for the longitudinal diffusion coefficient
D ≈ L2cor /τ we obtain
D⊥ ≈ b02 D . (2.241)
Such a representation for the effective diffusion coefficient of particles in the sto-
chastic magnetic field is called a “fluid limit” and appears to be one of the widely
encountered estimates of transport related to the stochastic magnetic field. It is im-
portant that the dependence on the amplitude of the magnetic field fluctuation b0 be
the same as in formula (2.238).
Many authors have tried to improve the methods of the quasi-linear approxima-
tion, since it has a very limited region of applicability. The Dupree papers [64–66]
are well known in this domain. Thus, the representation of the diffusion coefficient
in the form (2.238) will be valid only for the case when the diffusion displacement
in the transverse direction is much less than the transverse correlation scale length
b0 LZ  Δ⊥ .
The case of greatest interest arises when transverse correlation effects play a
central role:
b0 LZ ≥ Δ⊥ . (2.242)
Kadomtsev and Pogutse [67] suggested the use of a new approach and formulated a
criterion of its applicability in terms of the dimensionless parameter that character-
izes the ratio of longitudinal and transverse correlation effects:
b 0 LZ
Rm = > 1. (2.243)
Δ⊥
They related the regimes with Rm > 1 to the percolation character of the behavior of
streamlines [17]. Undoubtedly, it was a step forward, because the percolation meth-
ods are based upon the ideas of long-range correlations and fractality, which could
be relevant for “braided” magnetic field problems.

2.9 The Diffusive Renormalization

In the previous sections, different approximations of correlation effects were con-

sidered. One of the most effective is a diffusive approximation. Its effectiveness is
 2 Correlations and Anomalous Transport Models 99

related to the universality of the Gaussian distribution. Here, we will discuss the
diffusive representation of correlation effects by means of direct calculations of the
correlation function and by treating renormalized quasi-linear equations.

2.9.1 The Dupree Approximation

In the previous sections, we considered the quasi-linear approximation in the frame-
work of the hydrodynamic approach. However, the quasi-linear equations (which are
based on keeping a nonlinear term in the equation for mean density and using a non-
linear equation for perturbations) were first obtained in kinetics for considering the
problem of waves and particles interacting on the basis of Vlasov’s equation
∂f ∂f ∂f
+ V + E = 0, (2.244)
∂t ∂x ∂V

div E = 4πne f d V . (2.245)

Here, f is the velocity distribution function, E is the electric field and n is the plasma
density. The quasi-linear formulation of this problem has been repeatedly discussed
in detail [123–125]. Therefore, we will consider only those aspects that play an im-
portant role for subsequent considerations. A kinetic problem is naturally much more
complex. Thus, in the one-dimensional case the equations for the mean part of the
distribution function f0 and perturbation f1 have the form, which is analogous to
passive scalar equations (2.213), (2.214)
 
∂f0 e ∂f0
+ E = 0, (2.246)
∂t m ∂V
∂f1 ∂f1 e ∂f0
+V + = 0. (2.247)
∂t ∂x m ∂V
However, the diffusion equation in the velocity space, which was obtained as the
result of transformations

∂f0 ∂ ∂f0
= DV , (2.248)
∂t ∂V ∂V
with the quasi-linear diffusion coefficient
 2 
e |Ek |2
DV = dk, (2.249)
m ω(k) − kV
is added by the equation describing the energy dissipation of the electric field due to
the Landau damping
d
|Ek |2 = −2γk |Ek |2 , (2.250)
dt
where the characteristic frequency γk is defined by

∂f0
γk = 2πe2 ω δ(ω − kV ) dV . (2.251)
∂V
100 O.G. Bakunin

Here, ω(k) describes the frequency dependence on the wave number k, and |Ek |2
is the spectral function of the electric field. It is natural that the expression for the
quasi-linear diffusion coefficient (2.249) in a velocity space can be interpreted in
terms of the auto-correlation function of accelerations Ca ,
 ∞  2  ∞
e 
D≈ Ca (t) dt ≈ E(0)E(t) dt. (2.252)
0 m 0

One can see the analogy with the Taylor representation for the coefficient of turbulent
diffusion (2.6). However, in the case of phase space we deal with a more complex
problem. It is well known that the quasi-linear description of weak-turbulent plasma
is based on the notion of stochastic instability and randomness of phases. Indeed,
many theoretical and numerical investigations confirm the appearance of diffusion
in a space of velocities in the stochastic limit. In spite of the effectiveness of the
quasi-linear approximation, some correlation effects have not been considered. Thus,
the mixing process of stochastically instable trajectories leads to the nonlinear irre-
versible decay of correlations with the characteristic time
1
τk ∝ . (2.253)
(k 2 DV )1/3
Here, DV is the diffusion coefficient in velocity space and k is the wave num-
ber.
In the Dupree papers it was suggested considering the correlations decay by anal-
ogy with the Landau damping [64–66] using the frequency “renormalization” in the
form
i
ω(k) → ω(k) + . (2.254)
τk
Such an approach can be interpreted as the renormalization (modification) of the
equation for the perturbation of the distribution function f1
∂f1 ∂f1 ∂f0 f1
+ V + E = . (2.255)
∂t ∂x ∂V τk
Here, the term f1 /τk approximates the terms omitted in the quasi-linear approxima-
tion.
It is natural to consider the renormalization of the quasi-linear diffusion coeffi-
cient in terms of the autocorrelation function of accelerations,
 2
e    
C(t) = E x(t), t E x(0), 0 . (2.256)
m

Then, the particle velocity in the framework of one-dimensional electrostatic turbu-

lence is given by 
e t    
V (t) = V0 + E x(t ), t dt . (2.257)
m 0
 2 Correlations and Anomalous Transport Models 101

Representing the electric field as the totality of many independent Fourier compo-
nents, one obtains   
E(x, t) = Ek exp i(kx − ωk t) . (2.258)
k
The formal substitution of this expression into the formula for the correlation func-
tion yields
 2 
e     
C(t) = Ek exp i kx(t) − ωk t Ek  exp i k  x(0) . (2.259)
m
kk 

Then, by analogy with Corrsin, Dupree used the independence hypothesis:

 2 
e    
C(t) = |Ek |2 exp i kx(t) − ωk t + i kx(t) . (2.260)
m
k

For the Gaussian statistics one obtains exp A = exp A2  and hence the formula for
2

the correlation function takes the form

 2 

e k 2 x 2 (t)
C(t) = |Ek |2 exp i(kx − ωk t) − . (2.261)
m 2
k

Using the dimensional estimate dt d

V 2 (t) ≈ 4DV , it is easy to find the scal-
ing for x (t) ≈ 2DV t /3. The expression for the diffusion coefficient for one-
2 3

dimensional electrostatic turbulence for t → ∞ then takes the Dupree form [64–66]
 t
D= C(τ ) dτ
0
 2   ∞

e 1 2
= |Ek | exp i(kV − ωk τ ) − k DV τ dτ.
2 3
(2.262)
m 0 3
k

It is easy to note that this expression differs essentially from the quasi-linear one,
 2 
e
DQL =π |Ek |2 δ(ωk − kV ), (2.263)
m
k

where δ is the symbol of the Dirac function. Thus, the Dupree diffusivity scales
with Ek as D ∝ |Ek |3/2 , whereas the quasi-linear prediction is DQL ∝ |Ek |2 . Nat-
urally, the correlation effects approximation suggested by Dupree and based on the
independence hypothesis and dimensional estimates is fairly rough. Moreover, the
quasi-linear approximation is very effective for the overwhelming majority of mod-
els. However, it allows one to visualize correlation effects omitted in the quasi-linear
approach and opens new possibilities to obtain transport estimates [67, 71].
102 O.G. Bakunin

2.9.2 The Dupree Theory Revisited

The Dupree renormalized scaling DV ∝ |Ek |3/2 was tested in numerical test particle
simulations in [127] and later in [128]. The results were mixed, and no definitive con-
clusions drawn. The authors of [129] observed that the diffusivity numerically found
is significantly smaller than that predicted by the Dupree theory. Ishihara and Hirose
[130] confirmed their finding. Moreover, adopting the method proposed by Salat
[131], they recalculated the diffusivity without assuming Markovian process and
concluded that DV should be time-dependent [132]. An explicit analytical expres-
sion for the diffusivity has been represented by Salat [133] and Ishihara et al. [134].
It has been shown that in the asymptotic limit, DV scales with the turbulent field and
time as
DV ∝ |Ek |4/3 /t 1/3 . (2.264)
The predicted velocity variance [V ]2  ∝ t 2/3 in one-dimensional electrostatic tur-
bulence increases with time more slowly than the usual Brownian motion,
[V ]2  ∝ t. This indicates a diffusion process, which is not free but restricted and
dependent on the past history of particle trajectory.
The time integration is to be carried out along the perturbed particle trajectory
x(t) given by
x(t) = x0 + V0 t + x(t), (2.265)
where  
e t t  

x(t) = dt E x(t  ), t  dt  (2.266)
m 0 0
is the derivation from the free streaming trajectory x0 + V0 t.
Then, the velocity variance is formally given, with V = V0 , by
   t  t
 2  e 2 
 
V (t) = |Ek |2
dt ds  exp i(kV − ωk )s 
m 0 
t −t
k
   
× exp ik x(t ) − x(t  − s  ) .

(2.267)

For Gaussian statistics, the average in expression (2.267) can be approximated by

2 
    k  2 
exp ik x(t  ) − x(t  − s  ) ≈ exp − x(t  ) − x(t  − s  ) . (2.268)
2

In the quasi-linear theory, x = 0. This is equivalent to the assumption that the par-
ticles continue to experience the Eulerian field. In the original resonance broadening
theory by Dupree, the variance of particle trajectories is assumed to be independent
of the memory effects, which lead to the following approximation:
 2   2 
x(t  ) − x(t  − s  ) ≈ x(s  ) . (2.269)

The correlation function [x(t) − x(t − s)]2  was calculated more rigorously as
 2 Correlations and Anomalous Transport Models 103

follows. Each term in the expansion

 2   2     2 
x(t) − x(t − s) = x(t) − 2 x(t)x(t − s) + x(t − s) ,
(2.270)
is in the form of
 
x(t1 )x(t2 )
 2  t1  t  t2  t
e 1 2     
= dt1 dt1 dt2 dt2 E x(t1 ), t1 E x(t2 ), t2 . (2.271)
m 0 0 0 0

If the velocity variance is the result of diffusive process, we can make the following
approximation [132–134]:
 2  t1  t2
e     
dt1 dt2 E x(t1 ), t1 E x(t2 ), t2 ≈ 2DV min(t1 , t2 ), (2.272)
m 0 0

provided the time-dependence, if any, of DV is sufficiently weak. The substitution of

expression (2.272) into (2.271) yields
 t1  t2
 
x(t1 )x(t2 ) = 2DV dt1 dt2 min(t1 , t2 ). (2.273)
0 0

For t1 > t2 , the double integral reduces to (3t1 − t2 )t22 /6, and thus for t > s > 0, the
variance [x(t) − x(t − s)]2  becomes
 2  2
x(t) − x(t − s) = (3t − 2s)s 2 DV , (2.274)
3
which does depend on t as well as the relative time s. This non-Markovian nature
of the spatial variance will be responsible for the time-dependence of the velocity
diffusivity. Substituting this expression into the cumulant in the velocity variance
in expression (2.267), we finally obtain the following closed form for the diffusiv-
ity DV :
 2   t
e
DV (t) = |Ek |2
m 0
k


1
× exp i(kV − ωk )τ − k 2 DV τ 2 (3t − 2τ ) dτ. (2.275)
3

For resonant particles with V ≈ ωk /k, the upper limit of DV is given by

 2   t

e 1
DV = |Ek |2 exp − k 2 DV tτ 2 dτ. (2.276)
m 0 3
k
√ √
In the asymptotic limit t → ∞, the integral approaches 3π/2k Dt. Therefore,
104 O.G. Bakunin

the upper limit of the diffusivity is

 √ 2/3  2  2/3
3π e 1 1
DV max = 2
|E k |2
1/3
. (2.277)
2 m k t
k

Actually, in the approach suggested in [130–134], applying the substitution of a bal-

listic scaling x 2 (τ ) ∝ V 2 (t)τ 2 ∝ (DV t)τ 2 , where t is the parameter of the
integrand, for a Dupree dimensional approximation x 2 (τ ) ∝ DV τ 3 yields the
coordination of theoretical results and simulations.

2.9.3 The Taylor–McNamara Correlation Function

Taylor and McNamara [71] considered the problem of the direct calculation of the
Lagrangian correlation function for the description of strongly magnetized plasma.
However, we employ their method regardless of the plasma models. The analysis
carried out in [71] was based on the Fourier representation of Lagrangian velocities
appearing in the correlation function
   
C(t) = V x(t); t V x(0); 0
    
= Ṽk (t) exp ikx(t) Ṽk (0) exp ik  x(0) . (2.278)
k,k 

Here, · · · is the averaging symbol and Ṽk (t) is the Fourier transformation of the
velocity V (x, t) over the spatial x-coordinate. The next step in correlation decompo-
sition was called the “independence conjecture” [compare with (2.76)]:
    
C(t) = Ṽk (t)Ṽk (0) exp ik x(t) − x(0) . (2.279)
k

Taylor and McNamara then used the Corrsin and Dupree conjecture [28, 64–66] of
“the diffusion behavior of trajectories” [x(t) − x(0)]2 ∝ Dt. Here, D is the diffusiv-
ity, which depends on the model. This is in fact a “recipe for calculating the average”
of the value exp[ikx(t)] in accordance with the conventional formula

2 
A 
exp A = exp . (2.280)
2

Performing calculations yields exp[ikx(t)] = exp[−k 2 Dt]. The expression for

the correlation function was obtained in the form
  
C(t) = Ṽk2 exp −k 2 Dt . (2.281)
k

Prior to proceeding with the presentation of the results of [71], we consider formula
(2.281) from the “correlation” standpoint. This expression can be interpreted as the
sum of the Gaussian exponential correlation functions with “weight factors”, which
 2 Correlations and Anomalous Transport Models 105

is proportional to the turbulence spectrum E(k):

  
t
C(t) ∝ E(k) exp − . (2.282)
τk
k

Here, τk = 1/(k 2 D) is the characteristic time corresponding to the scale k. It is nec-

essary to take into account that this sum of the larger number of exponents can turn
out to be a nonexponential function. Similar expressions are most extensively em-
ployed to obtain correlation functions with power tails [17–22]. Another important
feature of formula (2.281) is the following formula for the diffusion coefficient:
 ∞ 
1  E(k) ∞  
D= C(t) dt = 2
exp −k 2 Dt d k 2 Dt
0 D k 0
k

1 
≈ E(k)/k 2 dk. (2.283)
D
We note the similarity of the resulting expression to the Howells result:
 ∞
E(k)
D2 = dk. (2.284)
k k2
One can see from the above analysis that the Howells ideas of the interaction of
scales and the correlation ideas of “diffusive behavior of trajectory” [28, 64–66] are
closely interrelated. Taylor and McNamara introduced a quantity S(t) = [x(t)]2 .
Then the expression for the correlation function takes the form

d d x 2 d2
C(t) ≈ D≈ ≈ 2 S(t). (2.285)
dt dt t dt
On the other hand, from formula (2.283) we obtain

   
C(t) = E(k) 1 − exp −k 2 S(t) dk. (2.286)

2
2 S(t) = F {S(t)}. After
d
In fact, one has a “Newtonian”-type differential equation dt
a simple transformation we obtain the final solution [71]:

dk    
D2 = 2
E(k) 1 − exp −k 2 S(t) . (2.287)
k
To obtain (2.284) it is sufficient to consider the case k 2 S(t)  1. Note that [71] does
not contain a reference to Howells’s paper [73]. Apparently, the Howells result has
become widely known more recently due to the Moffat analysis of turbulent transport
problems [32, 33].
In the framework of the Taylor and McNamara diffusion approximation the value
D is the effective diffusivity; this differs essentially from concepts based on the use
of “seed” diffusivity as the “decorrelation mechanism”. Thus, Wang and co-workers
[135] suggested a modification of the Corrsin conjecture (2.74) by a substitution of
106 O.G. Bakunin

the Taylor expression RT (t)2 for the mean square displacement 2D0 t:
 t
RD (t) ≈ 2
2
(t − t  )C(t  ) dt. (2.288)
0

Formal calculations lead to a complex nonlinear integral equation for C(t) [135].
For the purposes of this paper it is sufficient to consider the simplified model of the
diffusive evolution of the “correlation cloud” (2.98):

 t −d/2
V02 V02   
C(t) ≈ ≈ (t − t )C(t ) dt . (2.289)
nRD (t)d n 0

The simplest case considered by Wang and co-workers [135] corresponds to d = 2;

therefore, the approximation equation to define the Lagrangian correlation function
takes the form  t
V2
C(t) (t − t  )C(t  ) dt  = 0 . (2.290)
o n
Unfortunately, even the simplified version of the nonlinear equation cannot be ana-
lytically solved. However, numerical simulation that was carried out in [135] permits
one to consider complex correlation effects in a stochastic magnetic field.

2.9.4 The Kadomtsev–Pogutse Renormalization and the Stochastic Magnetic

Field

The obvious drawback of the quasi-linear theory is that the nonlinear term in the
equation for n0 is retained while the nonlinear terms in the equation for n1 are omit-
ted. Many authors have tried to refine the quasi-linear approximation. A detailed
analysis of such papers was carried out in [125]. In fact, the equation for n1 (2.215)
is linear and hyperbolic and it keeps the Lagrangian character of the correlations.
This opens up the possibility of describing the omitted correlation effects by includ-
ing the additional diffusive term.
In this context, it is expedient to present some of the results obtained by Kadomt-
sev and Pogutse [67] on anomalous electron transport in a magnetic field. They con-
sidered a three-dimensional problem in which a weak random field B  (Bx , BY , 0) is
superimposed on a strong constant field B(0,  0, B0 ) aligned with the z-axis. The for-
mal quasi-linear representation is valid only when the diffusion-related displacement
in the transverse direction is much less than the transverse correlation length (2.239).
Kadomtsev and Pogutse considered a more complex case than the quasi-linear one.
They introduced a continuity equation for the density of the magnetic field lines,

∂nb  B 
+ b∇⊥ nb (r⊥ , z) = 0, b = ≈ b0 , (2.291)
∂z B0
and represented nb as a sum of the mean density n0 = nb  and the fluctuation
component n1 ,
nb = n0 + n1 . (2.292)
 2 Correlations and Anomalous Transport Models 107

Here, b0 is the relative scale of the fluctuation amplitude and · · · is the averag-
ing symbol. The problem as formulated is close to problems (2.213) and (2.214) of
the quasi-linear diffusion of a passive scalar. In fact, the authors of [67] wrote the
equation for n0 in the traditional form:
∂n0  1  = 0.
+ ∇⊥ bn (2.293)
∂z
However, in the equation for n1 (2.214), they replaced the second-order terms
 
∂n1 ∂n1
v1 − v1 (2.294)
∂x ∂x

(which were omitted in earlier studies) by a diffusion term Dm ∇⊥

2 n . In essence, they
1
followed Corrsin’s and Dupree’s ideas and related the discarded correlation effects
to the diffusive spreading of trajectories:
∂n1  ⊥ n0 .
− D m ∇⊥
2
n1 = −b∇ (2.295)
∂z
Here, the effective diffusion coefficient was considered as diffusivity. This is similar
to the Taylor–McNamara model. The set of renormalized equations (2.293)–(2.294)
has retained a convenient form for solution. For passive scalar equations it corre-
sponds to the set of equations
 
∂n0 ∂n1
+ v1 = 0; (2.296)
∂t ∂x

∂n1 ∂n0 ∂ 2 n1
+ v1 =D 2 , (2.297)
∂t ∂x ∂x
which are similar to (2.213) and (2.215). Thus, they kept (2.297) linear but passed
from a hyperbolic equation of form (2.215) to the parabolic equation.
Applying the mathematical apparatus of Green’s functions to (2.297), we obtain
∂G
− D m ∇⊥
2
G = δ(r − r ). (2.298)
∂z
Kadomtsev and Pogutse derived the following equation for the mean density n0 :
∂n0
= D m ∇⊥
2
n0 , (2.299)
∂z
where the magnetic diffusion coefficient and the Fourier spectrum of perturbation
amplitudes are given by
 
1 b2 (k) 
Dm = d k, (2.300)
2 ikz + k⊥2D
m
108 O.G. Bakunin

1  
 =
b2 (k) r ) d r.
b(0)b(r) exp(−i k (2.301)
(2π)2
2 D , they got the quasi-linear expression
For kz > k⊥ m

π
Dm = 
d k b2 (k)δ(k z ) ∝ b 0 LZ ,
2
(2.302)
2
where LZ is the longitudinal correlation length.
In the case of strong correlations kz < k⊥ 2 D , the authors of [67] arrived at the
m
following result, which is similar to that obtained by Howells [73]:
 
1 b2 (k) 
2
Dm = 2
d k. (2.303)
2 k⊥

This result for the magnetic diffusion coefficient

Dm ≈ b0 Δ⊥ (2.304)

corresponds to the Howells estimate DH ≈ V0 Δ but for the anisotropic case. It once
again shows that, on the one hand, it is important to take into account the additional
correlation effects that are neglected in the quasi-linear approach and, on the other
hand, these correlations are closely related to interactions between different spatial
scales. Note that in contrast to the Dreizin–Dykhne model, here one deals with per-
pendicular correlations. Moreover, in spite of the simple form of the obtained esti-
mate (2.304), the linear character of the dependence of the effective diffusion coeffi-
cient Deff on the “stochastic layer” width Δ⊥ is used to describe turbulent transport
in models with convective cells, percolation transport, etc.

2.10 Anomalous Transport and Convective Cells

The analysis of complex structures plays an important role in the description of tur-
bulent transport in both fluids and plasma. On the one hand, in the presence of struc-
tures we deal with essential spatial inhomogeneity (regions of convective transport).
On the other hand, many correlation effects can still be approximated by diffusion
models. The focus of this section is the derivation of the expression for the effec-
tive diffusion coefficient, which is based on the balance of convective and diffusive
fluxes, in the convective cells system.

2.10.1 Bohm Scaling and Electric Field Fluctuations

The important peculiarity of the renormalized transport coefficients of Dupree,

Taylor–McNamara, and Kadomtsev–Pogutse is the linear dependence of the effec-
tive diffusivity on the perturbation amplitude, which differs essentially from the
quasi-linear square-law dependence. Thus, in the Dupree approach DV ∝ |Ek |;
in the Taylor and McNamara model Deff ∝ E(k)1/2 /k ∝ Vk ; and in the case of
 2 Correlations and Anomalous Transport Models 109

the Kadomtsev–Pogutse renormalization method one deals with the estimate Dm ∝

b0 Δ⊥ . In this relation, it is necessary to note that the question about the character
of the dependence of the effective diffusivity on the perturbation amplitude has not
only theoretical but also practical importance. For example, the analysis of transport
in magnetized plasma leads to the necessity to consider the character of the depen-
dence of turbulent diffusion across a magnetic field B on its magnitude. One of the
important approaches is the consideration of drift motions of plasma in crossed elec-
tric E and magnetic B fields:

[B × E]
 ∇ϕ
VE = c 2
∝ . (2.305)
B B
Here, VE is the drift velocity and ϕ is the potential of the electric field. Then, in terms
of the Kubo number Ku ≈ V0 /λω ≈ k 2 ϕ/ωB we can analyze the dependence
 2 βC  βC
k ϕ βC kϕ
D⊥ ∝ Ku ≈ βC
∝ VE ≈ . (2.306)
γB B
Here, k is the wave number and γ ≈ ω is the characteristic frequency. The conven-
tional representation of the collisional character of transverse diffusion in a strong
magnetic field leads to the estimate [5–8]
√
ne2 c2 me
D⊥ ≈ √ . (2.307)
B2 T
Here, e is the electron charge, me is the electron mass, T is the plasma temperature,
and n is the plasma concentration. Actually, this corresponds to βC = 2. The prog-
noses based on this formula provide fairly good confinement of plasma in magnetic
traps. However, much experimental data points to the incorrectness of this estimate.
The main reason that prevents good confinement is strong plasma turbulence. The
simplest Bohm consideration [136] of the effects of turbulent fluctuations of electric
fields leads to the linear dependence of the transverse diffusion on the perturbations
amplitude with βC = 1. The Bohm estimate of plasma diffusivity across a magnetic
field is based on the notion of the existence of eddies or nonuniformity in turbulent
plasma, which lead to chaotic fluctuations of electric fields. If we introduce a char-
acteristic scale lB corresponding to the structures under analysis then the simplest
correlation estimate of the diffusion coefficient is the expression

lB2
DB ≈ . (2.308)
τcor
Here, lB plays the role of the spatial correlation scale and τcor is the characteristic
correlation time that can be estimated from a dimensional analysis; introducing into
consideration the velocity VE characterizing drift motion in crossed electric δ E and
magnetic B fields yields
lB l2
τcor ≈ ≈ B B0 . (2.309)
VE cδϕ
110 O.G. Bakunin

Here, use is made of the expression for the drift velocity VE ≈ (c/B0 )δE ≈ (c/B0 ) ×
(δϕ/ lB ) and the dimensional estimate of electric field fluctuations through the po-
tential perturbation across the structure, δϕ ≈ δElB . Upon substitution we obtain the
Bohm scaling for transverse diffusion in magnetized turbulent plasma:
  
c cT eδϕ 2 1/2
DB ≈ lB VE ≈ δϕ ≈ , (2.310)
B0 eB0 T
where T is the plasma temperature. Here, use is made of energetic normalizing of
electric field fluctuations, since the dimensional consideration yields

eδϕ ≈ eδElB ≈ T . (2.311)

The absence of the characteristic scale of structures lB in the final expression is an

important particularity of the Bohm scaling, which attaches a “universal” character
to this estimate.
The simplest correlation interpretation of the appearance of the linear depen-
dence of Deff ∝ V0 ∝ Ku is the consideration of the decorrelation time τ in the
Taylor definition
DT ≈ V02 τ. (2.312)
Formally, the estimate τ ≈ 1/ω could be used as the decorrelation time. However, in
the presence of complex structures (like convective cells) trapping could be important
and one can estimate τ ≈ λ/V0 , where λ is the structure spatial scale. If the values
V0 are large then trapping is the main correlation mechanism, since λ/V0 < 1/ω,
and hence
λ
Deff ≈ V02 ≈ V0 λ. (2.313)
V0
In the next section, we will consider how to obtain the linear dependence of Deff on
V0 in more detail.

2.10.2 The Bohm Regime and Correlations

The scaling suggested by Bohm can be interpreted in terms of the renormalization

method that was used by Dupree [64–66] and Taylor and McNamara [71]. In this ap-
proach, the Bohm scaling is integrated in terms of the Howells interaction of different
scales, which is related to the “diffusion” character of decaying correlations. Thus,
the authors of [71] considered an auto-correlation function for the case of strongly
magnetized plasma by the guiding centers approximation
 2
c  
C(τ ) = δE⊥ (τ )δE⊥ (0) , (2.314)
B
where δE⊥ is the electric field fluctuation in the direction perpendicular to the mag-
netic field. Using the Fourier representation for electric field fluctuations
  
δE⊥ (x, t) = δEk (t) exp i kx(t) (2.315)
k
 2 Correlations and Anomalous Transport Models 111

and the independence hypothesis [28], the expression for the correlation function can
be rewritten in the form
 2 
c   
C(t) = δEk (t)δEk  (0) exp i kx(t) + k x(0)
B
kk 
 2 
c    
= δE(t)δE(0) exp i k x (t) . (2.316)
B
k

Here, x(t) is the diffusive particle displacement due to the presence of turbulent
fluctuations. On the basis of the Gaussian statistics one obtains the expression
    
 x (t) = exp −k 2 D⊥ (t) ,
exp i k (2.317)

which is analogous to the Corrsin and Dupree representations [28, 64–66]. Then,
the formal expression for the turbulent diffusion coefficient in terms of the Taylor
definition is given by
 ∞  ∞ c
2 
   
D⊥ = C(t) dt = δE⊥ (t)δE⊥ (0) k exp −k 2 D⊥ t dt.
0 0 B
k
(2.318)
Taylor and McNamara assumed that the spectrum of statistical fluctuations of electric
field δE 2 k is known and the decaying correlations have the exponential form
     
δE⊥ (t)δE⊥ (0) k = δE 2 k exp −k 2 D⊥ |t| (2.319)

with the characteristic correlation time in the diffusive form τ ≈ 1/k 2 D⊥ . Integrat-
ing the expression for the renormalized correlation function over time yields
 ∞ 2 
c  2  
D⊥ = dt δE k exp −2k 2 D⊥ t
0 B
k
 2 
c 1 δE 2 k
= . (2.320)
B D⊥ 2k 2
k

Applying the spectrum δE 2 k in the form [71]

 2 T
δE k = 4π , (2.321)
1 + (kλD )2

where λ2D = nT /4πe2 , and integrating over k, it is easy to obtain the formula
 2 
c d k 1
2
D⊥ = 4πT . (2.322)
B (2π) 2k [1 + (kλD )2 ]
2 2

Here, T is the temperature and n is the density. This expression can be transformed
112 O.G. Bakunin

into the form containing the Bohm factor DB ≈ cT /eB

 
cT e2 ∞ dk
D⊥ = . (2.323)
eB T 0 k[1 + (kλD )2 ]

Here, use was made of the integration over the azimuth angle θ : d k = k sin θ dk.
The divergence of this integral in the region of the small wave numbers k reflects
the slow correlation decay and the necessity for a more detailed consideration of
different scale interactions.

2.10.3 Convective Cells and Transport

The system of convective cells is one of the simplest models, permitting one to es-
timate transport related to self-organized structure. Thus, it is well known that low-
frequency drift waves can act as convective cells and cause rapid particle transport if
they are excited to a large amplitude [3–8]. The regular character of the location of
structure elements simplifies the analysis considerably. However, this model is im-
portant and has properties corresponding to a more complex system. The subsequent
progress of research on diffusion processes in systems with convective cells (see
Fig. 2.4) [80, 81] has led us to the understanding of the importance of the stochastic
layer width Δ and the convective fraction of the transport. The simplest estimate of
the longitudinal convective transport is as usual the quasi-linear expression
Δ
D ≈ V02 τ , (2.324)
λ
which is corrected by the geometrical factor Δ/λ to account for the fraction of space
that is responsible for the convection. Here, λ is the cell size and V0 is the charac-
teristic velocity of the convective flow. It is natural to use the characteristic time of
leaving the particle from the boundary layer as the correlation time

τ ≈ Δ2 /D0 . (2.325)

Fig. 2.4. System of convective cells

 2 Correlations and Anomalous Transport Models 113

Here, D0 is the “seed diffusion” coefficient. Then, the scaling for the convective
transport takes the form
Δ3
Deff ≈ V02 . (2.326)
λD0
However, the quasi-linear character of the dependence of D on V0 does not corre-
spond to the results of numerous simulations in the region
λV0
Pe ≈ ≥ 1. (2.327)
D0
A qualitatively new estimate could be the linear dependence Deff ≈ V0 Δ. Formally,
we must consider the effective diffusivity in the convective form

Deff ≈ λV0 P∞ , (2.328)

where P∞ is the fraction of space, which is responsible for convective transport. In

the case of convective cells, the value of P∞ can be estimated if we introduce the
cell size λ:
λΔ Δ
P∞ ≈ 2 ≈ . (2.329)
λ λ
This formula reflects the simple “topology” of the model of convective cells. The
diffusive estimate of the stochastic layer width Δ was considered in [80, 81],

D0 λ 
Δ= ≈ D0 τ . (2.330)
V0

Here, τ is the correlation timescale. This formula clearly defines the physical sig-
nificance of the particle number balance. The number of particles escaping from a
convective cell per unit time is
n
N D ∝ D0 λ. (2.331)
Δ
The convective flow along the boundary layer carries away a number of particles:

NC ∝ nV0 Δ. (2.332)

Since convective flows exist only in the fraction of space Δ/λ, we obtain
Δ
Deff ∝ λV0 = V0 Δ. (2.333)
λ
Note that the expression obtained is analogous to the nonquasi-linear result of
Kadomtsev and Pogutse (2.304). The authors of [80, 81] eventually arrived at the
following estimate for the turbulent diffusion coefficient:
 1/2
Deff = const · D0 V0 λ ≈ D0 Pe1/2 ∝ V0 . (2.334)

This representation of the result in terms of the Peclet number differs significantly
from both the quasi-linear and linear estimates Deff ∝ V02 .
114 O.G. Bakunin

2.10.4 Complex Structures and Convective Transport

In conditions where cells are not regular, the analysis of transport can be based on the
statistical properties of the streamline function landscape ψ(x, y). Convective flows
of particles along percolation channels contribute most to the effective diffusive co-
efficient. The topology of these channels is characterized by the second derivative of
streamline function ψ  (r). Therefore, introducing a characteristic scale of stream-
line function ψ0 and a characteristic spatial scale r0 , we obtain an estimate of the
channel width responsible for convection Δ:

ψ  (r)Δ2 ≈ D0 . (2.335)

Applying the scaling approach for ψ  , we obtain an expression for the definition of
layer width Δ:
ψ0
Δ2 2 ≈ D0 . (2.336)
r0
Simple calculations allow the derivation of the dependence of layer width Δ on the
flow parameters ψ0 ,D0 , r0 :
 1  1
D0 2 D0 2
Δ = r0 = r0 , (2.337)
ψ0 r0 V0
or in terms of the Peclet number,
1
Δ = r0 . (2.338)
Pe1/2
Here, V0 is the characteristic velocity scale.
Using the expression obtained above for effective diffusion in a system of random
convective flows in the form

Deff ≈ V0 Δ(ψ0 , D0 , r0 ), (2.339)

we easily find a scaling describing transport,

1
Deff ≈ D0 Pe 2 , (2.340)

which is in good agreement with the above.

This approach is an attractive one, which could possibly provide an alternative
starting point for the analysis of transport effects [40]. However, such an approach is
slightly naive and does not use all the possible information concerning flow topology;
therefore, it cannot be applied for a rigorous analysis of fractal streamlines. These
problems will be considered in the next sections in more detail.

2.11 Stochastic Instability and Transport

The problems of tracer description and the analysis of transport in a stochastic mag-
netic field have many traits in common. Thus, in the case of magnetized plasma
 2 Correlations and Anomalous Transport Models 115

we deal with the anisotropic medium in the presence of several “seed” diffusive
mechanisms simultaneously. Moreover, the stochastic instability of trajectories [74]
also leads to the appearance of new decorrelation mechanisms. In this section, the
Rechester–Rosenbluth transport model [75], in which stochastic instability plays a
major role, will be considered.

2.11.1 Stochastic Instability and Correlations

The problem of the divergence of nearby force lines (see Fig. 2.5) has been repeatedly
discussed in many books and articles [3–8]. It is obvious that this physical mecha-
nism has a significant influence on the process of particle decorrelation in a stochas-
tic magnetic field. For example, it “destroys” the double diffusion regime, which is
based on the repeated returns of magnetized particles under conditions of longitudi-
nal diffusive motion along the magnetic field (2.117). Ptuskin [137] considered this
effect in terms of Richardson’s relative diffusion [58]. This approach demonstrates
the possibilities of a nonquasi-linear method of transport description in the stochastic
magnetic field and the importance of using correlation ideas. In the simplest steady
case, the equation describing the walks of the separated force line can be expressed
in a form that is analogous to (2.1),

d r⊥ B
= . (2.341)
dz |B|

Here, z characterizes the distance that was traveled along the force line and B(z)
is the value of the magnetic field. In the framework of the simplified scenario—the
regular component of the magnetic field is absent, B = 0, and the absolute value
of the magnetic field is constant, /B/ = const—we obtain the following expression,
which is analogous to Taylor’s formula (2.6)
    z  ∞
Bi (z1 )Bj (z2 )
ri rj  = dz1 dz2 ≈ dz dg Cij (g). (2.342)
2
|B| −∞

Fig. 2.5. Stochastic instability of trajectories

116 O.G. Bakunin

Here, Cij (g) is the correlation function of the random magnetic field, which char-
acterizes the correlation decay along the chosen force line. Such a consideration is
correct only for cases where the force line length is much greater than the correla-
tion length Lz . It is natural that the expression obtained in the framework of Taylor’s
approximation (2.6),  ∞
d  2
ri ≈ dg Cii (g), (2.343)
dz −∞
can be employed for this case. Obviously, the nonformal analogy of the equations that
describe the particle walk and the force line walk allows us to consider the problem
of the relative divergence of two force lines [3–8]. The author of [137] considered
a relative displacement of two nearby force lines during the process of their random
walk:  z
B (1) (z) − B (2) (z)
r = r(1) − r(2) = dz . (2.344)
|B|
(1) (2)
It is convenient to suppose that r0 and r0 are the initial points and the force lines
pass through these points. The vector r connects two points of two different force
(1) (2)
lines that have started from the corresponding initial points r0 , r0 and have passed
an identical length z. Formal calculations allow us to obtain the expression
 z  ∞

ri rj  ≈ dz dg Cii (g, 0) + Cjj (g, 0)
−∞

− Cij (g, r ) − Cij (−g, r ) . (2.345)

Here,   denotes averaging over the ensemble. It is necessary to introduce an addi-

tional variable r in the expression for the correlation function Cij (g, r ), since
we deal with the correlations of the field on two different force lines. In the case
where correlations between the two force lines under consideration are absent, we
obtain the result that corresponds to the Taylor representation (2.6),
 z  ∞
 2
ri ≈ 2 dz dg Cij (g, 0). (2.346)
−∞

However, taking into account the correlation effects that arise over the small distance
r and using a decomposition over r in (2.345), the following expression was
obtained:   ∞
2 
 2 1 z ∂ Cii (g, 0)  2 
ri ≈ dz dg ri . (2.347)
3 −∞ ∂ρ 2
Here, ρ characterizes the distance across the tube of the force lines that corresponds
to using Cij (g, ρ). Now, it is easy to obtain the exponential estimate that character-
izes the process of divergence of nearby correlation force lines:

 ∞ 
 2  2 z ∂ 2 Cii (g, 0)
ri ≈ ri 0 exp dg . (2.348)
3 −∞ ∂ρ 2
 2 Correlations and Anomalous Transport Models 117

In spite of the qualitative character of the estimates, this result shows the necessity of
the modification of the dimensional subdiffusive expressions for particle transport in
the stochastic magnetic field (2.117), taking into account the decorrelation processes
of force lines.

2.11.2 The Rechester–Rosenbluth Model

Stochastic instability of trajectories is a nontrivial decorrelation mechanism [74, 75].

On average, two initially close streamlines diverge from one another according to the
law  
z
Δ(z) = l0 exp . (2.349)
LK
Here, l0 is the initial separation of the streamlines and z is the distance that was
passed along the streamline. The magnitude h = 1/LK is called the Kolmogorov
entropy and defined through
  
1 Δ(z)
h = lim ln . (2.350)
l0→0 ,z→∞ z l0
One can see that correlation scales need not be defined by the seed diffusion process
alone.
Rechester and Rosenbluth [75] assumed that decorrelation is related to stochastic
instability and not to particle collisions. In the problem as formulated the collisional
random walks of magnetized particles along and across streamlines of the magnetic
field play the role of seed diffusion. The simplest estimate can be obtained by a
consideration of the expression for the transverse diffusion coefficient of particles in
terms of the magnetic diffusion coefficient (2.112)

Δ2cor Lcor
Deff ≈ ≈ Dm . (2.351)
2τ τ
Here, Lcor is the longitudinal correlation length related to stochastic instability and
τ is the correlation time. This approach implies that the magnetic diffusion coef-
ficient Dm , the collisional longitudinal diffusion coefficient D , and the collisional
transverse diffusion coefficient D⊥ are known. The analysis is often carried out in
terms of the heat-conduction coefficient in order not to complicate the problem by
taking into account ambipolar effects. However, we will keep using the diffusion
symbols in order to retain the uniformity of the terminology. The values Lcor and τ
are interrelated by the expression for the longitudinal diffusion coefficient:

L2cor
D ≈ . (2.352)
2τ
We then obtain the following estimate for Deff :
2D
Deff ≈ Dm . (2.353)
Lcor
118 O.G. Bakunin

Fig. 2.6. The stochastic instability of phase element

Here, Lcor is the parameter of the problem. To define the value Lcor we use expres-
sion (2.349)
  
Δ(Lcor ) r0
Lcor ≈ LK ln ≈ LK ln . (2.354)
l0 l0
Here, r0 is the transverse spatial scale. The authors of [75] assumed that the values
r0 and LK are known and are related to the specific model. Therefore, to determine
Lcor it is necessary to determine only l0 .
For this purpose, Rechester and Rosenbluth considered a small element of evolv-
ing area (see Fig. 2.6), which has a spatial scale l0 . Because of stochastic instability
there will be two competing processes at the same time: the distance between stream-
lines will be exponentially increasing and the width of the area will be exponentially
decreasing in order to conserve the total area:
 
z
δ(z) = l0 exp − . (2.355)
LK
Here, δ is the width of the area. It is necessary to take into account the influence of
perpendicular diffusion processes that increase δ. The authors of [75] considered a
balance between these processes:
dδ δ
=− , (dδ)2 ≈ 2D⊥ dt. (2.356)
dz LK
Taking into account the expression that describes the longitudinal diffusive behavior
(dz)2 ≈ 2D dt, one obtains the formula

D⊥
l 0 ≈ δ ≈ LK . (2.357)
D
The Rechester–Rosenbluth model is based on the assumption of the deterministic
role of the stochastic Minstability of trajectories. However, there is also an alterna-
tive possibility related to the agreement between the longitudinal and transverse dif-
fusion mechanisms in a strongly anisotropic medium. Thus, in the case of strongly
 2 Correlations and Anomalous Transport Models 119

magnetized plasma we deal with D  D⊥ . Kadomtsev and Pogutse used this ap-
proach in their paper [67] that is devoted to the anomalous transport mechanisms
in a braided magnetic field. In contrast to Rechester and Rosenbluth, the authors of
[67] related longitudinal correlation length Lcor to the longitudinal diffusive process:
D ≈ L2cor /τ , and the decorrelation time τ was related to transverse diffusion:
D⊥ ≈ r02 /τ . Here, r0 is the characteristic scale in the transverse direction. Calcu-
lations then lead to the formula

D D⊥
Deff ≈ Dm , (2.358)
r0
which is one of the possibilities for relating the different diffusion mechanisms in the
anisotropic medium. There is a new kind of dependence on D here.
It is necessary to note that in the case of particle diffusion in the stochastic mag-
netic field we are dealing with a large variety of regimes: double diffusion (2.116),
ballistic behavior (2.113), fluid limit (2.241), the Rechester–Rosenbluth model
(2.353), and the Kadomtsev–Pogutse regime (2.358). Moreover, there are several
other regimes, which will be considered in the following sections.

2.11.3 Collisional Effects and the Stix Formula

The expression used for the effective diffusion coefficient (2.353) permits us to carry
out an interesting parametric analysis. For example, it is possible to estimate the
influence of collisions on the effective diffusion coefficient. Following Stix [138],
let us introduce the Ncoll factor accounting for the number of collisions during the
correlation time (compare it with the Kubo number Ku = V0 /λω):
τ
Ncoll ≈ . (2.359)
τcoll
Here, τcoll is the collisional timescale and τ is the correlation time. The expression
for the coefficient of longitudinal diffusion has the form

L2cor λ2
D ≈ ≈ coll . (2.360)
2τ 2τcoll
Here, λcoll is the collisional longitudinal mean free path. Hence, one can obtain

Lcor ≈ λcoll (Ncoll )1/2 . (2.361)

On the other hand, we can represent the longitudinal diffusion coefficient as

D ≈ λcoll V . (2.362)

Here, V is the particle characteristic velocity in the longitudinal direction. Hence,

we can rewrite expression (2.353) in the following form:
1
Deff ≈ Dm V √ . (2.363)
Ncoll
120 O.G. Bakunin

We see that collisional effects decrease the effective diffusion coefficient in com-
parison with the collisionless case Deff ≈ Dm V . However, in the framework of
the approach of [75, 138, 139] it does not lead to the appearance of double diffu-
sion (2.116).
This expression for Deff in the Rechester–Rosenbluth theory is reminiscent of
the Dreizin–Dykhne estimate
1
Deff ≈ V02 t , (2.364)
NI (t)
√
where NI (t) ≈ 2D0 t/a is the number of interactions between the particle and
transverse flows, which have the “collisional” meaning.
Stix [138] also suggested a somewhat different method to estimate the effective
diffusion coefficient for the Rechester–Rosenbluth regime. Considering Ncoll as the
main parameter of the model, the author of [138] used a transcendental equation to
define Ncoll . The basis of this equation is estimate (2.354),
 
Lcor
r0 = l0 exp . (2.365)
LK
However, the assumption was made that the value l0 is in agreement with the cor-
relation time τ , l02 ≈ D⊥ τ . Then, using the expression describing the relationship
between the longitudinal correlation length Lcor and the correlation time τ yields the
estimate for l0 , which differs from the Rechester–Rosenbluth result

D⊥
r0 = Lcor  Lcor . (2.366)
D

The equation for the longitudinal correlation length Lcor takes the form
  
D⊥ Lcor
r0 = Lcor exp . (2.367)
D LK

Here, as before, the parameters of the problem are D⊥ ,D , r0 , LK . However, we

can re-formulate this equation for Lcor in terms of the parameter Ncoll . Using the
expressions for D⊥ and D we obtain

Δ2⊥ λ2
D⊥ ≈ , D ≈ . (2.368)
τ τcoll
Here, Δ⊥ is the transverse correlation scale. Substitution of (2.368) into (2.367)
yields  √ 
 λ Ncoll
r0 = Δ⊥ Ncoll exp . (2.369)
LK
This transcendental equation for Ncoll can be used to calculate Deff in accordance
with formula (2.363).
 2 Correlations and Anomalous Transport Models 121

Concluding this section, note that the Kadomtsev–Pogutse result [67] can also
be rewritten in a form that uses the collisionless diffusion coefficient with the ad-
ditional factor. The particle longitudinal velocity enters into the expression for the
longitudinal diffusion coefficient:

D ≈ V 2 τcoll . (2.370)

On the other hand, the transverse diffusion coefficient has the form D⊥ ≈ Δ2⊥ /τcoll .
Here, Δ⊥ is the collisional transverse correlation scale. As a result of simple calcula-
tions we find the expression for the effective diffusion coefficient for the Kadomtsev–
Pogutse regime:
Δ⊥
Deff ≈ Dm Ve . (2.371)
r0
A comparison between formulae (2.358) and (2.371) allows us to see the essential
difference between these approaches. We have considered the Rechester–Rosenbluth
and the Kadomtsev–Pogutse models in terms of the correlation scale Δ⊥ and the
collisional time τcoll . But, it is easy to relate these values to the Larmor radius
ρe of electron and the collisional frequency ve to describe plasma physics prob-
lems [67, 75].

2.11.4 The Quasi-Isotropic Stochastic Magnetic Field and Transport

In the framework of the Rechester–Rosenbluth approach, the expression for longitu-

dinal correlation length  
r0
LCOR ≈ LK ln (2.372)
l0
contains the parameters LK , l0 , and r0 , and their definition depends on a selected
model. In the case of magnetized plasma, the effective approximation is given by

D⊥
l 0 ≈ LK and r0 ≈ ρe , (2.373)
D

where ρe is the Larmor radius of the electron. However, if the tangled magnetic field
is described by the only spatial scale lB , then the estimate used by Rechester and
Rosenbluth becomes incorrect. Quasi-isotropic stochastic magnetic fields play an im-
portant role in astrophysical problems. Thus, in describing heat-conduction processes
in a stochastic magnetic field in clusters of galaxies, there are serious difficulties be-
cause the transport observed considerably exceeds theoretical estimates [12]. Chan-
dran and Cowley suggested an interesting modification of the Rechester–Rosenbluth
scaling, in which the characteristic spatial scale of nonuniformity of tangled mag-
netic field lB is simultaneously the parameter describing electron capture by mag-
netic traps formed by a significant nonuniformity of a magnetic field in a longitudi-
nal direction. Formally, in the conditions when lB ≤ λB , where λB is the mean free
path, the correlation characteristics of stochastic magnetic fields can be represented
122 O.G. Bakunin

Fig. 2.7. Magnetic force lines 1–2, 3–4, and particle trajectory 1–4

Fig. 2.8. Magnetic traps

in the form  
lB
LCOR ≈ lB ln , (2.374)
ρe
lB2
Dm ≈ ≈ lB . (2.375)
lB
The authors of [12] assumed that transverse decorrelation in electron motion arises
along the whole distance of order lB and, at the same time, the same scale charac-
terizes the sizes of the magnetic traps, which electrons leave after gaining additional
energy in collisions. The situation under analysis is described by the following hier-
archy of scales:
ρe  lB ≤ λB  LCOR . (2.376)
The expression for the effective diffusivity in a quasi-isotropic stochastic magnetic
field takes the form:
LCOR D lB D
Deff ≈ Dm ≈ Dm ≈ D ≈ . (2.377)
τ LCOR LCOR ln(lB /ρe )
 2 Correlations and Anomalous Transport Models 123

The physical problem considered by Chandran and Cowley implies the use of the
expression for the electron heat-conduction coefficient of Spitser–Harm χSp as D ;
taking into account the one-dimensional character of electron motion along force
lines,
L2 χSp
D ≈ ≈ . (2.378)
2τ 3
The estimate of the value ρe in the conditions corresponding to clusters of galaxies
yields
lB /ρe ≈ 103 or LCOR ≈ 30lB ; (2.379)
therefore, the suggested approach gives the estimate for the effective heat conduction
coefficient:
χeff ≈ Deff ≈ 10−2 χSp . (2.380)
From the point of view of the explanation of observed distributions, this estimate is
not correct but (as will be discussed below) the Chandran–Cowley approach needs
only insignificant modification. The monoscale model in [12] implies that the MHD
turbulence has an isotropic character. However, the phenomenological approach,
which takes into account the contribution of different turbulence scales in the
value lB , appears to be more adequate for describing galaxy-cluster cooling
flows [148].

2.11.5 Quasi-Linear Scaling for the Stochastic Instability Increment

The relative displacement of two nearby force lines during the process of their ran-
dom walks is given by
 z
B (1) (z) − B (2) (z)
r = r − r =
(1) (2)
dz . (2.381)

|B|
It is then easy to obtain the expression describing the divergence of force lines of the
stochastic field for small values of r2 − r1 [67]
∂ ∂b
(r2 − r1 ) = b(z, r2 ) − b(z, r1 ) ≈ (r2 − r1 ). (2.382)
∂z ∂r
Formal calculations yield the exponential dependence:

 z 
∂b
r2 (z) − r1 (z) ≈ r(z = 0) exp dz . (2.383)
0 ∂r

The increment of stochastic instability can be found by averaging this expression

with an assumption about the Gaussian character of a random value b, which makes
it possible to calculate an average as follows:
  
exp A = exp A2 /2 , (2.384)
124 O.G. Bakunin

and hence:
 
Δ⊥ (z) = r2 (z) − r1 (z)

 z z    
1   ∂b(z , r) ∂b(z , r)
= r|z=0 exp dz dz . (2.385)
2 0 0 ∂r ∂r
The integral expression in this formula is analogous to the expression for the quasi-
linear diffusion coefficient. Simple transformations yield the increment of stochastic
instability γz in the form
  
1 ∞ ∂b(0, 0) ∂b(z, 0)
γz = dz. (2.386)
2 −∞ ∂r ∂r
In terms of the dimensionless parameter Rm, this result takes the form

b02 LZ Dm 1 2
γz ≈ ≈ ≈ R . (2.387)
Δ2⊥ Δ2⊥ LZ m

In terms of the Kubo number, γ ∝ Ku2 . Naturally, the limits of applicability of this
estimate coincide with the limits of applicability of the quasi-linear approximation,

Rm ≈ b0 LZ /Δ⊥ < 1 or Ku < 1. (2.388)

It is necessary to note that the features of topology play an important role in esti-
mating the increment of stochastic instability. Thus, in the problem of magnetically
confined high-temperature plasma, shear effects are of great importance [1–5]. Using
simple calculations and introducing an additional characteristic scale

LS = dq/dx ≈ const (2.389)

(for the case of constant shear), it is possible to re-define the value γZ . Here, q is the
shear of the magnetic field. The modified equations take the following form:

B = B0 eZ + q(x)eY + B  (x, y, z), (2.390)

d r⊥ (x, y)  ⊥ , z) + q(x)eY .
= b(r (2.391)
dz
In the case of small displacements δx and δy, the analogue of (2.382) is the following
set of equations [67]:
d ∂bX ∂bX
δx = δx + δy, (2.392)
dz ∂x ∂y


d ∂bY 1 ∂bX
δy = + δx + δy. (2.393)
dz ∂x LS ∂y
The equation of evolution of the value
   
r = δx 2 ; δy 2 ; δxδy (2.394)
 2 Correlations and Anomalous Transport Models 125

is written in the matrix form:

dr
= Ŵ r , (2.395)
dz
where the eigenvalues of the matrix Ŵ define the increment of stochastic instability
for the case of constant shear:
1
γZ = , where LS < LZ . (2.396)
(LZ L2S )1/3

Unfortunately, this important result obtained by Krommes in [38, 39] does not estab-
lish the direct relationship between γZ and Rm which characterizes the dependence
on the magnetic perturbation amplitude.

2.12 Fractal Conceptions and Turbulence

The use of the fractal conceptions to obtain scaling laws is widely practiced. Man-
delbrot’s papers [56, 140] have played a very important role in this field. Richardson
[58] suggested the first nontrivial scaling for turbulent diffusion. But many years
after the publication of this paper Richardson proposed another nonconventional ex-
pression to describe the length of very “tortuous” curves [141]. This very formula
stimulated the very interesting Mandelbrot papers. Mandelbrot introduced the no-
tion of the fractal dimensionality that essentially extends the region of applicabil-
ity of scaling methods in the turbulence theory and anomalous transport models. In
this section, we will consider several problems of turbulence theory in terms of the
fractal dimensionality. This allows us to move to the consideration of percolation
mechanisms of anomalous transport.

2.12.1 Fractality and Transport

A large variety of nondiffusive transport mechanisms lead to the necessity of using

different scaling laws. The scaling method appears to be very fruitful obtaining the
relationships between the parameters, which characterize the transport, correlation,
and geometric properties of the model under consideration. One of the reasons for
such efficiency is the possibility to describe the geometric properties of different
objects by using scaling terminology.
Such an approach was developed by Mandelbrot [56, 140] and obtained wide
use due to its exceptional universality. The scaling representation for the basic geo-
metric properties of models was suggested. In this approach the expression for the
curve length is both the simplest and a very important result. From the formal stand-
point [56, 57] the length of the “very tortuous curve” (the fractal curve) L(δ) can be
rewritten in the form
1
L ≈ δN (δ), N (δ) ∝ d . (2.397)
δ F
126 O.G. Bakunin

In this fractal approach the full length L is approximated by small segments of size δ,
N(δ) is the number of these segments, which are necessary for such an approxima-
tion, and dF is the fractal dimensionality of the curve [56, 57]. In the framework of
the conventional representation of the geometry of curves, we have to use the value
dF = d = 1. However, in this case the drawbacks of the conventional method of
length measurement by a “yardstick” (ruler) remain.
Mandelbrot considered the problem of the measurement of a tortuous seacoast
length in which the increase in measurement accuracy (the decrease in the value δ)
leads to a growth in the value N (δ) (dF > 1). From the formal standpoint this
approach yields
L(δ) ≈ δN(δ)|δ→0 → ∞. (2.398)
This means that such a fractal line embraces “almost” the full plane. There are ad-
vantages and disadvantages of such a representation. One of the advantages is the
possibility to describe the longest and most complex lines. Mandelbrot called them
the fractal lines.
A similar situation is for the fractal surface area A:
δ2
A ≈ δ 2 N (δ) ≈ . (2.399)
δ dF
Here, δ 2 is the small area of size of A for measurement and dF > 2 corresponds to
the fractal nature of the surface.
In general, we can obtain the expression for the fractal region in the form

Wd ≈ δ d N (δ) ≈ δ d−dF . (2.400)

Here, we are dealing with the fractal cases dF > d. It is possible to consider opposite
cases when dF < d. This situation can characterize the presence of a large number
of voids in the cluster. More detailed information can be found in many books and
review articles on fractal geometry and fractal models [56, 57, 140].
The simplest model, which permits us to analyze the fractal properties of trans-
port processes, is d-dimensional random walks:

√ Δ2cor
R≈ 2dDt, D≈ . (2.401)
2dτ
Here, Δcor is the correlation length and τ is the correlation time. For this case, it is
easy to obtain an expression that includes the fractal dimensionality of the Brownian
trajectory:
t 2dDt
L ≈ N Δcor ≈ Δcor ≈ 2 Δcor . (2.402)
τ Δcor
Hence, one can obtain a number of “steps” in the scaling form
1
N∝ , where dF = 2. (2.403)
Δ2cor
 2 Correlations and Anomalous Transport Models 127

Note that the value dF is not dependent on d. Here, we assume that Δcor is the small
quantity Δcor ≈ δ. This corresponds to the definition of the fractal curve (2.397).
However, it is possible to describe the fractal properties on the basis of using “large”
parameters R ∝ 1/δ, which corresponds to the “cluster” terminology. For example
the expression
N ∝ R dF (2.404)
is the equivalent definition of the number of molecules which form the fractal cluster.
Here, R is the radius of the region occupied with the cluster. In this case, we can
rewrite the Brownian scaling law in the following form:

t R2
N≈ ≈ ∝ R2. (2.405)
τ 2dDτ
Hence, dF = 2 and this result exactly corresponds to (2.405).
From the “fractal” point of view, the scaling laws describing anomalous trans-
port in terms of the Hurst exponent R ∝ t H can be interpreted analogously. Another
definition of the transport exponent, which is usually called the internal dimension-
ality [17, 57],
1
t ∝ N ∝ R dW , dW = , (2.406)
H
is often used.
We can relate the Hurst exponent H and the internal dimensionality of the ex-
ponent dW to the properties of the fractal cluster, where transport occurs. A simple
estimate, which implies that fractal diffusion is a slower process than the conven-
tional Brownian diffusion, is widely encountered:

R2 1
≈D≈ θ. (2.407)
t R
Then, simple calculations yield

R ∝ t 1/(2+θ) , dW = 2 + θ. (2.408)

This example demonstrates the method for obtaining the relationships between the
exponents that describe different physical properties of the system. In the case of
(2.407) the value dW reflects the character of transport processes and θ characterizes
the degree of complexity of the fractal structure.
The region of applicability of the fractal ideas for transport models is very wide.
Not only the walking particle trajectory, but also the great number of return moments
at the initial point, the diffusive front, etc. appear to be fractal values.

2.12.2 The Richardson Law and Fractality

There is a good example of the relation between fractal concepts and turbulence.
Even simple observation shows the fractal nature of the cloud boundary. In this con-
nection, great interest has arisen in the experimental measurement of a perimeter of
a rainfall P and a cloud area A, determined from radar and satellite data.
128 O.G. Bakunin

Lovejoy obtained the simple scaling that relates the fractal perimeter P of the
two-dimensional projections of the cloud region to its area A:

P 1/dF ∝ A1/2 . (2.409)

From the fractal point of view, this expression is not surprising. Thus, Mandelbrot
considered the formula for the length of the fractal curve that bounds the nonfractal
area:  dF  1−dF
λ δ
L ≈ δN(δ) ≈ δ ≈λ . (2.410)
δ λ
√
Here, λ is the characteristic spatial scale. If we suppose λ ∝ A, then the expression
for L takes the form
L ≈ δ 1−dF AdF /2 . (2.411)
Now it is easy to obtain the Mandelbrot relationship between the perimeter and the
area
P 1/dF ≈ L1/dF ≈ δ 1−dF A1/2 , (2.412)
where the value δ is the parameter. This exactly corresponds to the Lovejoy re-
sult (2.409).
The correctness of this result in a wide spectrum of parameters has led Hentschel
and Procaccia [78] to the idea of the relation between the fractal dimensionality
of the cloud perimeter and the universal properties of the Kolmogorov model of
isotropic turbulence. There is a simple method to obtain the relationship between the
fractal dimensionality of the perimeter and the exponent that describes the turbulence
spectrum. It based on the calculation of the effective rate of increase of the area of
the cloud cross section due to the turbulent pulsation of velocity:
dA
≈ lk V (lk )N (lk ). (2.413)
dt
Here, lk ≈ 1/k is the characteristic spatial scale, V (lk ) is the characteristic velocity
scale, and N(lk ) is the number of sections that approximate the fractal perimeter
of the cloud. If we use the scaling expressions for the spectrum of energy with an
arbitrary exponent ς and for the number of sites,

Vk2
E(k) ∝ ∝ k −ς , N (lk ) ∝ lkdF , (2.414)
k
then the substitution of (2.414) into (2.413) yields
dA 1−d −(1−ς)/2
≈ lk F . (2.415)
dt
However, it is obvious that the rate of increase of the area of the cloud cross-section
is independent of lk . Therefore, we obtain the following relationship:
1+ς
dF = . (2.416)
2
 2 Correlations and Anomalous Transport Models 129

Then in the Kolmogorov case with ς = 5/3, we obtain dF = 4/3. This result is in a
good agreement with the value defined by Lovejoy, dF = 1.35.
Even this simple estimate demonstrates the efficiency of using fractal ideas.
A more rigorous analysis [78] has led to the modification of the Richardson scal-
ing:  
Y (t)2 ≈ t 3+q , (2.417)
where the range for the exponent q is 0.45 > q > 0.15. Analogous results were later
obtained in the framework of using the Levy–Khintchine approximations [140].
Note that the problem of calculating the fractal object perimeter is, in general, a
rather complex problem. Moreover, in the problems of the description of turbulent
diffusion based on the percolation model, the exponent that characterizes the hull of
a fractal cluster (external perimeter) is a key parameter of the theory.

2.12.3 Intermittency and the Kolmogorov Law

The conception of the fractal character of geometrical values yields the possibility of
the essential modification of many results. Moreover, often the fractal representation
more adequately mirrors the essence of the problem under consideration. A good
example is the correction of the universal Kolmogorov spectrum of energy [63]. It
was clear at an early stage of development of Kolmogorov and Obuchov’s scaling
ideas that it is impossible to explain effects connected with intermittency [27, 30,
143] in the framework of the mono-parametric model. Many researches tried to cor-
rect “the law of 5/3” [63]. One of the most elegant models is the model of fractal
representation of regions of turbulence energy dissipation. Thus, in the classical for-
mulation the Kolmogorov idea can be represented by the dimensional estimates for
the dissipation rate εk ,

Vk2 V3
const = εk ≈ ≈ Vk3 k ≈ l , (2.418)
τ l
and the estimate of the energy spectrum in the space of the wave numbers k ≈ 1/ lk ,
 2/3
V2 εk 1
E(k) ∝ k ≈ . (2.419)
k k k

Mandelbrot [76] and then Fricsh, Sulem and Nelkin [77] have renormalized these
expressions using the fractal representation of energy dissipation regions. The frac-
tion of the volume corresponding to “one dissipation center” can be represented in
the form
V0 ld
Q≈ ≈ dk ≈ lkd−dF . (2.420)
N (lk ) l F
k
Here, N (lk ) is the number of “dissipation centers” in the region of the size lk and
Vl ≈ lkd is the volume of this region; d is the dimensionality of Euclidean space; and
dF is the fractal dimensionality of the “cluster” consisting of “dissipation centers”.
130 O.G. Bakunin

The expression for εk and E(k) can then be rewritten in the renormalized form:

Vl3
const = ε ≈ Vk3 kQ ≈ Q(l), (2.421)
l
Vk2
EF (k) ∝ . (2.422)
k
Then, upon performing calculations we arrive at the expression
1 1
EF (k) ∝ EK (k)k −(d−dF )/3 ≈ . (2.423)
k 5/3 k (d−dF )/3
Experiments are satisfactorily described by the value dF ≈ 2.8 [30]. However, there
are many papers that discuss dissipation on an eddy surfaces with dF = 2 or dissi-
pation on eddy filament, which can be considered in terms of self-avoiding random
walks (2.110) with dF = (2 + d)/3.
Thus, for example, in the case dF = 2 one obtains the energy spectrum
1
E(k) ≈ (2.424)
k2
that corresponds to the Taunsend suggestions [27], which can be interpreted in terms
of the constant spectrum of enstrophy (rot V )2 . Then the condition for the enstrophy
spectrum
(Vk k)2
Ew (k) ∝ ≈ const, (2.425)
k
in combination with the spectrum energy from definition (2.419), leads to the spec-
trum (2.424).
It is obvious that the Kolmogorov idea partially loses its initial universality after
we introduce the new parameter dF . However, at the same time, such a correction
essentially increases the possibilities of agreement between theory and experiment.

2.13 Percolation and Scalings

In this section, we discuss the possibility of explaining the anomalous superdiffusion
transport in the framework of the percolation threshold. The percolation approach
looks very attractive because it gives a simple and, at the same time, universal model
of behavior related to both the strong correlation effects and convective transport.
On the other hand, for two-dimensional systems it is possible to use computer sim-
ulation to check theoretical results. The scaling arguments for both steady and time-
dependent flows are presented.

2.13.1 Continuum Percolation and Transport

Kadomtsev and Pogutse [67] suggested a new method of analysis of the magnetic
transport effects under conditions where transverse correlation effects play a signif-
 2 Correlations and Anomalous Transport Models 131

Fig. 2.9. Random steady flow and fractal streamline

icant role, b0 LZ ≥ Δ⊥ (2.243). They used the percolation approach [49, 57] and
formulated the criterion for its applicability in terms of the dimensionless parameter
that characterizes the ratio of longitudinal and transverse correlation effects Rm  1.
The percolation methods are based on the ideas of “long-range” correlations and
could correspond to the two-dimensional anomalous transport pattern in turbulent
plasmas in the presence of a strong magnetic field. A physically clear presentation
of fundamental ideas in the percolation theory and the fractal concept can be found
in [49, 57]. In the subsequent discussion, we assume that the reader is familiar with
the basic definitions in these papers and consider the problems related to anomalous
transport. In the context of this approach, the streamlines Ψ = Ψ (x, y) are consid-
ered as coastlines in a hilly landscape flooded by water. It is expected that there is a
sharp transition from separated lakes on a boundless land to individual islands in an
infinite ocean. The percolation theory requires the existence of at least one coastline
of infinite length.
Kadomtsev and Pogutse [67] related the anomalous character of diffusion for
Rm  1 to the fractal character of the behavior of the streamlines of the two-
dimensional flow near the percolation threshold (see Fig. 2.9). They used the scaling
law for the length of a percolation line of flow:
1
L(ε) ∝ . (2.426)
ε 2.4
Here, ε is a small dimensionless quantity which characterizes the degree of deviation
of the system from the critical state (the percolation threshold), ε ≈ h/λV0 , where h
is the value of the streamline function Ψ = Ψ (x, y) near the percolation threshold,
λ is the characteristic scale, and V0 is the characteristic velocity of the flow. The
expression for L(ε) corresponds exactly to the fractal representation of the curve
length. There are advantages and disadvantages to such a representation. One of the
advantages is the possibility of describing the longest and most tortuous curves and
long range correlations [17, 20, 49, 57].
132 O.G. Bakunin

To describe effects related to the considerable increase in transport coefficients, it

is not sufficient to consider the fractal presentation of a streamline alone. Moreover,
the fractal character of the curve sometime leads to slower diffusion (subdiffusion).
Therefore, it is necessary to introduce another important value. In percolation theory
the correlation length a(ε) is the main magnitude characterizing spatial scales of the
system, which is located near the percolation threshold ε → 0:
λ
a(ε) = . (2.427)
|ε|ν
Here, λ is the geometric characteristic scale. Thus, the idea of long-range correla-
tions was realized in the framework of the percolation approach. However, there is
a problem, since the diffusion coefficient is directly related to the expression for
the correlation length: D ≈ Δ2cor /τ . Here, τ is the correlation time. In the case
under consideration, estimates yield Δcor ≈ a(ε)ε→0 → ∞. For this reason, per-
haps, Kadomtsev and Pogutse [67] based their consideration on the “diffusion renor-
malization” of the quasi-linear equations. However, in this approach the percolation
character of correlation effects was lost. It is not surprising that in the framework of
the classical diffusion equations we cannot use the percolation exponents ν, dF .
To develop the percolation approach, it is necessary to take into account the fact
that the percolation cluster occupies only a small fraction of the space. Therefore,
the value
Deff (ε) ≈ DC P∞ (ε) (2.428)
can be the estimate of the effective diffusion coefficient. Here, DC is the diffusion
coefficient which corresponds to transport on the percolation cluster and the value
P∞ (ε) defines the fraction of the space that is occupied by the percolation cluster. In
percolation theory, the scaling representations for Deff (ε) and P∞ (ε) are used:

DC (a) ∝ a −θ , Deff (ε) ∝ ε μ , P∞ (ε) ∝ ε β . (2.429)

On substituting (2.429) and (2.427) into expression (2.428) we obtain the relation-
ship between the different percolation exponents: μ = β − νθ . In the subsequent
consideration, we will use these results to analyze two-dimensional random flows.
Therefore, it is important to note that there is an exactly calculated value ν for two-
dimensional percolation problems: ν = 4/3 [17, 144–146]. To determine the trans-
port characteristics it is necessary to find the values θ and β, which depend on the
physical model.

2.13.2 Renormalization and Percolation

To solve real physical problems, the expression ε → 0 looks too abstract. Here, we
consider a simple method which permits us to obtain effective estimates of values
described by scaling laws (2.427), (2.428) by using the finite value of the percolation
parameter ε∗ instead of ε → 0. The correlation length is one of the most important
values describing transport. However, in a system of finite size L0 we cannot consider
 2 Correlations and Anomalous Transport Models 133

Fig. 2.10. Renormalization of the small parameter ε = p − pC for finite size samples

the infinite value

a(ε)|ε→0 → ∞. (2.430)
Here, it is relevant to introduce a new small “renormalization” parameter ε∗ [79] as
the value that gives the condition

a(ε∗ ) ≈ L0 . (2.431)

The simplest calculations yield

 1/ν
λ
ε∗ ≈ . (2.432)
L0
This result can be interpreted in the framework of percolation experiments with finite
size samples. Under these conditions, the percolation threshold arises when the value
of ε∗ slightly differs from zero and is situated in some ε interval. The estimate
obtained for ε∗ can be considered as the characteristic width of this diapason (see
Fig. 2.10)
ε ≈ ε∗ . (2.433)
Actually, we are dealing with the small parameter
λ
ε0 ≈  1, (2.434)
L0
which describes the real physical system with the characteristic scales L0 and λ. On
“renormalization”, we obtain a new percolation parameter
1/ν
ε ≈ ε∗ ≈ ε0 . (2.435)

It is natural that the value ε decreases if the system size L0 increases. Here, the
value of ε decreases if the system size L0 increases. Now we can calculate the
134 O.G. Bakunin

diffusion coefficient that is based on the estimate of the finite correlation length a:

a 2 (ε)
Deff ∝ . (2.436)
τ
Usually, the main problem of the percolation theory is to obtain interrelations be-
tween the percolation exponents. However, in the percolation models of turbulent
diffusion the key problem is to determine a small parameter ε0 and to find an ade-
quate renormalization condition for ε∗ . The next sections of this paper will consider
in detail several specific models of turbulent diffusion from this standpoint.

2.13.3 Graded Percolation

An important typical example of the small renormalization parameter is the consid-

eration of percolation in graded type systems. Trugman was the first to analyze this
situation [147]. The graded character of the model corresponds to the assumption
that the system is subject to a small external influence, which does not in general de-
stroy the percolation character of the system’s behavior, but can essentially change
its properties.
First, we will consider this method from the formal point of view. Let us intro-
duce a parameter ε0 characterizing the smallness of an influence. In contrast to the
renormalization, which uses the dependence of the percolation parameter on the sys-
tem size ε ≈ ε(L0 ) ≈ ε∗ , here we will deal with the spatial dependence, which
is related to the graded character of the problem ε ≈ ε(x). From the dimensional
consideration we can obtain the expression that characterizes the uncertainty of the
choice of the small parameter for these conditions, ε∗ ≈ ε ≈ ε  (x)a(ε∗ ). Then,
simple calculations yield
1 λ
a(ε∗ ) ≈ λ ≈  . (2.437)
v
ε∗ [ε (x)a(ε∗ )]v

After the dimensional estimate in the form ε  (x) ≈ ε0 /λ, we obtain the Trugman
renormalization condition for the correlation scale:
1 λ
a(ε∗ ) ≈ λ v
≈ v/(1+v) . (2.438)
ε∗ ε0

Here, the value

1/(1+v)
ε∗ = ε0  ε0 (2.439)
is the new renormalized small parameter. Note that the direct use of the value ε0 as
the parameter in the percolation dependences is not correct, since it characterizes the
destructive influence of a superimposed perturbation and not the degree of departure
of the system from the percolation threshold.
This method looks quite formal, but renormalization (2.439) was repeatedly used
to obtain information about the critical exponents that describe the hull of a perco-
lation cluster, to analyze transport in a system with shear flows, and to consider the
 2 Correlations and Anomalous Transport Models 135

models of multiscale percolation. In specific problems, which will be considered be-

low, a more detailed interpretation of a similar transformation will be given on the
basis of taking into account the physical parameters that influence the topology of
the flow under analysis. However, the first step is to obtain these physical values and
introduce the small parameter that characterizes a percolation threshold.
Using the graded percolation method allows one to describe the structure of the
diffusion front, which has a fractal nature [49, 57]. Another aspect of this problem
also exists. We can describe the external perimeter of the fractal cluster (the hull)
in terms of the diffusion front. From the diffusion standpoint the √ important value is
the average distance between the source and the “front” lF ∝ Dt. In the context
of the problem the value lF is the characteristic spatial scale. Such a probabilistic
formulation of the problem of the particle reaching the diffusion front is very fruitful,
since it is possible to interpret the problem in terms of the graded percolation (2.437),
where ε = ε(x). Then we can choose the value ε0 ≈ λ/ lF as the initial parameter of
the problem. In accordance with expression (2.439) we obtain
 (v+1)/v
a
lF (a) ≈ λ . (2.440)
λ

Here, a is the correlation size that characterizes the structure of the diffusive front.
Numerical and theoretical research on the external perimeter of the percolation clus-
ter has arrived at the conclusion that in the two-dimensional case the hull L(a) is
equivalent to the value lF (a) [17, 49, 57]. Hence, the fractal dimensionality of the
hull Dh is defined by the formula
1
Dh = 1 + = 7/4. (2.441)
ν
Indeed, rigorous mathematical consideration confirms this result [144–146].
Note that in the two-dimensional percolation models of turbulent diffusion [17,
148], the hull is considered as a percolation streamline. The characteristic scale of
the flow velocity V0 is usually a known parameter; then, the dimensional estimates
of many values can be obtained by way of combinations of the values V0 and L.
From this standpoint the relationship between Dh and ν allows all characteristics
of the system to be described in terms of the universal exponent ν that reflects the
correlation character of the percolation models.

2.14 Percolation and Turbulent Transport Scalings

The use of the “renormalization” method makes it possible to apply the percolation
approach to the description of turbulent transport in two-dimensional random flows.
The percolation scalings obtained by this method of the turbulent diffusion coeffi-
cient differ significantly from the quasi-linear one. This considerably increases the
range of problems that can be considered in the framework of the scaling approach.
136 O.G. Bakunin

2.14.1 Random Steady Flows and Seed Diffusivity

Isichenko et al. [82] realized the potentiality of the percolation approach. They con-
sidered a two-dimensional zero-average-velocity steady flow specified by the
bounded “common position” stream function Ψ (x, y). They implied an isotropic-
on-average oscillating function with a quasi-random location of saddle points along
its height. The following scales were selected:

Ψ
Ψ0 ≈ λV0 , λ ≈ . (2.442)
∇Ψ
The authors of [82] used formula (2.428) for effective diffusion and ideas of the
convective nature of the flow along the percolation streamline,
L(ε)Δ(ε)
Deff (ε) ≈ DC , (2.443)
a2
a2
DC (ε) ≈ . (2.444)
τ
Here, L(ε) is the length of the percolation streamline, a is the spatial correlation
scale [17], τ is the correlation time, and DC is the correlation contribution to the
effective diffusion coefficient. The value h = ελV0 is the percolation scale of the
stream function near a threshold. Effects of “long range correlations” enter into the
expression for the diffusion coefficient precisely through a(ε). Expression (2.443)
is similar to the formula for effective diffusivity in the convective cell system from
[80, 81], but here we are dealing with the dependence Δ = Δ(ε). Moreover, in
line with ideas in the literature on percolation theory, the authors of [82] suggested
“a renormalization”, i.e., a method of calculating the universal value of the small
parameter ε in their percolation diffusion theory. They identified the small “width”
of a percolation streamline with the small parameter of the percolation theory (see
Fig. 2.9):
Δ(ε) = λε. (2.445)
This renormalization has become the main step in [82] and has begun to be actively
used to solve other problems [17]. Representation (2.330) for the value Δ will be
used in this formula. However, in the percolation case, the correlation effects should
be taken into account by using L(ε) instead of the geometric scale λ,

D0 L(ε)
Δ(ε) ≈ . (2.446)
V0

This approach creates a dependence of Δ on the parameter ε. Now, we obtain the

equation for the determination of the “universal” value of ε∗ = h∗ /λV0 as a function
of the flow parameters D0 , V0 , λ,

D0 L(ε)
= ελ. (2.447)
V0
 2 Correlations and Anomalous Transport Models 137

The specific calculations can be completed by using the rigorous scaling results of
percolation theory [17, 49, 57] obtained for the correlation scale a and the length of
the fractal streamline L as functions of ε for the two-dimensional case,
 Dh
a 1
a(ε) = λ|ε|−ν , L(ε) = λ , ν = 4/3, Dh = 1 + . (2.448)
λ ν
The functional form of these dependences reflects the fractal and percolation be-
havior of the streamlines introduced in [67]. The solution of (2.447) in terms of the
Peclet number leads to the expressions

h∗ = λV0 ε∗ = λV0 Pe−3/13 ∝ V0

10/13
, (2.449)
 1/(3+ν)
1
Deff ≈ V0 Δ(ε∗ ) ≈ V0 λ = D0 Pe10/13 . (2.450)
Pe
This corresponds to the value μ = 1 [see (2.429)]. From the point of view of the
renormalization of the initial small parameter ε0 = 1/Pe, the expression for the
percolation small parameter can be obtained:

ε∗ = (ε0 )1/(3+ν)  ε0 . (2.451)

Some “arbitrariness” in expression (2.445) in the selection of the value λε, rather
than λε 2 or λε 3 , can be interpreted as a desire to have a universal small parameter,
which is analogous to a single characteristic scale—the correlation length in the the-
ory of phase transitions. Here, it should be particularly emphasized that the length of
the fractal percolation line is not infinitely large because the small parameter ε∗ does
not tend to zero but has a finite value ε∗ = h∗ /λV0 for all types of flows with the
characteristics D0 , V0 , λ. Therein lies the universality of formula (2.445).

2.14.2 The Spatial Hierarchy of Scales and Stochastic Instability

Note that the simplicity of the percolation estimate of turbulent transport is elusive. It
will suffice to recall in this connection the “hierarchy” of scales used by the authors
of [82] for their analysis,
 Dh
a a λ h∗
λ ≈L≈ a≈ ν λ ≈ Δ ≈ λε∗ . (2.452)
λ ε∗ ε∗ V0
7/3 4/3
Here, the values L(ε∗ ) ≈ λ/ε∗ and a(ε∗ ) = λ/ε∗ , which are responsible for
the percolation behavior, are not infinitely large, since ε∗ has a precise value. The
fraction of the volume occupied by the percolation streamlines is now estimated as
L(ε)Δ(ε) 4/3 1
P∞ = ≈ ε∗ ∝ . (2.453)
a(ε)2 a
This corresponds to the exponent β = ν [see (2.429)].
138 O.G. Bakunin

Besides the scaling laws (2.450), we can also obtain some additional information
that is useful for the subsequent analysis. Let us note that in terms of the Peclet num-
ber the percolation regime is intermediate between the regime of convective cells
from [80, 81] and the purely convective regime, where Deff ≈ λV0 . We can estimate
the range of the percolation scaling applicability in term of spatial scales. It is neces-
sary to take into account the finite size L0 of a real system. By analogy with (2.431)
we can consider the estimate

a(ε∗ ) = λPeν/(ν+3) ≤ L0 . (2.454)

Then, calculations yield the inequality for the Peclet number in the form
 (ν+3)/ν
L
1 < Pe < . (2.455)
λ
There is also a correlation meaning of percolation scaling. In the steady case the
correlation time τ is described by the scaling law
L(ε) λ 1 λ
τ≈ ≈  . (2.456)
V0 V0 ε∗7/3 V0

On the other hand, the correlation scale a can be represented in the form

Δ2
a ≈ ε∗ L ≈ ε∗ V0 ≈ ε∗ V0 τD  V0 τD . (2.457)
D
From the formal point of view one can expect that there exists an additional spatial
scale lS ,
L(ε)lS
≈ 1. (2.458)
λ2
Indeed, the scale
λ2 ε∗
lS (ε∗ ) ≈ ≈ λε∗ν+1 ≈ ε∗ν Δ(ε∗ ) (2.459)
a(ε∗ )
plays an important role in the analysis of the reconnection process [83]. This problem
will be discussed in the next section. Now the entire spatial hierarchy of scales is
given by
 Dh
a a λ
λ ≈L≈  a ≈ ν  λ  Δ ≈ λε∗  lS ≈ λε∗ν+1 ≈ ε∗ν Δ. (2.460)
λ ε∗ ε∗
This gives wide possibilities to obtain new scalings in the framework of the percola-
tion method.

2.14.3 Low Frequency Regimes

It is well known than the temporal reconstruction of a flow “topology” leads to con-
siderable change in the transport in comparison with the quasi-linear one. The best
 2 Correlations and Anomalous Transport Models 139

illustration of this fact is the dependence of the effective diffusion coefficient on the
Kubo number [3, 7, 8]. Following the ideas of [82], Grusinov, Isichenko, and Kalda
[83] considered the percolation limit of the turbulent diffusion of a passive scalar
in a time-dependent, incompressible, two-dimensional flow. In estimating the time
it takes the flow pattern to change completely as T0 ≈ 1/ω, the authors of [83]
considered the low frequency case:
V0
ω or λ  V0 T0 . (2.461)
λ
In this formulation of the problem, the lifetime of the individual percolation stream-
line τ is the main parameter. The standard expression can be used for the diffusion
coefficient,
a2
DC (ε) ≈ . (2.462)
τ
In the context of this problem the expression
1
τ ≈ε ≈ εT0 (2.463)
ω
was used, where ε is the small parameter of the problem, which is analogous to
that of [82]. In the time-dependent flow under consideration, one would expect a
universal result, provided it is possible to calculate a specific “universal” value ε∗ .
For this purpose, the authors of [83] suggested a simple expression accounting for
the convective nature of motion along the percolation streamline during the lifetime
of this streamline:
1 L(ε)
τ ≈ε = . (2.464)
ω V0
We see the analogy between this formula and formula (2.445). Using the scaling
laws from percolation theory yields a(ε) = λε −ν , L(ε) = λ(a/λ)Dh , ν = 4/3,
and Dh = 1 + 1/ν. Now, we easily obtain ε∗ = h∗ /λV0 as a function of the flow
parameters ω, V0 , λ,
 1/(2+ν)
λω
ε∗ = = Ku−3/10 ∝ ω3/10 . (2.465)
V0

Here, it is relevant to introduce the Kubo number Ku = V0 /λω instead of the Peclet
number, which was used for the analysis of a steady flow. From the standpoint of
the renormalization of the initial small parameter ε0 ≈ 1/Ku we obtain the expres-
sion
ε∗ = (ε0 )1/(2+ν)  ε0 . (2.466)
Upon direct substitution of the expression for ε∗ into (2.462) we arrive at the for-
mula
a(ε∗ )2
DC (ε∗ ) ≈ . (2.467)
τ (ε∗ )
140 O.G. Bakunin

Note that the dependence on T0 appears quite odd:

1/10 1
D C ∝ T0 ≈ . (2.468)
ω1/10
The slow “restructuring” of the flow is unlikely to result in a significant growth in
turbulent diffusion. The reason lies in the fact that we have not taken into account
the fraction P∞ of percolation streamlines in the total flow,
L(ε)Δ(ε)
P∞ = . (2.469)
a(ε)2
It is now evident that we need additional information on the value of Δ(ε), despite
the fact that we have calculated ε∗ . Unlike [82], we are forced to make an additional
assumption here other than expression (2.464). The authors of [83] assumed Δ(ε),
similarly to [82]. In fact, use is made of formulae (2.445) and (2.453):

4/3 λ
Δ ≈ ε∗ λ, P∞ ≈ ε∗ ≈ . (2.470)
a(ε∗ )
Calculations now lead to the final expression for Deff ,
 1/(ν+2)
1 7/10 3/10
Deff = DC (ε∗ )P∞ (ε∗ ) = λV0 ∝ V0 ω . (2.471)
Ku

The formula obtained reflects the universal character of percolation diffusion in

the time-dependent flows, since the value ε∗ depends only on the flow parameters
ω, V0 , λ.

2.15 The Temporal Hierarchy of Scales and Correlations

The consideration of the temporal hierarchy of scales of a percolation model allows
one to establish the bounds of applicability of the results obtained in the previous
section. Besides that, we will consider the method to obtain percolation transport
estimates that is based on the correlation scales balance, which makes it possible
both to reproduce known results and to obtain new scalings.

2.15.1 The Spatial and Temporal Hierarchy of Scales

Assuming small percolation parameters implies that ε∗ < 1 and, at the same time, it
is important to take into account the finite size of the system a(ε∗ ) ≤ L0 . From this
point of view we can find the inequality for the Kubo number, which describes the
time-dependent percolation,
 (ν+2)/ν
L
1 < Ku < . (2.472)
λ
 2 Correlations and Anomalous Transport Models 141

In the context of the definition of the Kubo number we noted that in low-frequency
regimes the real correlation scale Δcor is less than the frequency path lω = V0 /ω.
Indeed, the scaling (2.464) reflects the correlation meaning of the problem since

Δcor ≈ a ≈ ε∗ L ≈ ε∗2 lω  lω . (2.473)

Note that the condition a ≈ ε∗ lω leads to the renormalization τB ≈ T0 ≈ 1/ω.

On the other hand, additional estimates should be made of the effect of the diffu-
sion escape of particles from the streamlines. In fact, it is necessary to calculate the
time of the escape of particles from the streamlines τD . The formulae of [83] were
obtained in the framework of the condition τ < τD . To receive estimates, we make
use of the diffusion coefficient of streamlines Dψ and relate it to the coefficient of
“seed” diffusion D0 ,
h2
τD ≈ , Dψ = V02 D0 . (2.474)
Dψ
The applicability condition for the results [83] takes the form

h h2
< 2 . (2.475)
λV0 ω V0 D0

This is in fact a limitation on the value of the “seed” diffusion coefficient D0 ,

λ
D0 < hω(h) < hν+3 . (2.476)
V0
To conclude the discussion of this issue, we give the hierarchy of characteristic times
in the problem on percolation in a time-dependent flow,
 
1 h 1+ν h h2 1
= τS  τ ≈  2
≈ τ D  ≈ T0 . (2.477)
ω λV0 ωλV0 D0 V0 ω

Here, τS = lS /V0 is the characteristic time that describes the stochastic instability of
streamlines [83]. This means that one deals with the temporal interval τS  τ < τD .
In a similar manner to the above-considered hierarchy of scales in steady perco-
lation (2.460), this set of characteristic times (2.477) allows us to distinguish the
regimes of flow, where the “long-range” correlation effects become the principal
ones. The interpretation of renormalization conditions in terms of the characteristic
time is a very important aspect of the problem. For example, the condition for the
steady case (2.445) can be represented in the form of an equilibrium of the ballistic
and diffusive times:
L(ε∗ ) Δ2 (ε∗ )
τ≈ ≈ . (2.478)
V0 D0
It is easy to note that there are several characteristic times that do not play essential
roles at this stage:
h h
  τ. (2.479)
ωL(h)V0 ωa(h)V0
142 O.G. Bakunin

2.15.2 The Isichenko Intermediate Regime

Isichenko [17] considered the intermediate regime with the motion of particles along
the percolation streamline during the times t < τB ≈ L/V0 , which is analogous
to the fractal diffusion (this case differs significantly from the percolation model
of turbulent diffusion with t ≥ τB ). The author of [17] used the estimate of the
correlation scale in the form
   
L(t) 1/Dh V0 t 1/Dh
aI (t) ≈ λ ≈λ . (2.480)
λ λ
Here, the supposition was made that the test particle path at this stage is approxi-
mately ballistic, L(t) ≈ V0 t. To describe the initial stage of percolation t < τB ≈
L/V0 in the two-dimensional random flow the authors of [17] offered the expression
 1/Dh  
L V0 t 1/Dh
R 2 ≈ Deff t ≈ a 2 P∞ ≈ λa ≈ λ2 ≈ λ2 . (2.481)
λ λ

Here Dh = 1 + ν1 and ν = 4/3, and R is the mean free path of the particle. This
leads to the new subdiffusive regime

R ∝ t 1/2Dh = t 2/7 (2.482)

with the Hurst exponent

1
H = = 2/7. (2.483)
2Dh
This corresponds to the subdiffusion character and reflects the initial period when
the particle moved along the fractal streamline.
The model of the evolution of correlation scale aI (t) suggested by Isichenko can
be used to interpret and to analyze other percolation regimes [149]. Thus, simultane-
ously with the increase in the correlation scale aI (t) ≈ (L(t)/λ)1/Dh , it is necessary
to take into account the increase in the stochastic layer width Δ = Δ(t) which, in
the framework of percolation models of turbulent diffusion, is closely related to the
value of the small parameter ε∗ ≈ Δ/λ, and hence to the definition of the correla-
tion scale a ≈ λ/ε ν . Easy calculations allow one to obtain the expression describing
the increase in the correlation scale aD (t) due to the increase in the stochastic layer
width
λ
aD (t) ≈ . (2.484)
(Δ(t)/λ)ν
In the framework of mean field theory the consideration of the balance between aD (t)
and aI (t) makes it possible to find the estimate of the characteristic time t0 , which
has to be used to define the effective diffusion coefficient Deff . Thus, in the case
of the diffusive character of increasing the stochastic layer width Δ2 ≈ D0 t, one
obtains  
V0 t0 1/Dh λ
λ ≈ √ . (2.485)
λ ( D0 t0 /λ)ν
 2 Correlations and Anomalous Transport Models 143

Calculations lead to the estimate

   
λ2 1 1/(γ +3) λ2 D0 1/(γ +3)
t0 ≈ = , (2.486)
D0 Pe D0 λV0

which after substitution into the formula for the stochastic layer width yields
 1/(γ +3)
1 1
Δ ≈ (D0 t0 ) 2 ≈ λ . (2.487)
Pe

It is easy to see that this expression coincides exactly with the expression for the
steady case (2.449) and the corresponding estimate of the effective diffusion coef-
ficient is given by the formula Deff ≈ V0 Δ. Naturally, other estimates can also be
used to describe the growth of the stochastic layer width. Thus, in dynamical system
theory the linear estimate Δ ∝ t is widely used. In the context of the description of
time-dependence effects it is easy to represent this expression in the form

Δ(t) = (λω)t, (2.488)

where ω is the characteristic frequency. Then the consideration of the correlation

scales balance in the form
 
V0 t0 1/Dh λ
λ ≈ (2.489)
λ (ωt0 )ν

allows the estimate of the characteristic time t0 to be obtained:

   
1 λω 1/(ν+2) 1 1 1/(ν+2)
t0 ≈ ≈ . (2.490)
ω V0 ω Ku
The expression for the stochastic layer width then takes the form corresponding to
the percolation model (2.465)
 1/(ν+2)
1
Δ = λε∗ = λ , (2.491)
Ku
and hence the estimate of the turbulent diffusion coefficient is given by expres-
sion (2.471). Note that the approach under consideration makes it possible to use
the correlation scales balance as the basis for constructing new percolation transport
models based on approximations Δ(t).

2.15.3 Dissipation and Percolation Transport

In this section, we consider the influence of weak dissipative effects on the perco-
lation layer width. The presence of even small dissipation can considerably change
the character of streamlines behavior [17, 149]. It is possible to introduce the rate of
energy dissipation εK by analogy with isotropic turbulence theory [27, 30], εK =
144 O.G. Bakunin

[m2 /c3 ] = const. Note that in the framework of such an approach even a simple
dimensional estimate of the characteristic time τK ≈ (εK k 2 )−1/3 together with the
Fourier representation of the Einstein–Smolukhowski nonlocal functional yields the
classical Richardson law [62] for the relative diffusivity: R 2 ∝ t 3 . Here, k is the
wave number.
In the percolation case under consideration, the weak dissipation leads to an in-
crease in the stochastic layer width. We can estimate this at the initial stage of the
percolation process by the linear dimensional expression
 1/3
εK
Δ ≈ V∗ t ≈ λ 2 t. (2.492)
λ
Here, V∗ ≈ δ/τK characterizes the rate of growth of the stochastic layer width and
δ is the spatial dissipative scale. Formally, the value δ can be considered as an in-
dependent parameter. However, in the framework of the hierarchy of spatial scales,
which corresponds to the percolation model
λ a
Δ ≈ λε  λ  a ≈ ν
L≈ , (2.493)
ε ε
the estimate δ ≈ Δ is fairly adequate for the weak dissipation since Δ is the least
among spatial scales. Then simple calculations yield
 1/5
εK
Δ≈λ 2 t 3/5 . (2.494)
λ
At the initial stage we deal with two competing processes: an increase in the correla-
tion scale a due to growth of the distance passed by the test particle, and a decrease
in the correlation scale due to dissipative effects. Indeed, in the case under consider-
ation the dissipation leads to an increase in the stochastic (percolation) layer width
Δ(t) and hence to the decrease of the correlation scale:
   ν
Δ(t) −ν λ
aD (t) ≈ λ ≈λ . (2.495)
λ λ(εK /λ2 )1/5 t 3/5
On the other hand, it is also necessary to take into account the increasing correlation
scale a due to the percolation character of streamlines
   
L(t) 1/Dh V0 t 1/Dh
aI (t) ≈ λ ≈λ . (2.496)
λ λ
Here, the supposition was made that the test particle path at this stage is approxi-
mately ballistic, L(t) ≈ V0 t.
In the framework of scaling arguments we consider the balance between these
processes as the balance of the correlation scales aI (t0 ) ≈ aD (t0 ) ≈ a,
 ν  
λ V0 t0 1/Dh
3/5
≈ . (2.497)
λ(εK /λ2 )1/5 t λ
0
 2 Correlations and Anomalous Transport Models 145

Here, t0 is the mean field estimate of the characteristic time that corresponds to the
correlation scale balance. The solution of this equation yields
 5/12  7/36
λ λ2
t0 ≈ . (2.498)
V0 εK

This expression allows us to define the small percolation parameter that characterizes
the analyzed model based on approximation (2.445) of the stochastic layer width,
 λ )1/3 1/4
Δ (εK
ε= ≈ . (2.499)
λ V0

Now, one can obtain the scaling for the percolation layer width:
 1/5  1/4
εK 3/5 U
Δ(t) ≈ t0 ≈λ . (2.500)
λ2 V0

Here, U = (εK λ)1/3 is the dimensional estimate of the characteristic velocity that is
related to dissipative effects. Note that the very small percolation parameter
1/4
ε = ε0 = ((εK λ)1/3 /V0 )1/4 differs from the initial dimensionless parameter
ε0 = U/V0 = (εK λ)1/3 /V0 .
In the framework of the percolation approach, the expression for effective dif-
fusivity has the form (2.450) Deff ≈ V0 Δ(ε), which mirrors the key role of the
stochastic layer width Δ. The expression for the effective diffusivity takes the form
 1/4
(εK λ)1/3 3/4 1/12
Deff ≈ V0 Δ(ε) ≈ V0 λ ∝ V0 εK . (2.501)
V0

Note that such a form is true only for the small values of the parameter ε0 = U/V0 =
(εK λ)1/3 /V0 . On the other hand, it is necessary to take into account that correlation
size is limited by the sample size

a(ε) < L0 . (2.502)

Here, L0 is the characteristic external size of the system. We, in fact, obtain the
condition  1/ν  
λ (εK λ)1/3 1/4
<ε=  1. (2.503)
L0 V0
In relation to the consideration of the correlation scales balance method, it is inter-
esting to observe the dynamics of growth of two-dimensional small-scale structures
λ(t) near the percolation threshold. Computer simulations of turbulent flows often
need in obtaining a scaling describing small-scale structure evolution. Formally, we
can write the following expressions:

λ(t) ≈ ε ν a(t), (2.504)

146 O.G. Bakunin
 Dh
a(t)
L(t) ≈ λ ≈ V0 t. (2.505)
λ
Then, simple calculations yield

λ(t) ≈ ε∗ν a(t) ≈ ε∗ (λ)ν (V0 t)1/Dh , (2.506)

where ε∗ is the renormalized percolation parameter, which could be taken from

known percolation models. For instance, in the presence of seed diffusivity,
  1   1
1 ν+3 1 ν+3
ε≈ ∝ . (2.507)
Pe λ
After substitution one obtains
 1 ν+3
λ ≈ V02 D0ν+1 2ν+4 t 2ν+4 ∝ t 13/20 ≈ t 0.65 , (2.508)

where ν = 4/3 was used, and the correlation scale a(ε) is much less than the particle
diffusive path
a ≈ ε 3 V0 τD . (2.509)
Formula (2.508) leads to the simplest percolation estimate of the growth of the char-
acteristic scale of small structures in the system under analysis. Similar regimes were
obtained by simulations of astrophysical turbulence, where the value of the exponent
describing structure evolution was found to be 0.7 [103].

2.16 The Stochastic Magnetic Field and Percolation Transport

The results obtained in the framework of the time-dependent percolation model are
expressed in terms of the Kubo number. This makes it possible to use percolation
scalings to describe stochastic magnetic fields, since there exists a direct analogy
between the Kubo number and the magnetic Kubo number that is responsible for
correlation effects in a “braided” magnetic field.

2.16.1 The Stochastic Magnetic Field and Percolation Transport

We have considered the percolation approach from the simplest models that allow
us to obtain the diffusion coefficient of magnetic streamlines Dm . The percolation
model was suggested in order to describe cases with
b 0 LZ
Rm ≈ > 1. (2.510)
Δ⊥
Here, the value Rm is analogous to the Kubo number Ku = V0 T0 /λ. Therefore,
we can use the result of the time-dependent percolation case to analyze a “braided”
magnetic field. Isichenko [150, 151] applied this analogy directly. He rewrote the
 2 Correlations and Anomalous Transport Models 147

expression for Dm in the form (2.471)

−1/(ν+2)
Dm = b0 Δ⊥ Rm . (2.511)

Here, the value of the relative amplitude of the stochastic magnetic field b0 is used
instead of the velocity V0 , Δ⊥ corresponds to the spatial scale λ, and the value LZ
that enters into quasi-linear expression (2.239) corresponds to T0 that was used in
defining the Kubo number Ku = V0 T0 /λ. Many authors have carried out numerical
simulations [152–154] which allow us to consider that the analytical result (for the
7/10
monoscale model with the exponent 7/10) Dm ∝ b0 is in qualitative agreement
with the simulation results 0.6–0.8.
Isichenko [150, 151] also suggested using the percolation method to describe the
particle transverse transport in a stochastic magnetic field. Earlier, we considered the
interpretation of the flow parameters V0 , λ, T0 . In order to obtain the complete pattern
it is necessary to interpret the values
 Dh
−ν a
a(ε) = λ|ε| , L(ε) = λ , Δ ≈ λε. (2.512)
λ
To retain the physical meaning of these values we assume that the length of the per-
colation streamline on the plane, which is perpendicular to the magnetic field, can be
estimated in the form L ≈ b0 z. Here, z is the distance traversed along the force line.
We will consider the value Δb ≈ εΔ⊥ as a width of the stochastic layer Δ, which is
responsible for the particle percolation transport. The correlation (percolation) scale
a(ε) characterizes the transport properties of the system in the transverse direction
in accordance with the expressions considered above. Then, in terms of magnetic
diffusion we can obtain the expression for the particle transverse transport:
z(τ ) −1/(ν+2) z(τ )
Deff ≈ Dm ≈ b0 Δ⊥ Rm . (2.513)
τ τ
Here, τ is the correlation time. This expression implies that b0 z(τ ) > a.
Another important case is “ballistic” transport under the conditions when
b0 z(τ ) < Δ⊥ . For the same conditions the author of [150, 151] considered the esti-
mate [see the “fluid limit” (2.241)]

z(τ )2
Deff ≈ b02 . (2.514)
τ
There also exists an intermediate regime that corresponds to the “initial” stage of the
motion along the percolation streamline (2.480) when a > L ≈ b0 z > Δ⊥ . Here, it
is convenient to use the running correlation scale a0 (τ ),
   
L 1/Dh b0 z(τ ) 1/Dh
a0 ≈ Δ⊥ ≈ Δ⊥ , (2.515)
Δ⊥ Δ⊥
as a parameter of the problem. This corresponds to the intermediate regime (2.482) in
ν/(ν+2)
the time-dependent model. This estimate is correct if a0 (t) < a(ε∗ ) ≈ Δ⊥ Rm .
148 O.G. Bakunin

We can easily calculate the boundary value to be

Δ⊥ Dh ν/(ν+2)
zm ≈ Rm . (2.516)
b0
Then, applying the results obtained earlier one can obtain

a02
Deff ≈ P∞ (a0 ). (2.517)
τ
In the case under consideration the full hierarchy of spatial scales can be presented
in the form
 
−1/(ν+2) b0 zm 1/Dh
Δb (ε) ≈ Δ⊥ Rm < b0 z ≤ Δ⊥ < Δ⊥
Δ⊥
ν/(ν+2)
≤ Δ⊥ Rm ≈ a(ε∗ ). (2.518)

Here, the appearance of “intermediate” scales is related to the ballistic character of

the estimate of the transverse displacement L ≈ b0 z.
Isichenko [150, 151] suggested the consideration of two mechanisms of motions
along the z-axis related to a magnetic force line. The diffusion mechanism is

z ≈ D t for t > τcoll . (2.519)

The ballistic (kinetic) mechanism corresponds to the expression

z = V0 t for t < τcoll . (2.520)

To define the correlation times τ the author of [150, 151] used several models con-
sidered earlier: the percolation estimate τ ≈ εT0 ; the Kadomtsev–Pogutse estimate
τ ≈ Δ2b /D⊥ ; and the correlation time that corresponds to the Rechester–Rosenbluth

model with Lcor ≈ D τ .
A good example of the efficiency of the suggested approach is the possibility
of obtaining characteristic transport regimes. Thus, the use of diffusive and ballistic
expressions in the formula for Deff (2.514) leads to the fluid limit Deff ≈ b02 D .
On the other hand, the collisionless result Dm V0 can be obtained by the substitu-
tion of the ballistic expression into the formula for Deff (2.513). It is obvious that
many possible combinations lead to a large variety of regimes. The author of [150,
151] considered in detail all the possible situations and the corresponding graphs are
shown.

2.16.2 Percolation and the Kadomtsev–Pogutse Scaling

It is possible to consider percolation effects in the stochastic magnetic field in the

context of the well-known Kadomtsev–Pogutse scaling (2.358). The model sug-
gested in [67] is based on the assumption of the presence of a strong longitudinal
magnetic field. For these conditions the longitudinal diffusion coefficient is much
 2 Correlations and Anomalous Transport Models 149

greater than the transverse diffusion coefficient D > D⊥ and the choice of the
value ε0 = D⊥ /D as an initial small parameter is fairly natural. The renormaliza-
tion condition, which allows the true small parameter to be obtained, has to mirror
the competition between the longitudinal and transverse decorrelation mechanisms.
Isichenko [150, 151] suggested reformulating the particle balance condition in
the stochastic layer (2.331), (2.332) with allowance for the diffusion character of
longitudinal motions:
n n
D⊥ L ≈ D Δ. (2.521)
Δ L
Here, n is the particle number density, L = L(ε) is the length of the fractal stream-
line, and Δ = Δ(ε) is the width of the stochastic layer. We can rewrite this condition
in terms of the equivalence of decorrelation times, by analogy with the convective
cells case:
L(ε)2 Δ(ε)2
τ≈ ≈ . (2.522)
D D⊥
Equations (2.521) and (2.522) are the conditions to determine the small parameter of
the percolation problem:
 
D⊥
ε∗ = ε∗ (ε0 ) = ε∗ . (2.523)
D
To solve these equations we have to find the value Δ(ε). However, the result of
interest to us can be obtained from the formal definition of the value Deff (2.443):

a 2 LΔ LΔ
Deff ≈ ≈ . (2.524)
τ a2 τ
It is easy to see that in our case
Δ L 
Deff ≈ D ≈ D⊥ ≈ D D⊥ , (2.525)
L Δ

where Δ/L ≈ D⊥ /D . Using the percolation expression for L(ε) and conjecture
Δ = ελ, one can obtain the expression for the correlation scale:
 
D ν/(2ν+4)
a=λ . (2.526)
D⊥
Hence, the percolation interpretation of the Kadomtsev–Pogutse scaling is related to
the renormalization of the initial small parameter in the following form:
1/(2ν+4)
ε∗ = ε0 . (2.527)

Now it is possible to express all the values by means of the small parameter ε0 and
the percolation exponent ν.
Obviously, the percolation interpretation for scaling (2.358) is not necessary.
However, the calculations point out the possibilities that exist for the generalization
of both the renormalization condition and the alteration of the small parameter ε0 .
150 O.G. Bakunin

2.16.3 Percolation Renormalization and the Stochastic Instability Increment

The consideration of time-dependent two-dimensional random flow is of particular

interest because there is no exponential separation of initially nearby streamlines in
the steady case in a bounded two-dimensional area. Actually, in the steady case we
deal with the Hamiltonian one-dimensional problem, and only in the case ω = 0
is there the possibility to investigate stochastic behavior, since the time-dependence
adds a degree of freedom. The percolation approach to the consideration of the hier-
archy of spatial and temporal scales allows us to treat long-range correlation effects
in terms of simple scalings. One of the important problems here is to obtain esti-
mates of the stochastic instability increment. In the framework of the mono-scale
approach, the scaling for the stochastic instability increment was obtained in the pa-
per by Grusinov, Isichenko, and Kalda [83]. Analyzing the spatial scales hierarchy
in a two-dimensional random time-dependent velocity field, the authors of [83] esti-
mated the square corresponding to the stochastic layer Δ as

L(ε∗ )Δ(ε∗ ) ≈ a(ε∗ )λε∗ /ε∗ ≈ aλ  λ2 . (2.528)

Naturally, this is because there are many streamlines in the stochastic layer. At the
same time, this means that there exists a spatial scale lS that characterizes a single
streamline. Indeed, a possible estimate of lS is the expression [40]

λ2
lS (ε) ≈ ≈ λε ν+1 ≈ εΔ  Δ ≈ λε. (2.529)
L(ε)
The characteristic reconnection time of a pair of nearby separatrixes can be estimated
as
1 VS λω L(ε)ω
γS ≈ ≈ ≈ ≈ , (2.530)
τS (ε) lS (ε) lS (ε) λ
where VS = λω is the dimensional estimate of velocity of separatrix motion. On the
other hand, the balance of characteristic times τS ≈ τB , which has the form
λ L(ε∗ )
τS = ≈ = τB , (2.531)
L(ε∗ )ω V0
allows the small percolation parameter ε∗ to be defined:
  1   1  3
λω 2(ν+1) 1 2(ν+1) 1 14
ε∗ ≈ ≈ ≈ , where ν = 4/3. (2.532)
V0 Ku Ku
The final expression for the stochastic instability increment γS then takes the follow-
ing form:
1 L(ε∗ ) √
γS ≈ ≈ω ≈ ω Ku. (2.533)
τS λ
It is easy to take into consideration here the logarithmic factor, which leads to the
result √
γS ≈ ω Ku/ ln Ku for Ku > 1. (2.534)
 2 Correlations and Anomalous Transport Models 151

Since the Kubo number Ku can be easily interpreted in terms of the magnetic Kubo
number Rm ≈ b0 L0 /Δ⊥ , it is interesting to compare the obtained expression with the
Kadomtsev–Pogutse result [67], γZ ≈ Rm 2 /L for the quasi-linear case. We see that
Z
the stochastic instability increment in the percolation case (long-range correlations)
is characterized by smooth scaling, which is analogous to the dependence of the
effective diffusion coefficient on the Kubo number.

2.17 Percolation in Drift Flows

Drift effects play an important role in turbulence. Even a small drift velocity can
significantly change the character of turbulent transport. In this section, we consider
both the steady and time-dependent models of drift effects in terms of the renormal-
ization of the small percolation parameter.

2.17.1 Graded Percolation and Drift Flows

The problem of the influence of the small external perturbation on the percolation
system has been considered in terms of the graded approach [147]. Yushmanov [84]
analyzed the influence of a small drift velocity Ud on the fractal topology of stream-
lines (see Fig. 2.11) in the framework of the Trugman theory:
V = V0 + Ud , Ud  V0 . (2.535)
Here, V0 is the mean flow velocity. Zeldovich [155] has already formulated this prob-
lem, but at that time the renormalization methods had not yet been developed. The
simplest way to alter the small parameter is by using the value
Ud
ε0 = . (2.536)
V0
However, in this approach the fractal character of percolation streamlines is com-
pletely lost. Yushmanov suggested the use of the following dimensional estimate of

Fig. 2.11. Disconnected streamline and drift flow

152 O.G. Bakunin

the drift velocity:

a(ε) L(ε)
Ud = P∞ (a), a(ε) ≈ λ|ε|−ν , τ (ε) ≈ . (2.537)
τ (ε) V0
For P∞ he used an additional expression, which was suggested in the steady case,
P∞ ≈ λ/a. Simple calculations allow us to obtain the parametric dependence for the
renormalized small parameter ε∗ on the flow parameters V0 and Ud ,
 1/(1+ν)
Ud
ε∗ = , (2.538)
V0

where ν = 4/3. It is easy to see that this expression completely coincides with the
Trugman result (2.439) and can be interpreted in terms of the streamline function,
Ψ1 εΨ0
Ud ≈ ≈ . (2.539)
a(ε) a(ε)
Here, we are dealing with the conditions

Ψ1  Ψ0 ≈ λV0 and a(ε)  λ. (2.540)

One obtains an estimate that is related to the finite size of a system a(ε∗ ) ≤ L0 :
V0 (λ/L0 )(ν+1)/ν < Ud < V0 . Moreover, we can note how the spatial hierarchy of
scales is included in the description of perturbations [156]:
λ λ Δ
Ud ≈ ε V0 ≈ V0 ≈ V0 . (2.541)
a(ε) L(ε) a(ε)
We see that parameter λ does not enter into the expression for the renormalized
value ε∗ . It is possible to determine some velocity U , which will be much less
than Ud :
Δ
U≈ V0  Ud  V0 . (2.542)
L(ε)
This problem will be considered below in connection with the study of compressibil-
ity effects.
There are also alternative possibilities for the renormalization. For example, we
can consider the following expression:

a(ε) εa(ε)
Ud = ε = V0 . (2.543)
τ (ε) L(ε)
This case corresponds to setting P∞ (ε) ≈ ε [compare with (2.453)]. However, in
contrast to the Yushmanov case, these estimates lead to the expression Ud ≈ ε 2 V0 ,
which does not contain the percolation exponent ν. Undoubtedly, this is an important
drawback. Therefore, taking into account the transport effects, we will follow the
Trugman renormalization (2.439).
 2 Correlations and Anomalous Transport Models 153

Formally applying the results of the previous sections we can obtain the estimate
of the diffusion coefficient in terms of the stochastic layer width Δ ≈ λε∗ :
 3/7
Ud 4/7
Deff ≈ V0 Δ(ε∗ ) ≈ λV0 (ε0 )1/(1+v) ≈ λV0 ∝ V0 . (2.544)
V0

A more complex situation arises when both the drift effects and the time-dependence
characterized by the time T0 ≈ 1/ω exist simultaneously in a system.

2.17.2 Low Frequency Regimes and Drift Effects

The analysis of drift effects on the basis of Trugman’s renormalization does not take
into account processes related to the temporal fluctuations of the velocity field. More-
over, in the framework of the “graded” approach (2.539) the correlation time τ plays
a formal role because to define the effective diffusion coefficient Deff = V0 Δ, the
conjecture Δ = λε∗ is usually used. Nevertheless, the scaling approach allows us
to consider percolation transport in the presence of both drift effects and fluctua-
tions. The main parameter that characterizes the temporal dependence of the stream
function is the characteristic time:
Ψ
τD ≈ . (2.545)
∂Ψ/∂t
The percolation character of transport is manifested in the estimate Ψ ≈ ε∗ Ψ0 ≈
ε∗ V0 λ. Here, ε∗ is the small parameter of the percolation model. On the other hand,
it is necessary to take into account the “graded” character of perturbations (2.539),
1/(1+ν)
Ψ ≈ Ud a(ε∗ ), ε∗ = ε0 . (2.546)

Then, estimates yield:

∂Ψ
≈ Ud a(ε∗ )ω. (2.547)
∂t
Here, Ud is the drift velocity, a is the correlation scale, and ω is the characteristic
frequency of the stream function alteration. Calculations yield
1 ν/(1+ν)
τ= ε . (2.548)
ω 0
Yushmanov [84] suggested using the simplest quasi-linear dependence τ (ω) ≈ 1/ω
to represent the effective diffusion coefficient (2.428):

a2 a(ε∗ )
Deff ≈ P∞ ≈ Ud2 τ . (2.549)
τ λ
Here, P∞ ≈ λ/a(ε∗ )  1, which corresponds to the drift model [84] and ε∗ =
(Ud /V0 )1/(1+ν) , which corresponds to the Trugman approach. Upon substituting the
154 O.G. Bakunin

expression for τ = 1/ω into (2.549), we obtain the scaling

   
Ud2 1 ν/(1+ν) Ud2 V0 4/7 10/7
Deff ≈ ≈ ∝ Ud . (2.550)
ω ε0 ω Ud
This dependence shows that there is an essential difference between transport in
the presence of velocity field fluctuations and the steady case (2.544). However, the
method of calculating the effective diffusion coefficient Deff , which was suggested
by Yushmanov, only corrects the percolation expression (2.549), whereas the cor-
rect method would be to include the frequency ω into the renormalization condition
that characterizes transport in the drift flow in the presence of the time-dependent
perturbation of the stream function. Thus, in the steady case [82] the balance of
characteristic times was the main condition for the renormalization
Δ(ε)2 L(ε)
τ≈ ≈ . (2.551)
D0 V0
In considering the fluctuations of the stream function, the diffusion estimate of the
characteristic time is given by (2.474)
h2 (ελV0 )2
τD ≈ ≈ , where DΨ ≈ V02 D0 . (2.552)
DΨ DΨ
However, in considering the drift effects we will use the correlation estimate [84]
Ψ ≈ Ud a(ε∗ ). The important aspect is that the streamline diffusivity DΨ is respon-
sible for the physical mechanism of the distortion of streamlines. Let us approximate
the value DΨ in the form that mirrors the certain character of the external influences
(drift flow and time dependent perturbation):

DΨ ≈ (Ψ )2 ω ≈ Ud2 a(ε)2 ω. (2.553)

The new equation for the small percolation parameter ε∗ takes the form [157]
(ελV0 )2 L(ε)
2
= . (2.554)
2
Ud a(ε) ω V0

In fact, we have renormalized the value DΨ ≈ V02 D0 in expression (2.191) in accor-

dance with mechanisms distorting streamlines (the drift flow with the characteristic
velocity Ud and temporal fluctuations with the frequency ω):
 2
Ud
D0 → a(ε)2 ω. (2.555)
V0
Solving (2.554), we obtain a new small parameter for the problem of percolation
transport in the presence of both the drift flow Ud and fluctuations with the charac-
teristic frequency
  2   1
Ud 3(1+ν) 1 3(ν+1) 2
−3 1
ε∗ ≈ ∝ Ud7 V0 7 ω 7 . (2.556)
V0 Ku
 2 Correlations and Anomalous Transport Models 155

Note that the characteristic size λ is included in the definition of the small parame-
ter in contrast to the Yushmanov model. The expression for the effective diffusion
coefficient takes the form
  2   1
Ud 3(ν+1) λω 3(ν+1) 2 1
Deff ≈ V0 λε∗ ≈ ∝ Ud7 ω 7 . (2.557)
V0 V0
This result differs significantly from the quasi-linear representation (2.550).
The increase in the number of parameters allows us to consider different transport
regimes. For example, in the case Ud = λω we obtain
  2   1
Ud 3(1+ν) 1 3(ν+1)
ε∗ ≈ . (2.558)
V0 Ku
The physical meaning of this regime becomes clear when we consider the renor-
malization condition L(ε)/V0 = 1/ω and compare it with (2.464). This shows the
absence of the intermediate characteristic time τD in this case. It is possible to obtain
other estimates also based on the conditions Ud  λω or Ud  λω.

2.17.3 Compressibility and Percolation

In the models of turbulent diffusion, the subdiffusion character of transport can be

related to the presence of compressibility effects and the “trapping” character of the
interaction between a passive tracer and vortex structures. The analysis of the com-
pressibility effects is important for two-dimensional flows since in two-dimensional
incompressible flows the subdiffusion mechanism cannot be realized [114].
The author of [17] mentioned the computer simulation of the influence of small
compressibility effects on a percolation flow. The monoscale approach based on

V ≈ [∇Ψ × ez ] + ε∇Ψ (2.559)

was suggested. In this equation, the second term describes the deviation effect of
the real velocity from the streamline. The results of the computer simulation are
given in [17] and clearly show the appearance of “trapping” in the flow. We see
that the dimensional estimate of the flow pattern “distortion” by the compressibility
effects is similar to the “graded” estimate. However, we can obtain a new type of
“renormalization” for this case if we note that in the framework of the graded method
the following relationship was used:
λ Δ(ε∗ )
Ud ≈ V0 ≈ V0 ≈ V0 ε∗ν+1 . (2.560)
L(ε∗ ) a(ε∗ )
Applying the assumption of the “weakness” of compressibility effects we can obtain
another condition of renormalization to describe the velocity deviations:
Δ(ε∗ )
Uc ≈ V0 ≈ V0 ε∗ν+2 . (2.561)
L(ε∗ )
156 O.G. Bakunin

It can also be interpreted by taking into account the particle balance (compare with
the convective cells case)
nUc L(ε∗ ) ≈ nV0 Δ(ε∗ ). (2.562)
Hence, in the problems accounting for a weak compressibility, the new parameter of
smallness is given by
  1
Uc 2+ν 1
ε∗ = = (ε0 ) 2+ν  ε, (2.563)
V0
where ν = 4/3. In the framework of the monoscale approach it is easy to obtain the
dependence for the effective diffusion coefficient:
3/10 7/10
Deff ≈ V0 Δ(ε) = V0 λε∗ ∝ Uc V0 . (2.564)
Considering the compressibility effects on the basis of using the small parameter
ε0 ≈ Uc /V0 , we are in fact dealing with the Kubo number. Thus, if we introduce the
characteristic time τc ≈ λ/Uc then the expression for the small percolation parame-
ter takes the form
  1   1
λ ν+2 1 ν+2
ε∗ ≈ ≈ . (2.565)
V0 τc Ku
As was mentioned earlier, there is a close interrelation between the compressibility
effects and trapping; [158] was devoted to the analysis of the trapping based on the
analysis of the correlation mechanisms. Even simple estimates of the correlation time
τ ≈ λ/V0 + τT (2.566)
using the characteristic time of finding the particles in traps τT = τT (Ku) show the
essential alteration of the transport character with Ku  1:
λ2 Ku λ2
Deff ≈ ≈ D0 τ (Ku)
≈ . (2.567)
τ 1 + T τ Ku τT (Ku)
The analysis of the regimes in [158] was also carried out using the power form of
correlation functions. Moreover, the model approximation of the streamline function
was applied. However, the method suggested in [158] differs significantly from the
percolation method, since the analysis was carried out on the basis of the special
trajectory ensemble. This approach will be discussed.

2.18 Multiscale Flows

The above-considered percolation approach to turbulent transport problems was
based on the monoscale representation of a flow with the parameters V0 , λ. The
natural generalization of this model is the analysis of multiscale flows where the hi-
erarchy of spatial scales is related to the velocity hierarchy. This makes it possible to
treat the anomalous transport in terms of the correlation function exponent and the
Hurst exponent.
 2 Correlations and Anomalous Transport Models 157

2.18.1 The Nested Hierarchy of Scales and Drift Effects

Here, we consider the generalization of the monoscale percolation model by the

analysis of multiscale flows with the hierarchy of spatial scales, which is related
to the velocity hierarchy. The main difference of this multiscale approach from the
Kolmogorov hierarchical model of turbulence is the assumption of the percolation
character of the streamline behavior. Such a model has been considered by Isichenko
and Kalda in [85, 86]. The expression for the streamline function Ψλ as a function
of the parameter λ was presented in the scaling form with the stream function expo-
nent M:  M
λ
Ψλ ≈ Ψ0 . (2.568)
λ0
Here, Ψ0 is the scale of the streamline function that corresponds to the characteristic
spatial scale λ0 and the characteristic velocity V0 . In agreement with the dimensional
estimates, we can obtain the expression for the velocity Vλ that corresponds to the
scale λ:  M−1
Ψλ λ
Vλ ≈ ≈ V0 . (2.569)
λ λ0
In the framework of the percolation approach, the most intensive streamlines will
contribute most to the transport. Two different situations occur with respect to the
value of M.
Let us consider the first case, M > 1. Here, the characteristic flow pattern is
determined by the maximum value of λ ≈ λm for such a system.
The second case of interest arises when M < 1 and appears to be more complex.
The authors of [85, 86] considered the hierarchy of spatial scales,

λ0  λ1  λ2  · · · , (2.570)

and the corresponding hierarchy of velocities,

V0  V1  V2  · · · . (2.571)

Here, the total velocity V = V0 + V1 + V2 + · · · is the superposition of flows en-

tering in the hierarchy. Formally, this flow pattern is similar to the hierarchy used
by Kolmogorov to describe isotropic turbulence [27, 29, 30, 63]. In that case, the
Kolmogorov scaling law for the spectrum of energy E(k) is one of the important
characteristics, E(k) ∝ Vk2 /k ∝ 1/k 5/3 . Here, k is the wave number. The esti-
mate of the velocity gives Vk ∝ 1/k 1/3 ≈ λ1/3 . A comparison with the multiscale
expression yields M = 4/3 > 1. An analogous situation arises when considering
the spectrum for the two-dimensional model E(k) ≈ 1/k 3 , which is more relevant
for the two-dimensional percolation case under consideration. We see that the Kol-
mogorov approach differs significantly from the case considered in [85, 86] where
M < 1. Isichenko and Kalda based their analysis of the hierarchy on Trugman’s
ideas of “graded” percolation. In the previous section, we noted that this approach
[147] is equivalent to the Yushmanov relationship [84]. These approaches are based
158 O.G. Bakunin

Fig. 2.12. Hierarchy of scales

upon the concept of the distortion of a small scale related flow, by way of superim-
position of a weaker drift flow related to a large scale over V = V0 + V1 . In the case
of the scale hierarchy the Trugman renormalization [84] can be represented in the
form
1 Vi+1
ε∗ (i) = ε0 (i) 1+ν  ε0 (i) ≈ . (2.572)
Vi
Following the ideas of the monoscale percolation approach, the authors of [85, 86]
suggested
 Dh
ai 1
ai (ε∗ ) = λi ε∗−ν , Li (ε∗ ) = λi , P∞ (i) ∝ . (2.573)
λi ai
Here, the values a0 , Δ0 , L0 correspond to the scale λ0 . The values P∞ (i) and Δi are
interdependent,
Li Δi
P∞ (i) = . (2.574)
ai2
Therefore, we obtain a relationship that is similar to the case of monoscale perco-
lation Δi (ε∗ ) ≈ ε∗ λi . An assumption was made about the hierarchical flow pattern
based on the system of nested scales (see Fig. 2.12),

ai (ε∗ ) ≤ λi+1 . (2.575)

This condition plays an important role in the multiscale approach, since the value ε∗
can be expressed as a function of λi+1 and λ1i :
 
Vi+1 λi+1 M−1
ε0 (i) ≈ ≈ . (2.576)
Vi λi
Then, simple calculations yield
  ν(1−M)
λi λi+1 1+ν
< 1. (2.577)
λi+1 λi
 2 Correlations and Anomalous Transport Models 159

This condition means that the assumption of nested scales is correct only for
1
− < M < 1. (2.578)
ν
In fact, we have a double inequality for the value M that describes the case of interest.

2.18.2 The Brownian Landscape and Percolation

The scaling representation of properties within closed fractal loops plays an impor-
tant role in the problem of the description of turbulent diffusion in two-dimensional
geometry. In this paper we have considered the possibility of using the exact scaling
dependences for a “cluster hull” to describe the percolation transport. There is one
more interesting problem related to the fractal dimensionality of single loops in the
framework of the modeling of the “rough surface” problem. Moreover, the multiscale
approach [85, 86] is the “percolation version” of the rough surface problem.
The simplest interpretation of this problem [56, 57] is based on the representation
of a “rough” 1D + 1D landscape as a graph of one-dimensional random walks in x–t
axes, where the t-axis can be interpreted as a horizontal coordinate and the x-axis
can be a vertical one. Then, different values of the Hurst number correspond to dif-
ferent types of landscape “roughness” (x)2  ∝ t 2H . This implies that the “rough
landscape” is a statistically self-affine fractal over a corresponding range of length
scales with the characteristic Hurst exponent equal to the roughness
 exponent H (see
Fig. 2.13). For such landscapes the mean height difference (h)2 between the pairs
of points separated by a “horizontal” distance r is given by

(h)2 ∝ (r)H . (2.579)
It is easy to generalize this representation for the case of a rough surface with an-
other dimensionality. Note that Isichenko and Kalda [85, 86] considered a similar

Fig. 2.13. The Brownian landscape

160 O.G. Bakunin

model where the streamline function Ψλ is used as the “height” characteristic of the
two-dimensional random field Ψλ ≈ Ψ0 (λ/λ0 )M . This is analogous to the model of
continuum percolation; the loops arise when considering a coastline that was formed
under the conditions of a “landscape flooded by water”. In this connection, there is a
problem in obtaining the relationship between the fractal dimensionality characteriz-
ing a single loop D̃h and the Hurst exponent H (or the stream function exponent M).
Here, we will use another symbol D̃h instead of the hull percolation exponent Dh ,
since in order to describe the Brownian surface the percolation ideas were used as
approximations only.
The theoretical probabilistic approximation is, as usual, the simplest method.
The authors of [159] suggested the use of the model of self-avoiding random walks
to describe the single loop character. The functional suggested by Flory (2.110)
could be an adequate model. However to describe cases with different Hurst ex-
ponents it is necessary to use the probability density function with the arbitrary val-
ues H : R ∝ N H , instead of the Brownian case, where H = 1/2. Then, the expres-
sion for the probability of self-avoiding Brownian motion takes the form
 ∞    
1 1 R2
PS (t) = exp − d (N )2 exp − (dR)d . (2.580)
−∞ R (N H )d N 2H

We assume that the main contribution to the integral comes from the extremum of
the integrand,  
1 R2
min (N ) + 2H .
2
(2.581)
Rd N
Performing calculations we can obtain the scaling

N ∝ R (d+2)/2(1+H ) . (2.582)

The authors of [166] suggested considering the fractal dimensionality

d +2
dF = (2.583)
1+H
for the two-dimensional case (d = 2) as a fractal dimensionality of the single contour
loop (coastline) of a self-affine surface with the Hurst exponent H :
2
D̃h = . (2.584)
1+H
Now it is possible to make several simple estimates. The value of H = 1 yields
result which corresponds to the linear type of behavior with D̃h = 1. The random
walk with H = 1/2 corresponds to D̃h = 4/3.
However, approximation (2.584) is not correct in the region of small H since the
following condition must be realized:
1 7
D̃h ≤ Dh = 1 + = . (2.585)
ν 4
 2 Correlations and Anomalous Transport Models 161

This condition has a clear physical meaning. The fractal dimensionality of the hull
that has a percolation nature has to be larger than the fractal dimensionality of the
coastline of the self-affine surface.
Isichenko and Kalda [85, 86] suggested another version of the approximation D̃h
based on the hierarchy of nested scales (2.570). In the framework of this approach
the correct description of the regimes, where D̃h tends to the hull exponent Dh ,
is possible, since the multiscale approach is based on “graded” percolation. Under
these new conditions the expression for the value Dh , with a0 → λ1 , should be
reconsidered from the standpoint of the interaction of nearby scales of the hierarchy.
Recall that Dh = 1 + 1/ν; therefore, in the monoscale approach we have L(a) ≈
a/ε∗ . Then, using the assumption a0 → λ1 , the authors of [85, 86] suggested the
simple approximation
λi
L(λi ) ∝ . (2.586)
ε∗ (λi )
In the framework of this approach, the calculations yield the renormalized value D̃h ,
ν 1 + νM
D̃h = Dh (1 − M) + . (2.587)
1+ν 1+ν
For the two-dimensional case, the following simple expression was obtained:
10 − 3M
D̃h = , (2.588)
7
where D̃h (−1/ν) = Dh and D̃h (1) = 1. However, this result has not passed recent
tests [165] because there is a rigorous result, which gives the value D̃h (0) = 3/2.
A new approximation for 0 < M < 1 was suggested, namely D̃h = (3 − M)/2,
which is in agreement with the rigorous result for D̃h (0) [163, 164].
Unfortunately, this new approximation does not give any information about the
behavior of D̃h in the region of negative M. This is not surprising, since in the frame-
work of the Brownian theory of “rough” surfaces we cannot interpret the values
H < 0. However, the “graded” percolation method used in [85, 86] allows these val-
ues to be analyzed. In the next section, we will consider the physical interpretation
of the exponent M from the correlation point of view.

2.18.3 Correlations and Transport Scalings

Completing the consideration of the basic assumptions of the multiscale approach,

let us focus on the physical meaning of the parameter M. Obviously, this parameter
governs the character of a flow. We can relate it to the correlation properties of a
flow. The simple estimate for the spatial correlation function in the scaling law form
is given by
1
C(λ) ∝ V (λ)2 ∝ λ2(M−1) ∝ α . (2.589)
λ C
Here, αC is the correlation exponent that describes the rate of decay of the correlation
function
162 O.G. Bakunin

αC = 2(M − 1). (2.590)

Cases where M ≈ 1 correspond to the scaling law C(λ) ≈ const. In the cases
M ≥ −1/ν the correlation function is “steepest”. The condition of applicability of
the Isichenko–Kalda hierarchical model takes the following form [111]:

0 < αC < 7/2. (2.591)

In subsequent considerations, in order to describe transport properties we will use

the exponent αC together with the exponent M. The above analysis permits us to
interpret the scaling estimate of transport, because the rigorous methods often show
the relationships between the Hurst exponent and the correlation exponent.
Following the ballistic character of the estimates of the percolation effects,
Isichenko and Kalda [85, 86] suggested a scaling law for the calculation of the Hurst
exponent H in the multiscale case:
 M−1
λ(t)
λ(t) ∝ V (λ)t ≈ V0 t. (2.592)
λ0

Then, simple calculations make it possible to obtain the expression

 1/(M−2)
V0 t
λ(t) ≈ λ0 ∝ tH . (2.593)
λ0

Hence, the Hurst exponent is

1 2
H = = . (2.594)
2−M 2 + αC
This ballistic estimate (2.592) looks too rough [compared with (2.177)]. However, in
terms of the correlation exponent αC , one obtains the rigorous result for incompress-
ible flows with the power correlation function [51] where aC < 2. We see that the
steeper correlation functions correspond to the lower values of the Hurst exponent.
The correlation function C(λ) ∝ 1/λ2 corresponds to the case of classical diffusion
with H = 1/2.
The efficiency of similar estimates allows us to consider more complex flows,
where the effects of the anisotropy of the medium play an important role, together
with the multiscale effects λ⊥ (t) ∝ V⊥ (λ⊥ , λ )t. Formula (2.592) does not take into
account the anisotropy effects; however, one can consider an analogous approach for
the simplest anisotropic case with separated spatial scales [17, 72]:

λ⊥ (t) ∝ V⊥ (λ )t. (2.595)

Here, λ⊥ is the perpendicular displacement and λ is the longitudinal displacement.

If the case under consideration shows a diffusion character in the longitudinal√motion
(the double diffusion, the Dreizin–Dykhne model, etc.), then we obtain λ ≈ 2D0 t.
 2 Correlations and Anomalous Transport Models 163

Upon substitution of this estimate into (2.595) we find the scaling

 (1−M)/2
λ20
λ⊥ (t) ∝ V0 t (1+M)/2 . (2.596)
D0

The expression for the Hurst exponent takes the form [167, 168]
1+M αC
H = =1− , (2.597)
2 4
where 0 < αC < 7/2. The comparison between (2.597) and (2.594) shows the coin-
cidence of both these dependences at two points, αC = 0 (H = 1 is the ballistic case)
and αC = 2 (H = 1/2 is the classical diffusion equation). The characters of both
these dependences H (αC ) are similar. Moreover, the exponent M really characterizes
not only scaling but also the physical properties of the flow. Thus, for M = 1/2 we
obtain the Dreizin–Dykhne result H = 3/4, which is in agreement with the isotropic
case (2.594), where for M = 1/2 one can obtain H = 3/2, which corresponds to the
“Manhattan grid” flow [17] (the generalization of the shear flows model [72]). Note
that for αC > 2 scaling (2.597) yields the subdiffusive regime, which contradicts the
initial assumptions about the incompressibility of the flow and using the streamline
concept.

2.18.4 The Diffusive Approximation and the Multiscale Model

Besides the ballistic approach (2.592), it is possible to consider the diffusion approx-
imations
λ2 ≈ D(λ)t. (2.598)
Here, D(λ) is the coefficient corresponding to the scale λ. However, the expressions
for D(λ), which were suggested by Isichenko and Kalda, are based on the approx-
imation (2.588) of the exponent D̃h . This is not surprising, since in the monoscale
approach the expression for Dh = 1 + 1/ν also plays a key role. The diffusion coef-
ficient D(λ) can be expressed in the form that is analogous to the monoscale case:

λ2
D(λ) ≈ P∞ (λ), (2.599)
τ
where P∞ (λ) ≈ L(λ)Δ(λ)/λ2 . The authors of [85, 86] suggested an approximation
of P∞ (λ) in the form corresponding to the monoscale approach (2.453):
λi
P∞ (i) ≈ . (2.600)
ai (λi )
Then, calculations yield the expression
 4(1−M)/7
λ0
P∞ (λ) ≈ . (2.601)
λ
164 O.G. Bakunin

This result allows the width of the percolation layer to be defined:

 M
λ2 λ
Δ(λ) ≈ P∞ (λ) ≈ λ0 . (2.602)
L(λ) λ0

Introducing the correlation time τ into the ballistic form τ ≈ L/V0 into the expres-
sion for D(λ) leads to the Koch–Brady result (2.594). In this case, the conditions
M > 0 (αc < 2) are automatically satisfied, since D(λ) has to increase if λ in-
creases:  M
λ
D(λ) ≈ V0 Δ(λ) ≈ V0 λ0 . (2.603)
λ0
A new regime arises when the correlation time is considered in the diffusive form
τ = Δ2 (x)/D0 :
L(λ)
λ2 ≈ D0 t. (2.604)
Δ(λ)
Simple calculations [85, 86] lead to the scaling for the Hurst exponent
7
H = . (2.605)
10M + 4
Here, −2/5 < M < 1. It easy to see that the scaling obtained by Isichenko and
Kalda depends essentially on the choice of the expression for L(λ). Thus, for the
approximation of L(λ), the substitution of ai for λi was used, whereas for the ap-
proximation of F (λ) the conventional form of ai was retained. It is easy to explain
the choice made in [85, 86], since the expression in the form L(λi ) ≈ ai /ε leads to
the value D̃h (1) = 0, which is absolutely unacceptable. On the other hand, approxi-
mation (2.321) provides the correct limit for negative values of M: D̃h (−1/ν) = Dh
and the realization of Mandelbrot’s condition D̃h < 2 − M with M < 1. From
the modern standpoint correct calculations could be based on the recent results of
Kondev et al. [162, 163] and Kalda [164, 165]:
3−M
D̃h = . (2.606)
2
Then, upon carrying out calculations we obtain
 11(1−M)/7
L(λ) L2 (λ) λ
D ≈ D0 ≈ D0 2 ≈ D0 . (2.607)
Δ(λ) λ P∞ λ0

The diffusion coefficient D(λ) increases with λ for the case when M < 1, yielding
the superdiffusion character of the behavior in the region 0 < M < 1:
7
H = . (2.608)
3 + 11M
Note that the use of new approximation (2.606) leads to alteration in the formula
for the diffusion coefficient D(λ)that is based on the ballistic expression for τ ,
 2 Correlations and Anomalous Transport Models 165

since
1
D(λ) ≈ V0 Δ(λ) ∝ . (2.609)
L(λ)
In this case the new scaling for the Hurst number does not coincide with the expres-
sions by Koch and Brady (2.594).
However there are also other possibilities, since the approximation of P∞ (λ)
in the form (2.600) is not universal. Thus, the estimate P∞ ≈ ε0 ≈ (λ/λ0 )M−1
can be used in the region of values M ≤ 1. Another way is based on using the
united approximation formula on the basis of three characteristic values of D̃h (M)
at the points M = 1, M = 0, and M = −1/ν for the whole diapason of para-
meters −1/ν < M < 1. The estimates considered show that diffusive approxi-
mation (2.598) has several degrees of freedom, which leads to considerable uncer-
tainties in the results. Nevertheless, the multiscale approach essentially increases
the possibilities of applying percolation ideas for the description of turbulent diffu-
sion.

2.18.5 Stochastic Instability and Time Scales

To consider stochastic instability in the framework of the multiscale approach it is

convenient to introduce the hierarchy of characteristic times (frequency scales)
 G
1 λ λ0
tλ ≈ ≈ , (2.610)
ωλ λ0 V0
where G is the frequency exponent, which permits modeling both ballistic G = 1
and diffusive regimes G = 2. Based on the multiscale results obtained by way of
multiscale approximation, we can consider the stochastic instability increment in
terms of the monoscale definition
VS L(λ)λωλ
γ̃S (λ) ≈ ≈ ∝ λD̃h −G−1 . (2.611)
lS λ2
Introducing the local Kubo number Kuλ corresponding to the selected scale λ yields
Vλ
Kuλ ≈ ∝ λG+M−2 . (2.612)
λωλ
Then, in terms of the Kubo number the expression for the stochastic instability in-
crement takes the form
1+G−D̃h (M)
γ̃S (λ) ∝ Kuλ 2−G−M
. (2.613)
As already mentioned, there is still no correct approximation for D̃h (M) in the whole
interval
−1/ν < M < 1. (2.614)
However, the exact value of D̃h is calculated for M = 0:
D̃h (0) = 3/2; (2.615)
166 O.G. Bakunin

in combination with the assumption of the ballistic character of particle motion along
percolation streamlines it could yield a scaling similar to the monoscale estimate.
Indeed, if
M = 0, D̃h = 3/2, and G = 1, (2.616)
we obtain [40, 148] 
Vλ 
γ̃S (λ) ∝ ∝ Kuλ . (2.617)
λωλ
Note that the passage from the monoscale percolation model to the multiscale one
makes it possible to describe essentially nondiffusive transport regimes in both iso-
tropic and anisotropic models. Therefore, it is only natural to apply the multiscale
description for other problems, where long-range correlation effects play a signifi-
cant role.

2.18.6 Isotropic and Anisotropic Turbulent Energy Spectra

The important aspect of turbulent transport is to establish a relation to spectral tur-

bulence characteristics. Thus, in the framework of the scaling approach, the most
universal is the Kolmogorov model with a constant energy flux over a spectrum [27].
In the case of isotropic hydrodynamical turbulence, the expression for the energetic
spectrum E(k) ≈ Vk2 /k can be obtained from the condition that the dissipation rate
is constant:
Vk2
εK = = const. (2.618)
τCASC
Here, Vk is the rate of turbulent pulsations corresponding to the wave number k ≈
1/λk and the characteristic time of nonlinear interaction τCASC ≈ (Vk k}−1 . Simple
calculations lead to the Kolmogorov spectrum E(k) ∝ 1/k 5/3 for three-dimensional
isotropic hydrodynamical turbulence. The multiscale model of percolation transport
considered in the previous sections is also based on the scaling representation for ve-
locity. However, the consideration of percolation transport channels system requires
the separation of characteristic scales in the framework of the hierarchy

λ0 ; λ1 = μλ0 ; λ2 = μ2 λ0 ; ··· λm ; μ  1. (2.619)

Analysis shows that the limit of applicability of the multiscale model is μ ≈ 2. On

the other hand, the multiscale percolation model is closely related to drift effects,
which, naturally, lead to a considerable difference between the percolation scaling
for velocity Vk ∝ (1/k)M−1 , where −1/ν < M < 1, and the expression for two-
dimensional isotropic turbulence [148]

Vk ≈ E(k)k ≈ 1/k, where E(k) ∝ 1/k 3 . (2.620)

The consideration of the multiscale transport model in the framework of MHD tur-
bulence is also closely related to the energetic spectrum form. Formally, the time of
 2 Correlations and Anomalous Transport Models 167

nonlinear interaction of waves is defined as

 m−1
Vp 1
τCASC ≈ , (2.621)
Vk ωp

where ωp = Vp k, Vp is the phase velocity and m characterizes the type of nonlinear

interaction (m = 3 for a three-wave interaction, m = 4 for a four-wave interac-
tion, etc.) The MHD turbulence model of Iroshnikov and Kraichnan [172, 173] for
Alfven’s waves is widely practiced now:
 2
VA 1 B0
τCASC ≈ , Vp ≡ VA = √  Vk , m = 3,
Vk VA k 4πn (2.622)
τA  τCASC .

Here, VA is the Alfven velocity and n is the plasma density. Calculations based on
the constancy of the dissipation rate εK ≈ Vk2 /τCASC yield E(k) ∝ 1/k 3/2 for the
turbulence spectrum. In terms of energy dissipation,

δE ≈ εK τA ≈ Vk3 /VA , (2.623)

the model from [172, 173] describes the diffusive character of energy transfer in
1/2
nonlinear wave interaction Vk2 ∝ τCASC :
 2 
Vk2 VA τCASC
≈ ≈ . (2.624)
δE Vk τA

Moreover, Kraichnan noted [40] that in the framework of the isotropic model of
MHD turbulence there is a linear dependence of the dissipation rate on the character-
istic time εK ∝ τA , whereas in the classical Kolmogorov approach εK ∝ 1/τCASC .
Naturally, the isotropic model is inadequate for describing turbulent transport
processes in a strong magnetic field. Indeed, in the case of strong MHD turbulence,
the separation of longitudinal and transverse scales plays an important role. The scal-
ing approach to the description of essentially nonisotropic MHD turbulence was sug-
gested by Goldreich and Sridhar [176]. It was based on the balance of characteristic
times in Alfven’s MHD turbulence
1 1
≈ k VA ≈ k⊥ V⊥ (k⊥ ) ≈ , (2.625)
τA τ⊥
where ωA = 1/τA is Alfven’s frequency, VA is Alfven’s velocity, k ≈ 1/ l is the
longitudinal wave number, k⊥/ ≈ 1/ l⊥ is the transverse wave number, V⊥ (k⊥ ) is the
scale of transverse velocity related to the spatial scale k⊥ , and τ⊥ is the dimensional
estimate of the time of nonlinear interaction that characterizes the turbulent cascade
in the transverse direction to the magnetic field with the constant dissipation rate

V⊥3
ε⊥ = = const. (2.626)
l⊥
168 O.G. Bakunin

This considerably differs from isotropic representations, which were used by Irosh-
nikov and Kreichnan [172, 173]. Expression (2.625) is, naturally, only an approxi-
mation analogous to the percolation renormalizations (2.531), but its efficiency has
been confirmed repeatedly by simulations, which demonstrate the correctness of the
scaling
1 VA l⊥ VA l⊥ VA 2/3
l ≈ ≈ ≈ 1/3
≈ 1/3 l⊥ . (2.627)
k V⊥ (ε⊥ l⊥ ) ε ⊥
The relationship between the longitudinal and transverse scales in the form l ∝ l⊥ α,

where α = 2/3, corresponds to strong MHD turbulence [176]. Note that the method
of separation of longitudinal and transverse spatial scales is also fairly effective in
analyzing multiscale percolation flows.

2.18.7 The Multiscale Model of Transport in a Tangled Magnetic Field

The incorporation of the complete hierarchy of spatial scales considerably improves

the agreement of theoretical predictions and observation results. Thus, the authors of
[54] obtained a correct estimate of electron heat-conductivity in galaxy clusters; the
assumption was made that the characteristic spatial scale lB in model [12] must be
calculated taking into account the anisotropic MHD turbulence spectrum. They used
the phenomenological model of Goldreich and Sridhar [176], where the longitudinal
and transverse scales are related by l ∝ l⊥α , where α = 2/3 corresponds to strong

MHD turbulence, and α = 4/3 corresponds to the intermediate turbulence regime.

This allows us to assume that the correlation scales characterizing transport in a
stochastic magnetic field have the form
 
LCOR Δ⊥ α
≈ . (2.628)
lB lB

This estimate differs significantly from the Chandran and Cowley estimate LCOR ≈
30lB and for transverse displacements Δ⊥ ≈ lB it leads to the value of longitudinal
correlation length
LCOR ≈ Δ⊥ ≈ lB  30lB . (2.629)
Then, the electron heat-conductivity estimate has the form
lB χSp
χeff ≈ D ≈ , (2.630)
LCOR 3
which agrees well with the data of astrophysical observations. To substantiate ob-
tained scalings, let us consider the model equation describing the separation of ini-
tially close force lines. The approximation equation describing the exponential di-
vergence of force lines Δm could take the form

d 2 Δ2
Δm ∝ m . (2.631)
dl lB
 2 Correlations and Anomalous Transport Models 169

However, the exponential regime for Δ2m > lB2 has to move into the diffusive one,
dΔ2m l2
≈ Dm ≈ ⊥ . (2.632)
dl lB
From this standpoint, it is natural to describe intermediate regimes by the modifi-
cation of the factor 1/ lB in (2.631), taking into account the increasing role of large
scales.
Considering the scale hierarchy corresponding to the model of a quasi-isotropic
stochastic magnetic field, the authors of [54] used the scaling from the model of
strong Alfven’s turbulence [176]
 α
l l⊥
≈ , (2.633)
lB lB
with the hierarchy of scales
lmin < l⊥ < lB ≤ LCOR . (2.634)
Then, in terms of the wave numbers k⊥ ≈ l1⊥ and k ≈ l1 , it is convenient to rewrite

model equation (2.631) in a form that takes into account the contribution of different
scales of hierarchy in the representation of LK ≈ lB :
  1/ lmin
dΔ2m   2  1/Δ⊥ k (k⊥ )
≈ Δm k (k⊥ ) d ln k⊥ + 2
d ln k⊥ , (2.635)
dl 1/ lB 1/Δ⊥ k⊥
1/α
where k ∝ k⊥ ; hence, this equation describes transition to regimes with
Δα⊥ ≈ l⊥
α
≈ l . (2.636)
It is important to note that in spite of using the multiscale approach relating lon-
gitudinal and transverse scales in the strong Alfven’s turbulence, the characteristic
spatial scale lB of a tangled magnetic field appears to be the universal parameter of
model
lB ≈ LCOR ≈ LK ≈ Δ⊥ , (2.637)
and the characteristic correlation scale does not enter the final expression for χeff .

2.19 Subdiffusion and Traps

In the presence of structures, it is possible that increasing effective transport owing
to convective flows and trapping lead to subdiffusion. In this section we consider
some simple approximations of trapping effects that allow us to obtain estimates of
the effective diffusion coefficient in the presence of spatially distributed traps, and
on comb structures. In the models of turbulent diffusion, the subdiffusion character
of transport can be related to the presence of compressibility effects and the “trap”
character of interaction of a passive tracer with vortex structures. Note that with-
out compressibility in two-dimensional flows the subdiffusion mechanism cannot be
realized.
170 O.G. Bakunin

2.19.1 The Balagurov and Vaks Model of Diffusion with Traps

One of the simplest methods of approximation of compressibility effects in the anom-

alous transport models relates to using different “trap” mechanisms. We will analyze
a particular physical model of diffusion in a medium with traps on the basis of the
well-known paper by Balagurov and Vaks [46]. Using theoretical probabilistic es-
timates, we can derive scaling relations, which can be interpreted on the basis of
relaxation functions. On long time scales, the diffusion of particles in a medium with
traps is governed by the fluctuating character of the appearance and disappearance
of regions free of traps. We introduce the trapping probability (i.e., the probability of
a particle being captured in a trap) in terms of the Poisson distribution:
 
t
Pc ∝ exp − . (2.638)
τD

Here, τD ≈ R 2 /D is the characteristic time scale on which the particle diffuses

through the medium until it reaches the boundary of the trap-free region of the radius
R and D is the local diffusion coefficient. We assume that trap-free regions obey the
Poisson distribution,
  d 
R
PT ∝ exp − , (2.639)
R0
where R0 is the mean radius of the trap-free regions in the space of dimensionality d.
Now, we can estimate how the radius R(t)of the trap-free region should change in
time in order for the survival probability to be the highest:
  d−1 
R t
P = Pc PT ∝ exp − − . (2.640)
R0 τD
Calculating the time derivative of the argument of the exponential functions in this
expression, we obtain
R(t) ∝ t 1/(2+d) . (2.641)
For d > 0, the diffusion described by this scaling is obviously slower than that
described by the classical diffusion scaling [170]

R 2 (t) ∝ t 2/(2+d)  t. (2.642)

In the language of fractional derivatives, this indicates that a fractional derivative

with respect to time can serve as a model equation for describing diffusion in the
situation at hand:
 t
∂ γ ñ(k, t) ∂ ñ(k, τ ) 2
= dτ, γ = . (2.643)
∂t γ ∂t 0 (t − τ )γ 2+d
Note that the problem of diffusion in a medium with traps is not necessarily related
to such issues as condensed states or chemical reactions. The ideas associated with
traps are also used in the study of particle trapping by vortices in order to describe the
 2 Correlations and Anomalous Transport Models 171

behavior of a passive scalar in a turbulent field [17, 18, 158, 169] or in the analysis
of the correlation functions of a turbulent field. In what follows, we will consider the
problem in which traps in a medium manifest themselves in particle diffusion in a
magnetic field with “braided” force lines.

2.19.2 Subdiffusion and Fractality

√
The diffusion approximation λ ∝ 2Dt of longitudinal motion in double diffu-
sion and the Dreizin–Dykhne models lead us to expect that the subdiffusive char-
acter of transport can also be interpreted by means of the “renormalization” that is
analogous to t → tδN/N. Indeed, the diffusive character of motion in the one-
dimensional case leads to numerous “returns”, hence the particle takes part in the
transverse diffusion process only a fraction of the total time t. Let us assume the
fraction of the time during which the particle takes part in the transverse diffusion
has a fractal character with the dimensionality dF . Actually, we can use the repre-
sentation
δN
Δ2⊥ ≈ D⊥ t , (2.644)
N
where δN ∝ (t/τ )dF , N ∝ t/τ . Here, τ is the longitudinal correlation time
and D⊥ is the transverse diffusion coefficient. Simple calculations yield the expres-
sion  
D⊥
Δ⊥ ≈
2
t dF . (2.645)
(τ )dF −1
Changing to the symbols D⊥ ≈ Dm L /τ⊥ , D0 ≈ L2 /τ used above, we find
that   
D0 τ t dF
Δ⊥ ≈ Dm
2
. (2.646)
τ⊥ τ
In the case of “double” diffusion dF = 1/2, we obtain the expression for double dif-
fusion (2.117). Note that the cases where dF = d/2 are very frequent. For example,
in the percolation models of turbulent diffusion [17], the fraction of the percolation
streamlines is
a dF 1
P∞ ≈ 2 ≈ , dF = d/2 = 1. (2.647)
a a
The diffusion equation corresponding to the specific fractional value dF has the form
of an equation with fractional derivatives [18–22]. Indeed, based on the definition of
the Hurst exponent H , one obtains

Δ⊥ (t) ∝ t H , H = dF /2. (2.648)

The rigorous theory for such problems is based on continuum time random walks
(CTRW) or on the diffusive models for a medium with “traps” [18–22]. To describe
the transport in a stochastic magnetic field, this approach was used in [20] very suc-
cessfully. The main assumption in that approach is the consideration of the probabil-
ity Φ(t) of a particle remaining in a trap during the time t. The following function in
172 O.G. Bakunin

Fig. 2.14. Comb-like structure

the scaling law form was usually used:

1
Φ(τ ) ∝ , (2.649)
τγ
where γ = dF for γ < 1. Here, γ is the parameter of the probability function. There
exists a simple estimate for Φ(t) based on the probability of the return (2.104) given
as
Δ
Φ(t) ∝ ρ(0, t)Δ ∝ √ . (2.650)
4πD0 t
Here, Δ is the correlation scale. We obtain the relation between the probability of
finding the particle in the trap and the “return” probability. We can imagine the
“traps” as teeth in the “comb-like structure” (see Fig. 2.14). If the teeth length tends
to infinity, then our expression (2.650) is satisfactory. The models with teeth, which
have a limited length and number of teeth and which have fractal structures, appear
to be the natural generalization of the “comb-like” model [18–20].

2.19.3 Comb Structures and Transport

Anomalous transport on comb structures plays an important role due to the fairly
universal kind of trap topology. Therefore, we will consider a more general comb
structure model, where the length of the “teeth” is distributed in accordance with
power laws. We will consider the structure that consists of a backbone directed along
the z-axis and orthogonal teeth connected to this backbone. The distances between
the teeth Δ are identical; however the tooth distribution along the length is given by
the scaling
γ −1
l0 S
f (l) = . (2.651)
l
Here, f (l) is the probability density of finding a tooth with the length l; l0 is the
tooth characteristic length, and γS is the characteristic exponent. In the model under
analysis, the minimum tooth length l0 must be of order the characteristic longitudinal
scale Δ,
l0 ≈ Δ. (2.652)
 2 Correlations and Anomalous Transport Models 173

On the other hand, it is necessary to take into account the renormalization condition
of the distribution function f (l):
 ∞
f (l) dl = 1, (2.653)
l0
 ∞
f (l)l n dl = ∞, (2.654)
l0
where n = 1, 2, 3, . . . . In such a formulation of the problem, the description of
anomalous transport on comb structures was repeatedly discussed in [171–173].
Therefore, here we develop only qualitative estimates.
Following Lubashevskiy and Zemlianov [171], let us define the mean length l
that describes the spatial scale corresponding to diffusive walks along teeth with the
diffusion coefficient D0 :
 √D0 t  γS
  D0 t l0 1
l(t) lf (l) dl ≈ √ ∝ γ /2 . (2.655)
l0 l 0 D 0 t t S

The mean time T  that the particle diffuses along a tooth can be estimated as
 
Δ l0 D0 t γS /2
T  ∝ t∝ ∝ t γS /2 . (2.656)
l D0 l02
Then, the mean-squared displacement is estimated by the scaling
 
  D0 t γS /2
R 2 = x 2 (t) ≈ D0 T  ≈ l0 ∝ t γS /2 . (2.657)
l02
On the other hand, one can obtain the estimate of the probability of avoiding a return
to the backbone Φ(t):
T  1
Φ(t) ≈ ∝ γ /2 . (2.658)
t t S
Note that in the case of comb structures with the power form of length distribution,
the expression for Φ(t) differs essentially from the diffusive estimate for regular
comb structures:
1
Φ(t) ∝ 1/2 . (2.659)
t
The consideration of the probability Φ(t) in the framework of the more rigorous
approach (continuous time random walk) makes it possible to obtain not only scal-
ings for the mean-squared displacement but also fractional differential equations to
describe particle density. This approach will be developed in the next section.

2.20 Continuous Time Random Walks

The nonlocal nature of turbulent diffusion has stimulated the search for equations that
differ significantly from the conventional diffusive representation. An elegant inte-
174 O.G. Bakunin

gral equation corresponding to this problem was suggested by Einstein and Smolu-
chowski. However, memory effects were not included in this equation. To describe
trapping and subdiffusive regimes, the continuous time random walk model was sug-
gested [174, 175]. It is possible to combine both these approaches to describe the
anomalous transport, where memory and nonlocality effects play important roles.

2.20.1 The Montroll and Weiss Approach and Memory Effects

A careful analysis of the problems involving random-walk processes shows that

a fundamentally important role is played by the transition probability density. In
Markov’s approach, the transition probability density is assumed to depend on the
spatial variable, W (Δ), where Δ is a spatial step. Montroll and Weiss [174] used
a fundamentally different dependence—they assumed that the transition probability
density depends on time, ψ(t). They also introduced a physically clear quantity—the
probability of not undergoing a transition from a point y to any other points during a
time t:  t
ΦY (t) = 1 − ψY (t) dt. (2.660)
0
The subscript Y in the functions Φ and ψ served merely to mark an arbitrarily cho-
sen point. The function Φ(t) reflects the relaxation properties of the system. In the
simplest case, the function Φ(t) is represented in the form of the Poisson distribution

Φ(t) = exp(−t/τ ), (2.661)

where τ is the mean time between transition events. The function Φ(t) can also be
represented in some other forms capable of reflecting the characteristic behavior of
relaxation character of systems:
√
Kohlrausch relaxation function Φ(t) = exp(− αt), (2.662)

algebraic relaxation function Φ(t) = (αt)−γ , (2.663)


Montroll function Φ(t) = exp − lnβ [−αt] . (2.664)
Here α, β, and γ are the characteristic parameters of the problem. In the Montroll–
Weiss theory, the function Φ(t) plays a governing role. The authors of [174] suc-
ceeded in writing an elegant chain equation for the probability of a randomly walking
particle to be at the point x at time t:
 t
P (x, t) = Rp (x, τ )Φ(t − τ ) dτ. (2.665)
0

Here, Rp (x, t) is the probability of transitions from other points to the point x during
the time interval (t; t +dt). The functions on the right-hand side of this equation have
an essentially similar physical meaning as the function ψ [see definition (2.660)].
Consider the point x1 and let the function ψX1 be represented as a sum of the prob-
ability densities for transitions from the point x1 to all allowed points x. Then, we
 2 Correlations and Anomalous Transport Models 175

have ψX1 (t) = ψ(x1 → x, t) and, consequently,
X1

 t
Rp (x, t) = Rp (x1 , τ )ψ(x1 → x, t − τ ) dτ + P (x, 0)δ(t), (2.666)
X1 0

t
where 0 ψ(x1 → x, t) is the probability density for a transition from the point x1 to
the point x at the time t. Note that the function ψ depends not only on the time t but
also on the relative spatial positions of the points x1 and x, in which case we have
 ∞  t
dx ψ(x1 → x, t) dt = 1. (2.667)
−∞ 0

The probability density P (x, t) is related to the particle density by

N P (x, t)
n(x, t) = , (2.668)
δx
where N is the total number of particles and δx is a volume element. Applying
the Laplace transformation in time and the convolution theorem, we obtain from
expressions (2.310) and (2.666)
 
s P̃ (x, s) − P (x, 0) = R̃p (x1 , s) − R̃p (x, s) ψ̃(x1 → x, s), (2.669)
X1

where, in accordance with expression (2.310), P̃ and R̃p are related by

P̃ (x, s) s P̃ (x, s)
R̃p (x, s) = =  . (2.670)
Φ̃(s) 1− X1 ψ̃(x1 → x, s)

Then, returning to the physical variables, we arrive at the Montroll–Weiss equation

∂   t 
P (x, t) = P (x1 , τ ) − P (x, τ ) F (x1 → x, t − τ ) dτ, (2.671)
∂t 0
X1

where the memory function F is defined in terms of its Laplace transform,

s ψ̃(x1 → x, s)
F̃ (x1 → x, s) =  . (2.672)
1 − X1 ψ̃(x1 → x, s)

In what follows, we will be interested in the functions that depend only on the
difference between x and x1 ; this corresponds to the case of a uniform medium
F (x1 → x, s) = F (x − x1 , s).
Assuming that the variable x takes on continuous values, we can generalize
(2.671) to a sort of the Smoluchowski–Chapman–Kolmogorov equation (2.50) with
memory effects:
 ∞  t
∂
P (x, t) = dx1 dτ P (x1 , τ )F (x − x1 , t − τ ) + Q, (2.673)
∂t −∞ 0
176 O.G. Bakunin

where Q is expressed in terms of Laplace transforms as follows:

Q̃(x, s) = P (x, 0) − P̃ (x, s)F̃ (x − x1 , s). (2.674)

The assumption that the memory function is of a multiplicative nature yields

F (x − x1 , t − τ ) = G(x − x1 )MB (t − τ ). (2.675)

Here, G corresponds to the kernel of the Chapman–Kolmogorov functional and MB

is the memory function. Switching now to Fourier transforms in x and Laplace trans-
forms in t, we arrive at the following equation for the particle density:

˜ s) − ñ(k, 0) = − s ψ̃(s) 
˜ s).
s ñ(k, 1 − G̃(k) ñ(k, (2.676)
1 − ψ̃(s)

˜ s) is both the Fourier and Laplace transformation of the density n(x, t).
Here, ñ(k,
It is easy to draw an analogy between this equation and the Einstein functional equa-
tion. Obviously, under the conditions MB (t) ∝ δ(t), M̃B (s) = const the Montroll–
Weiss equation passes over to the Smoluchowski–Chapman–Kolmogorov equation.
In fact, choosing the Poisson distribution for the function Φ(t) ensures the required
limiting transition for the equation with memory effects. Telegraph equation (2.229)
was derived for an exponential memory function MB (t) = exp(−t/τ ) and for a
Gaussian memory function with 1 − G̃(k) = −Dk 2 . It is of interest to note that,
although the equation considered above and the memory function MB (t) both have
a simple form, the expression for Φ(t) is fairly complicated in structure [18, 19].
Hence, we see that it is necessary to choose different model functions for different
physical situations.
There are numerous investigations of the continuous time random walk models.
Fortunately, several detail reviews [18–22] have been published recently. Therefore,
in the next sections we will consider a few examples closely related to the renormal-
ization of quasi-linear equations.

2.20.2 Fractional Differential Equations

An important physical quantity in the description of random walk processes with

memory is the mean waiting time t until an event occurs:
 ∞
t = tψ(t) dt. (2.677)
0

This time is analogous to the mean length of the jump in the theory of Markovian
processes. This is not surprising because, in the approach based on memory effects,
the transition probability density Φ(t) is an analogue of the function W (y). For the
Poisson distribution (2.661), we have t = τ .
 2 Correlations and Anomalous Transport Models 177

An important particular case of relaxation functions is represented by those that

decrease according to a power law,
 γ
τ
Φ(t) ∝ , 0 < γ < 1. (2.678)
t
In this case, the mean waiting time until an event occurs tends to infinity:
 ∞
t = tψ(t) dt → ∞. (2.679)
0

The power relaxation functions were found to provide an efficient tool for analysis
of transport processes [21, 22]. For long times t, simple manipulations yield the
following expression for M̃B (s):

s Φ̃(s)
M̃B (s) = ≈ s Φ̃(s) = (1 − γ )s γ , (2.680)
1 − s Φ̃(s)
where (z) is Euler’s gamma function. The equation describing memory effects
takes the form

˜ s) = 1 − G̃(k) ñ(k,
s Φ̃(s)ñ(k, ˜ s) + ñ(k, 0)Φ̃(s). (2.681)

The expression
˜ s) ≈ s γ ñ(k,
s Φ̃(s)ñ(k, ˜ s) (2.682)
can be interpreted as a time derivative of order γ [21, 22, 174–176]:
 t
˜ γ ˜ ∂ γ ñ(k, t) ∂ ñ(k, τ )
s Φ̃(s)ñ(k, s) ≈ s ñ(k, s) → = dτ. (2.683)
∂t γ ∂t 0 (t − τ )γ
Representing the results in such a manner facilitates interpretation of scaling rela-
tions of the form γ
R(t) ∝ t αL , (2.684)
which have found increasingly wider application in the analysis of fractional deriva-
tives:
∂ γ n(x, t) ∂ αL n(x, t)
= + Q(n, x, t). (2.685)
∂t γ ∂x αL
Here, αL , γ are the parameters and Q is the additional term related to the initial
conditions.

2.20.3 The Taylor Definition and Memory Effects

The approximations of the function ψ(x1 → x; t), which were used in the previous
sections, allow one to investigate the models in which it is easy to use the decoupling

ψ(x, t) = ϕ(x)ψ(t). (2.686)

178 O.G. Bakunin

However, many important results can be obtained in the framework of a more general
analysis. Thus, consider the system of the Montroll–Weiss equations in the form
 t
P (x, t) = P (x1 , τ )ψ(x − x1 , t − τ ) dτ + Φ(t)δ(x), (2.687)
x1 0


ψ(t) = ψ(x; t) = ψ(k = 0; t). (2.688)
x
Then, upon applying both the Laplace and Fourier transformations one obtains

P̃˜ (k, s) = P̃˜ (k, s)ψ̃(k,

˜ s) + Φ(s), (2.689)

1 − ψ̃(s)
Φ(s) = . (2.690)
s
Formal calculations yield the expression for P̃˜ (k, s) in the form

1 − ψ̃(s)
P̃˜ (k, s) =
1
. (2.691)
s ˜
1 − ψ̃(k, s)
Following the approach developed in [177] it is easy to obtain a formal expression
for the mean squared displacement:

  ∂ 2 P̃ (k, t)
R 2 ≡ x 2 (t) = x 2 P (x, t) dx = − . (2.692)
∂k 2 k=0
On the other hand, the expression suggested by Taylor (2.6) relates the mean square
displacement to the correlation function
 t
dR 2
=2 C(τ ) dτ . (2.693)
dt 0

The authors of [176] used the Fourier transformation of

C̃(s)
s R̃ 2 (s) = 2 (2.694)
s
in order to establish the relationship between the Lagrangian correlation function
and the memory function. The comparison between expressions (2.692) and (2.694)
yields this relationship in terms of the Laplace transform

ψ̃(s)
C̃(s) = ζ 2 s , (2.695)
1 − ψ̃(s)
where we used the Gaussian approximation for the step length distribution
˜ 
ψ̃(k, s) = ψ̃(s) exp −ζ 2 k 2 . (2.696)
 2 Correlations and Anomalous Transport Models 179

It is well known that the exponential correlation function corresponds to the conven-
tional Brownian motion. Therefore, the correctness of the formula obtained in [177]
must be first proved for ψ(t) = δ(t − τ ). Then, substituting ψ̃(s) = exp(−sτ ) it is
easy to find
s
C̃(s) = ζ 2 sτ . (2.697)
e −1
To compute the Laplace inverse of this expression one keeps the three lowest orders
of the exponential as sτ → 0 (the observation time is much lower than the micro-
scopic waiting time: t  τ ) and obtains
 2  
2ζ t
C(t) = exp −2 . (2.698)
τ2 τ
This result corresponds exactly to the Langevin equation
2 ζ2
V̇ (t) = − V + ξ(t), (2.699)
τ τ
where V (t) is the Brownian particle velocity and ξ(t) is the white noise random
function.
Now we are able to consider more complex models, where the memory function
is represented by the stable law

ψ̃(s) = exp −(sτ )γ (2.700)
with the parameter σ ; 0 < γ < 1. Simple calculations in the framework of continu-
ous time random walks allow the following expression to be obtained:
ζ2 tγ
R2 = 2 , (2.701)
τ σ Γ (1 + γ )
where Γ is the Gamma function symbol. The expression for the Laplace transform
of the velocity correlation function takes the form
ζ 2 1−γ s 1−γ
C̃(s) = s , (2.702)
τ [1 + (τ γ s γ /2)]
where the three lowest orders in sτ were kept to compute exp[(sτ )γ ]. Formal calcu-
lations [177] yield the final expression for the correlation function in the form

 γ 
2 ζ 2 2γ −2 t
C(t) = γ t Eγ ,2γ −1 −2 , (2.703)
τ τ τ
where 1/2 < γ < 1 since we restrict the parameter γ in order to express C(t) in
terms of the Mittag–Leffler functions Eγ ,2γ −1 . This result leads to the power tail
asymptotic
1
C(t) ∝ 2−γ . (2.704)
t
This agrees well with the simple estimates based on the transport scaling R 2 ∝ t γ ,
since C(t) ≈ R 2 /t 2 ≈ t γ −2 . Hence, the method suggested in [177] is effective for
analyzing the correlation properties of the system in which memory effects play an
important role.
180 O.G. Bakunin

2.21 Fractional Differential Equations and Scalings

A fractional differential equation is an especially effective tool for investigating
anomalous transport. These equations allow us to obtain scalar probability den-
sity functions based on scaling representation of waiting time distributions. On the
other hand, the consideration of correlation effects in the framework of renormalized
quasi-linear equations can also lead to the appearance of fractional derivatives. In this
section, we consider the Richardson law, trapping in vortex structures, transport in a
stochastic magnetic field, and shear flow systems in terms of fractional differential
equations.

2.21.1 The Klafter, Blumen, and Shlesinger Approximation

Applying the continuous time random approach makes it possible to consider both
nonlocality and memory effects by the approximation of model function ψ(x, t).
Blumen, Klafter, and Shlesinger [142] suggested using the advantages of this method
to describe the Richardson relative diffusion R 3 ∝ t 3 . Recall that in the Monin model
[62] use was made only of the dimensional estimate of nonlocality effects by the
power approximation of the kernel of the Einstein–Smoluchowski functional (2.52)
1/3
G(k) ∝ εk k 2/3 . (2.705)

The authors of [142, 178] suggested the dynamical interpretation of nonlocality and
memory effects by using the model function ψ(x, t) in the form
 
x
ψ(x, t) = ϕ(x)ψ(t/x) = ϕ(x)δ t − , (2.706)
V (x)

where, through the δ function, x and t are coupled. Here, the functions ϕ(x) and
V (x) are represented by the following scalings:
1
ϕ(x) ∝ , (2.707)
x 1+βR
V (x) ∝ x γR . (2.708)
Then, the case γR = 0 corresponds to the ballistic model and the case γR = 1/3
corresponds to the Kolmogorov scaling
1
Vk2 ∝ E(k)k ≈ . (2.709)
k 2/3
Then, for γR = 1/3, using the expression for mean-squared displacement in the form

∂2
R 2 = − 2 P̃ (k, t) , (2.710)
∂k k=0
 2 Correlations and Anomalous Transport Models 181

where the expression for P̃ (k, t) is given by the formula (2.691), the authors of [142]
obtained the following relationships:
1
R2 ∝ t 3 for βR ≤ ; (2.711)
3
2 1 1
R 2 ∝ t 2+ 3 (1−βR ) ≤ βR ≤ ;
for (2.712)
3 2
1
R 2 ∝ t for βR ≥ . (2.713)
2
One can see that for βR ≤ 1/3 the Richardson law R 2 ∝ t 3 is obtained, as we
could anticipate. Moreover, in [142] the modified model was considered, where the
intermittency effects are included:
 
1 d − dF 1 μF
V (x) ∝ x , γR = +
γR
= 1+ . (2.714)
3 6 3 2
Then, the mean-squared separation of two particles is given by
12 1 − μF
R 2 ∝ t 4−μF , βR ≤ ; (2.715)
3
1−β
2+6 4−μR 1 − μF 10 − μF
R2 ∝ t F , ≤ βR ≤ ; (2.716)
3 6
10 − μF
R 2 ∝ t, ≤ βR . (2.717)
6
Here, the scaling exponents depend on the index βR as well as the fractal dimension
dF = d − μF . Note that the scaling for
1 − μF
βR ≤ (2.718)
3
was previously obtained by Hentschel and Procaccia [78], who used a much different
approach.

2.21.2 The Stochastic Magnetic Field and Balescu Approach

Fractional differential equations are especially relevant for analyzing different anom-
alous transport mechanisms. Since double diffusion corresponds to the subdiffusion
character of transport with H = 1/4, the corresponding equation could be in agree-
ment with the continuous time random walk approach. This model was considered
by Balescu [179]. The authors of [179–189] investigated the transport effects in the
stochastic magnetic field in terms of the correlation function. Here, we represent this
approach in the simplified form, which is related to earlier discussions on the renor-
malization of quasi-linear equations (2.213) and (2.214). Instead of introducing the
additional diffusion term D(∂ 2 n1 /∂x 2 ) that describes transverse correlation effects,
the author of [179] kept one of the usually omitted terms v1 (∂n1 /∂x), which allows
the memory effects to be described. As a result, the transformations put (2.214) for n1
182 O.G. Bakunin

into the form

∂n1 ∂n1 ∂n0
+ v1 = −v1 . (2.719)
∂t ∂x ∂x
This equation can be considered as a first-order linear hyperbolic equation with the
source term
∂n0
I (x, t) = −v1 , (2.720)
∂x
where the derivative ∂n0 /∂x is the parameter of the problem. We also supplement the
equation with the uniform initial condition n1 (x, 0) = 0. This formulation is similar
to (2.215) but here we deal with the characteristic

z = x − v1 (t − t1 ) (2.721)

instead of the characteristic of the conventional quasi-linear approach z = x −

v0 (t − t1 ),  t
∂ n (z, t)
n1 (x, t) = − v1 (t1 ) 0 dt1 . (2.722)
0 ∂z
We substitute this expression for n1 into (2.213) and perform simple manipulations
to obtain  t
∂n0   ∂ 2 n0 (z, t1 )
= v1 (t)v1 (t1 ) dt1 . (2.723)
∂t 0 ∂z∂x
Now one can see that the use of the correlation function in the power form

C(t) = MB (t) ∝ (τ/t)α (2.724)

leads to the continuous time random walk representation for the transport equation
 t 2  αC
∂ 2 n0 (x, t) D ∂ ∂ τ
=− n0 (x, τ ) dτ. (2.725)
∂t 2 τ ∂t 0 ∂x 2 t −τ
Here, D0 is the seed diffusivity, τ is the characteristic time, and αC is the correla-
tion exponent. Actually, the renormalized quasi-linear equations, together with the
approximation of the correlation function, make it possible to obtain the fractional
differential equation  
∂ 2n D0 ∂ αC ∂ 2 n
= − , (2.726)
∂t 2 τ ∂t αC ∂x 2
which corresponds to the continuous time random walk with the scaling for the Hurst
exponent
R 2 (t) ∝ t γ = t 2−αC , (2.727)
γ αC
H = =1− , αC < 2. (2.728)
2 2
However, in this simplified approach we do not consider the aniso tropy effects and
longitudinal collisional diffusion, which are essential for describing the anomalous
transport in the stochastic magnetic field. Balescu et al. [179, 180] considered all
these aspects in detail. The final equation for the transverse transport with the modi-
 2 Correlations and Anomalous Transport Models 183

fied memory function  

D0 τ αc
MB ≈ D0 δ(t) + (2.729)
τ t
takes the form
  αC
∂n0 (x, t) ∂ 2 n0 D0 t ∂2  τ
= −D 0 − n0 (x, t )  dt  , (2.730)
∂t 2 ∂x 2 τ 0 ∂x 2 t −t
which allows the investigation of both conventional diffusion and subdiffusive be-
havior. Note that the authors of [179–189] carried out a thorough analysis of the
correlation function. They interpreted different regimes of the anomalous transport
in the stochastic magnetic field on the basis of the renormalized quasi-linear equa-
tions with
H = 1 − αC /2, (2.731)
where the case with αC = 3/2 corresponds to the double diffusion (2.117).

2.21.3 Longitudinal Correlations and the Diffusive Approximation

The Kadomtsev–Pogutse analysis [67] was based on the ballistic representation of

longitudinal (z-axis) motion and the diffusion approximation Deff ∇⊥ 2 n of trans-
1
verse correlation effects. In fact, the opposite case corresponds to the shear flow
model (2.177) and the double
√ diffusion (2.114) where longitudinal motions have the
diffusion character λ ≈ 2D0 t. From this standpoint, it is easy to obtain an equa-
tion for the passive tracer density under conditions when longitudinal correlation
effects can be approximated by the longitudinal diffusion term D0 ∂ 2 n1 /∂z2 [181].
Thus, in the two-dimensional case the corresponding renormalized equations have
the form  
∂n0 ∂n1
= − VX (z) ; (2.732)
∂t ∂x
∂n1 ∂ 2 n1 ∂n0
= D0 2 − VX (z) . (2.733)
∂t ∂z ∂x
Here, in contrast to [67], the diffusion coefficient D0 characterizes the seed diffusion.
The dependences n0 = n0 (x, t) and n1 = n1 (x, z, t) were used to describe the two-
dimensional case. Indeed, the author of [182] obtained a similar set of equations by
averaging the diffusion equation with the random convective term
∂n ∂n
= D0 n − VX (z) . (2.734)
∂t ∂x
Using the Laplace transformation over t and the Fourier transformation over z, one
obtains from (2.732) and (2.733)

∂ 2 ñ0
s ñ0 (s, x) − n0 (x, 0) = D̃(s) , (2.735)
∂x 2
184 O.G. Bakunin
    
1 L0 ∞ exp[− s|z − z |2 /D0 ]
 
D̃(s) = lim dz dz √ VX (z)VX (z ) .
L0 →∞ 2L0 −L0 −∞ D0 s
(2.736)

The aim of the author of [182] was to obtain a diffusion equation for the model of
random drift flows [72]. This corresponds to the condition z → z (the condition
of “return” of the particle to the initial point). A fractional differential equation was
found,

∂ 3/2 n0 (t, x) ∂2 t n0 (t  , x) dt  V02 a ∂ 2 n0 (t, x) n0 (0, x)
= √ 
= √ − √ 3/2 . (2.737)
∂t 3/2 ∂t 2 0 π(t − t ) 2D 0 ∂x 2 2 πt

Indeed, the “renormalization” of the quasi-linear equations allows us to obtain the

transport equations, which differ significantly from the classical diffusion equation
[20–22]. Now, we will consider using the correlation properties of a system of ran-
dom flows (see Fig. 2.15). The correlation function can be represented in the power
form. We can change the form of the equation for n0 by means of a more detailed
consideration of the function K(|z − z |) = VX (z)VX (z ) in (2.736). This function
describes the correlation properties. The question of the correlation nature of this
function was discussed in detail in [182]. However, the choice of the form of the
function K was not important there because Dreizin and Dykhne considered the
model taking into account the return effects (z → z ). Let us consider the power
approximation of the function K(w),

 V02
K(w) = K |z − z | ∝ . (2.738)
1 + w αC
The power approximations of the correlation function are often used for obtaining the
scaling law [17–22]. Using the symmetry of integral (2.736) and the known formula
of Kampe-de-Feriet [25–27], we can easily simplify integral (2.736). In terms of

Fig. 2.15. Zonal flows system

 2 Correlations and Anomalous Transport Models 185

the Laplace transformation, the equation that is determined takes the form
 αC /2−1 2
V02 s ∂ ñ0
s ñ0 (s, x) − n0 (x, 0) = √ . (2.739)
2D 0 2 ∂x 2
Changing to the dependence in time, we obtain the subdiffusion equation [17–22]
∂ γ n0 ∂ 2 n0
= D eff − Q(t, x), (2.740)
∂t γ ∂x 2
V02 a αC n0 (0, x)
Deff = , Q= √ . (2.741)
(2D0 )αC /2 2 πtγ
Here, the order of the derivative with respect to time γ depends on the parameter αC
(2.740),
αC
γ = 2H = 2 − , 0 < αC < 4, (2.742)
2
which describes correlation properties in the longitudinal direction.
In the case of an anisotropic medium this relationship can be related to the
Corrsin functional in the form (2.179)
 ∞  
λ2⊥ V02 z2 dz
C(t) ≈ 2 ≈ exp − √ . (2.743)
−∞ 1 + (z/z0 )
t αC 4D0 t 4πD0 t
Here, V0 and z0 are the dimensional parameters of the correlation function. More-
over, it is possible to recognize the Corrsin integral in the expression for the Laplace
transformation (2.736). Using the dimensionless variable z2 /4D0 t in order to calcu-
late the integral, we obtain the simple estimate in the scaling form

λ2⊥ const
C(t) ≈ 2
≈ α /2 and λ⊥ ∝ t 1−αC /4 . (2.744)
t t C
Actually, this is scaling law (2.742). Note that the regime with αC = 1 (H = 3/4),
which was considered in [72, 120], has a clear physical interpretation in terms of the
spectrum Sc (k) [120]:
 
Ṽx (k)Ṽx (k  ) = Sc (k)δ(k − k  ). (2.745)

Here, Ṽx (k) is the Fourier representation of Vx (z) and δ(k − k  ) is the Dirac func-
tion. For k  1 and the power form of Sc (k) ≈ k αC −1 the regime with H = 3/4
corresponds to a shear velocity field, which is white noise.

2.21.4 Vortex Structures and Trapping

Cardoso et al. [183, 184] carried out experimental research on anomalous diffusion
of scalars in the field of several vortices. Actually, the authors of [183, 184] studied
trapping effects related to the capture of test particles by vortices (see Fig. 2.16),
which plays an important role. Naturally, in such a formulation of the problem con-
tinuous time random walks are an adequate theoretical model.
186 O.G. Bakunin

Fig. 2.16. Vortex trapping regions

From the formal standpoint we must introduce a scaling to describe the waiting
time distribution
1
ψ(t) ∝ γ −1 . (2.746)
t
Then, the mean waiting time between two “jumps” is estimated by the expression
 tmax
3−γ
τT ≈ tψ(t) dt ∝ tmax ≈ t 3−γ . (2.747)
0

On the other hand, the full walking time can be estimated through τT and the number
of pauses N,
t ≈ N τT ≈ t 3−γ N. (2.748)
Hence, one obtains the scaling for N in the form:

N ∝ t γ −2 . (2.749)

Since the mean squared displacement is related to N by the scaling R 2 ∝ N , one

arrives at the relationship
√ γ /2−1
R∝ N ≈t . (2.750)
The experimental results obtained by Cardose et al. allow one to define scaling for
the mean squared displacement in the following form

R ∝ t 1/3 , (2.751)

where the Hurst exponent H = 1/3 and the exponent γ = 8/3. On the other hand,
the probability density function ψ(t) corresponding to these scalings was success-
fully measured [183, 184], which permits us to consider the continuous time random
walk prediction as correct.
 2 Correlations and Anomalous Transport Models 187

Naturally, the subdiffusive regime that was experimentally found is not the only
one possible. In the presence of vortex structures, there are both trapping effects and
flights. Thus, the authors of [185, 186] experimentally investigated test particle trans-
port in an almost two-dimensional flow in an annular tank (“Texas experiments”).
The tank was rotating at about 1 or 2 Hz and the bottom was sloped to simulate
β-effects. Because of the rapid rotation, the flow was quasi-two-dimensional. Two
types of particle trajectories were found. Particles within a vortex remain trapped
for very long time (stick). Particles in the azimuthal jet experience prolonged flights
around the circumference of the tank. Because the vortex pattern is not perfectly sta-
tionary particles alternate, apparently randomly, between flying in jets and sticking
in vortices. The authors of [185, 186] obtained superdiffusive scaling for the angular
displacement  2
θ ∝ t γ , (2.752)
where γ ≈ 1.4–1.7. It was also possible to observe conventional diffusion with
γ = 1 by breaking the azimuthal symmetry of the forcing the flow.

2.21.5 Correlations and Trapping

Fractional differential equations appear to be the relevant tool to visualize nondif-

fusive scalings. Instead of the classical scaling r 2 ∝ Dt, which corresponds to the
conventional diffusive equation for the description of the double diffusion regime,
use was made of
∂ 1/2 n ∂ 2n
1/2
= Deff 2 , (2.753)
∂t ∂r
1/2
where the expression ∂∂t 1/2n is the nonlocal fractional differential operator describing
temporal alteration of density accounting for long-range correlation effects [22, 42].
This operator can be interpreted as a fractional time derivative of order α [121, 181]:
 t
∂ γ n(r, t) ∂ n(r, τ )
= dτ. (2.754)
∂t γ ∂t 0 (t − τ )γ
Representing the results in such a manner facilitates interpretation of the nondiffusive
scaling relations. Such an approach is rather unusual but it is based on simple ideas.
Thus, arising nondiffusive effects are related to particles staying in “traps”. If the
probability of the particle staying in a trap during the time t is described by the
scaling  γ
τ0
Φ(t) ∝ , (2.755)
t
where τ0 is the characteristic time and 0 < α < 1, then the modified diffusive
equation is given by the expression

∂γ n ∂ 2n
= D eff . (2.756)
∂t γ ∂r 2
188 O.G. Bakunin

The particle does not disappear in the trap but only, with some probability, is kept
in every trap; this leads to the nonlocal fractional form of the differential equation.
In the case of double diffusion γ = 1/2, the expression for Φ(t) can be easily
explained from the physical point of view. Indeed, if we use the Gaussian expression
for the probability of finding a particle in a small vicinity Δ of the point r at the
moment t,  
Δ r2
P (r, t) ≈  exp − , (2.757)
4πDq t 4Dq t
then the probability of a particle returning to the initial point r = 0, in the framework
of one-dimensional diffusion, is written as
Δ
P (0, t) ≈  . (2.758)
4πDq t

Here Dq is the particle’s collisional diffusivity.

In the case of moving charged particles in a braided magnetic field, the constant
return of charges to the initial point, which is caused their one-dimensional diffusive
motion along force lines, prevents significant transverse displacements. Therefore,
in the model of transport in the braided magnetic field it is natural to use the estimate
for the probability
Δ
Φ(t) ≈ P (0, t) ≈  . (2.759)
4πDq t
A corresponding fractional differential equation with the exponent γ = 1/2 for the
electron density was obtained in [20], where careful calculations of correlation func-
tions were carried out, but the direct relation to the probability of returns at the initial
point P (0, t) was not established. In the presence of strong anisotropy several spatial
and temporal correlation scales describe the stochastic magnetic field and they differ
from correlation scales, which correspond to particle transport. Therefore, scaling
arguments (2.117) look oversimplified. Nevertheless, the development of methods to
describe anomalous transport in stochastic magnetic and electric fields in the con-
ditions of anisotropy allows one to obtain visual estimates [171]. To describe trans-
port in comb-like structures, use is made of independent estimates of characteris-
tic times of particles staying in traps or on a bone. Then, using the approximation
from [171] we can introduce the effective time of diffusion in the transverse direc-
tion:
δ⊥
teff ≈ t, (2.760)
LTrap
where δ⊥ is the perpendicular correlation scale related to the stochastic magnetic
field and LTrap is the correlation scale related to trapping mechanisms due to particle
returns. The analysis of correlation effects [171] leads to a simple estimate of the
value LTrap through the effective time of particles staying in traps

LTrap ∝ tTrap . (2.761)

 2 Correlations and Anomalous Transport Models 189

Based on this approximation, it is easy to obtain the expression for teff in terms of
the probability of charged particles returning:

Φ(t) ≈ δ⊥ / 4πDq t. (2.762)

Since
tTrap ≈ Φ(t)t, (2.763)
we can write 
δ⊥ δ⊥ l  πt
teff ≈ t≈ ≈ 2l , (2.764)
LTrap Dq Φ(t) Dq
where l is the longitudinal correlation length related to longitudinal diffusion. The
expression for the transverse displacements in a static braided magnetic field is given
by  
Dq δ⊥ 
2
r⊥ ≈ Deff teff ≈ Dm teff ≈ Dm ≈ Dm 4πDq t, (2.765)
l Φ(t)
2 ∝ t 1/2 .
which corresponds to the scaling for double diffusion r⊥

2.22 Correlation and Phase-Space

There are certainly deep connections between the conventional space approach to
transport and the phase-space approach. The Hamiltonian approach gives the ad-
vantage of using degrees of freedom to treat nonlocality and memory effects in the
framework of phase-space. The kinetic model provides the possibility of describing
ballistic modes and establishing the relationship between different exponents and
distributions.

2.22.1 The Corrsin Conjecture and Phase-Space

The models considered show the efficiency of applying the Corrsin conjecture to
describe the superdiffusion behavior of a passive scalar. However, long-range corre-
lations play a significant role in considering trapping in turbulent flows. Results of
the direct numerical simulation have led Vlad et al. [158] to the conclusion that side
by side with the trajectories of a percolation character there exist considerable parts
of space where transport is defined by streamlines of the trapping type. By analogy
with the percolation approach [17] based on a careful consideration of the structure
of the percolation streamline (e.g., the hull of a percolation cluster), Vlad, Spineanu,
Misguich, and Balescu suggested that the trapping of test particles could be analyzed
using a model representation of the trap streamline system. Moreover, the essential
modification of the Corrsin conjecture was done in [158].
Here, we describe this new approximation briefly. Thus, from the standpoint of
the authors of [158] it is possible to keep the Corrsin factorization (2.76) but to
considerably modify the trajectory ensemble under consideration. Indeed, the sub-
190 O.G. Bakunin

ensemble of trajectories in which the resulting displacement corresponds to some

fixed value λ could be described as follows:
   
V x(0), 0 V x(t), t X(t)=λ . (2.766)

The author of [158] considered a system of sub-ensembles, where the value of

the initial velocity is fixed in each such system:
      
CV (V0 , t) = V x(0), 0 V x(t), t X (t)=V = V0 V x(t), t X (t)=V . (2.767)
0 0

On the one hand, this model leads to averaging of sub-ensembles over the velocity
with the kinetic distribution f (V0 , t),
 ∞
C(t) = f (V0 , t)CV (V0 , t) dV0 . (2.768)
−∞

On the other hand, this gives the possibility of more detailed analysis of the stream-
line behavior. This is not surprising because for the new conditions the approximation
of expression (2.767) by means of the specially chosen function VC (x, t) permits us
to investigate the specific system of trajectories in the framework of
dx   
= VC (x, t) = V x(t), t V . (2.769)
dt 0

The option of an approximating function defines the character of trajectories and

makes it possible to “visualize” correlation effects, which in this formulation of
the problem are also determined by the expression for VC . Actually, in this ap-
proach the value V0 VC (x, t) replaces the expression for the Eulerian correlation
function CE (x, t). The simplest example of the approximation VC (x, t) was exam-
ined in [187] in the form that allows one to easily solve (2.769) and at the same time
to satisfy the trapping character of transport
   
x t
VC (x, t) = V02 exp − exp − . (2.770)
λ τ

Here, λ is the characteristic spatial scale and τ is the characteristic time. Simple
transformations of (2.769) yield the expression for the displacement

  
|V0 |τ t
x(V0 , t) = sign(V0 )λ ln 1 + 1 − exp − . (2.771)
λ τ

Then, taking into account the estimate x(t → ∞) ≈ λ ln[1+|V0 |τ/λ] ≈ λ ln(1+Ku),
which mirrors trapping, Vlad, Spineanu, Misguich, and Balescu carried out calcula-
tions of the correlation function (2.768) and the diffusion coefficient using the one-
dimensional Maxwellian distribution
 
1 V2
f (V0 ) = √ exp − 02 . (2.772)
π VT VT
 2 Correlations and Anomalous Transport Models 191

Here, VT is the characteristic velocity of the one-dimensional Maxwellian distribu-

tion. Calculations yield
 ∞  
τ V02 V02
DT = √ dV0 exp − . (2.773)
πVT −∞ (1 + |V0 |τ/λ)2 VT2

The results of the calculations allow us to obtain the quasi-linear expression DT ≈

V02 τc ∝ Ku2 for the case Ku  1, and “flat” scaling DT ∝ Kuσ with σ ≈ 0 for the
long-range correlation case Ku  1.
The approach considered was subsequently developed in a number of papers
[158, 187–189] where the two-dimensional model is analyzed in the framework of
the special Hamiltonian function Ψ (the stream function) characterizing the trapping
in the flow under consideration:
d x   
x , t) × eZ = V x(t), t V ,Ψ .
= −∇Ψ ( (2.774)
dt 0 0

Here, Ψ0 is the initial value of the streamline function. Streamlines obtained in this
approach are closed curves, except the single streamline, which is a straight line
along the initial velocity V0 . In other words, the particles moving along closed stream-
lines do not make an essential contribution to the transport, if considered trajectories
are localized. Actually, this case is opposite to the percolation one where a perco-
lation streamline embraces almost the whole plane. The author of [158] found the
universal character of the dependence of the effective diffusion coefficient DT on
the Kubo number for Ku  1, DT ≈ D0 Kuσ . Here, σ ≈ 0.62 and D0 is the
conventional (seed) coefficient of diffusion. This differs from the percolation result
σ = 7/10 suggested in [83]. The majority of results in the model of decorrelation
trajectories [158, 187, 188] were obtained by means of simulations. This is natural
for a complex model where structures play an essential role.

2.22.2 The Hamiltonian Nature of the Universal Hurst Exponent

The problem of the description of nonlocality and memory effects is important not
only for conventional space but also for phase-space. Thus, in 1940 Kramers [190]
pointed out the difficulties encountered in an attempt to obtain the diffusion equation
in ordinary coordinate space

∂n ∂ 2n ∂
=D 2 − (V0 n) (2.775)
∂t ∂x ∂x
from the simplest kinetic equation which includes spatial nonuniformity,
 
∂f ∂f ∂f 1 ∂ kT ∂f
+V − F (x) = Vf + . (2.776)
∂t ∂x ∂V τ ∂V m ∂V
Here, f (t, V , x) is the particle distribution function, V is the velocity, T is the tem-
perature, τ is the characteristic time, and m is the mass of the particle. Even here a
192 O.G. Bakunin

demand arose for a nontrivial approach with integration over a “model trajectory”
r = r0 + V τ in lieu of “conventional averaging” with the fixed value r0 . Here, r0
is an arbitrary initial point. This corresponds to the system of characteristic lines
dV /dt = −V /τ and dr/dt = V . From this point of view one can see that it is
possible to describe the spatial nonuniformity of the distribution function f at scales
λ ≤ V0 τ : f (t, V , x) ≈ f (t, V , x + λ). This means that only local effects are de-
scribed by (2.776). However, this argument was not effective enough for the intro-
duction of corrections to the kinetic equation at that time. Kramers in fact pointed
out the conventional character of the diffusion equation in use and to its close rela-
tion to the notions of the behavior of correlation functions. However, there are now
papers of interest where fractional differentials are recommended for use in the de-
scription of the nondiffusive kinetic effects [191–194]. Naturally, relations between
kinetic models, diffusion equations, and probabilistic estimates have to exist. On the
one hand, the kinetic approach makes it possible to correctly take into account bal-
listic effects that are related to the convective fraction of hydrodynamic flows. On
the other hand, the increase in the number of degrees of freedom, which is related
to the velocity incorporation as an independent variable, permits the description of
nonlocal effects related to spatial nonuniformity and stochastic layers.
Thus, recently Zaslavsky and Edelman [194] suggested a fruitful approach to re-
late phase-space properties of the sticky island boundary to continuous time random
walk scalings. A regime close to the ballistic motion was considered on the basis of
the fairly universal Hamiltonian function [74]
Heff = b1 (φ)2 + b2 g − b3 (g)3 . (2.777)
Here, φ and g are the deviations of the phase φ and the energy g values on the
trajectory from their special quantities φ∗ and g∗ on the ballistic trajectory, which
corresponds to an initial stage when an island is born. The ballistic kind of behavior
is related to long-range trapping near the boundary of the sticky island [196]. It is
possible to estimate the escape probability distribution ψ(t) from the boundary layer
in terms of the phase volume Γ :
1 1
ψ(t) ∝ ≈ . (2.778)
Γ φg
The authors of [195] suggested a solid estimate for the relationship between φ and
g that is based on the Hamiltonian function (2.777): φ ∝ g 3/2 . Then, one can
express the phase volume in the scaling form
Γ ∝ g 5/2 . (2.779)
This simple estimate mirrors the advantages of using additional degrees of free-
dom, which arises in the framework of the Hamiltonian approach. Escaping from
the boundary layers implies that during an initial period g scales with time in ac-
cordance with g ∝ t. Simple calculations [195] yield
1 1 1
ψ(t) ∝ ≈ 5/2
≈ 5/2 . (2.780)
Γ g t
 2 Correlations and Anomalous Transport Models 193

This means that γ = 3/2. Now it is possible to obtain the Hurst exponent that
describes transport effects (2.684):
γ 3
H = = . (2.781)
2 4
This approach looks very attractive due to a universal kind of assumption about the
relationship between ψ and Γ (2.778). Indeed, different models give exponents
[195, 196] which are very close to H = 3/4.

2.22.3 The One-Flight Model and Transport

The large variety of anomalous transport models leads to a search for “hyper-
scalings” and relationships between exponents. This is not surprising since the anal-
ogous situation arose in considering phase transitions and percolation models. The
investigation of relationships between the kinetic and correlation approaches plays
an important role here. Thus, the simplest correlation scaling is the power approxi-
mation of the Lagrangian correlation function
1
C(t) ∝ , (2.782)
t γc
which, in accordance with the Taylor formula (2.6), yields the scaling for the mean
displacement:
R ∝ t 1−γc /2 , 0 < γc < 2. (2.783)
This means that the Hurst exponent is expressed in terms of the temporal correlation
exponent γc . Note that relations between the Lagrangian correlation function and the
probability density ψ(t), which plays an essential role in continuous time random
walks, could exist.
Indeed, Zaslavsky [195] considered a ballistic model when the velocities V (0)
and V (t) that enter into the Lagrangian correlation function C(t) = V (0)V (t)
belong to the same flight. The analysis of [196] was based on the approximation of
the anomalous diffusion in a map model [197] where the relationship
3−γ
H = (2.784)
2
was proposed. Here, γ is the exit time exponent (2.678). Formal calculations make
it possible to obtain from (2.784) and (2.783) the scaling

γc = 1 − γ . (2.785)

Zaslavsky suggested an interesting interpretation of this relationship in terms of the

probability density ψ = Φ  (t) ∝ 1/t γ +1 . Actually, in the framework of the “one-
flight” model, the correlation function C(t) is proportional to the probability Pesc (t)
that a particle will stay in the same flight at least during the time interval t. From the
194 O.G. Bakunin

formal standpoint it can be rewritten in the functional form

 ∞  ∞
C(t) ∝ Pesc (t) ≈ dt1 dt2 ψ(t2 ). (2.786)
t t1

One can see that the scaling interpretation of this expression leads to γc = 1 − γ .
It is natural to consider other approximations of relaxation functions Φ(t) as
well. The Kohlrausch relaxation law (2.662) is an important example:
  
t
Φ(t) ∝ exp − . (2.787)
τ

Of particular interest is the fact that the Kohlrausch slowed relaxation law is related
to the Laplace transformation of the familiar Levy’s law for jumps with the exponent
αL = 1/2:
 ∞
√
exp(− ws) = exp(−sx)fαL (x) dx, (2.788)

0
  (2.789)
1 w w
fαL (x) = 3/2 exp − .
2x π 4x

This simple formula clearly shows a close relationship between the memory and non-
locality effects. Physically, this relationship is not surprising. A particle that stays in a
trap in phase space is not involved in events (does not undergo collisions). However,
in conventional coordinate space, such a collisionless particle is transported over a
large distance during the time it stays within the phase-space trap. In this sense, col-
lisionless particles cannot be regarded as being involved in a conventional diffusion
process, in contrast to the particles that undergo collisions.
The physical meaning of formal relationship (2.788) can easily be understood by
treating its integral part as the averaging procedure for the Poisson law:

exp(−sx) = exp(−x/x0 ). (2.790)

As an example, let us consider the case x = V , s = 1/x0 ≈ t/L0 , w = V0 , where

V is the particle’s velocity V0 is the characteristic velocity, and L0 is the size of the
region over which the averaging is performed. In this case, we have
    ∞  
V0 Vt
exp − t = exp − f (V ) dV . (2.791)
L0 0 L0

As a result, we see that the Kohlrausch relaxation law describes the Poisson’s prob-
ability for a particle not to undergo collision in a region of size L0 during the time t,
averaged by means of a Levy distribution with αL = 1/2:
 
1 V0
f (V ) = f1/2 (V ) ∝ 3/2 exp − . (2.792)
V 4V
 2 Correlations and Anomalous Transport Models 195

Note that power tails play an important role in the kinetic consideration of electron
anomalous transport [198, 199] in the high temperature nonuniform plasma. Actu-
ally, for strongly nonequilibrium systems where accelerating and trapping mecha-
nisms are present, it is necessary to take into account nondiffusive effects in describ-
ing the processes in phase-space.

2.22.4 Correlations and Nonlocal Velocity Distribution

From the common standpoint, just the distortion of the Maxwellian particle veloc-
ity distribution function leads to the difference between the transport equation in
coordinate space and the classical diffusion one. It is obvious that the probabilistic
interpretation is also effective for this case. Thus, the ballistic particle motion can
be interpreted as trapping in phase space, since if collisions (interactions) are ab-
sent then the particle has constant velocity and hence does not change its position in
the velocity space. The effectiveness of the one-flight approximation [197] indicates
the possibility of using the ballistic character of motion for obtaining simple scaling
estimates. Thus, the probability Φ(t) of avoiding an “event” (capture by a trap, for
example) during time t can become the grounds for building a simple kinetic model.
It is possible to represent this probability as a result of averaging over the ensemble
of particles with the velocity distribution function f (V ).
Indeed, if the characteristic distance between traps is R, then the probability of
avoiding trapping for a particle with velocity V could be the estimate in the form
t Vt
w(V , t) ≈ 1 − ≈1− . (2.793)
T R
Here, T = R/V is the characteristic time for the particle with velocity V to reach a
trap. This linear estimate can be interpreted as decomposition over the small parame-
ter V t/R  1 of the conventional Poisson probability of avoiding an “event” during
time t, exp(−V t/R) ≈ 1 − V t/R. In the framework of the randomization method,
we can obtain the integral
 V t<R
Φ(t) = w(V , t)f (V ) dV . (2.794)
0

Note that the upper limit of the integral is not constant: V < R/t. Then, the simple
transformations of this expression yield the functional equation [200]
 R/t   
Vt
Φ(t) = 1− f (V ) dV . (2.795)
0 R

Following the Montroll–Weiss ideas [21, 22] we can assume that the function Φ(t)
is known, or could be approximated by one of the characteristic probabilistic distrib-
utions. Then, it is possible to solve this integral equation. Using a double differenti-
ation over t, we obtain the equation for the definition of the distribution f (V ). Thus,
196 O.G. Bakunin

after the first differentiation of the functional, we obtain

 R/t
d V
Φ(t) = −ψ(t) = − f (V ) dV . (2.796)
dt 0 R
Here, we take into account that the upper limit of the considered integral depends on
time: V = R/t. The next differentiation yields

R2 d 2 R2 d
f (V ) = 3 2 Φ(t) = − 3 ψ(t) . (2.797)
V dt t=R/V V dt t=R/V

It is a very simple and at the same time nontrivial relationship, which connects kinetic
and probabilistic characteristic functions. Now we can use characteristic functions
for Φ(t). In the first case, we assume that Φ(t) has the exponential Poisson form,
which corresponds to the absence of memory effects in the framework of continuous
time random walks [18–22],
 
t
Φ(t) = exp − . (2.798)
τ
Here, τ is the characteristic time that is the parameter of the problem. Formal calcu-
lations then permit obtaining the mean waiting time by conventional averaging:
 ∞  
d 1 t
t = tψ(t) dt = τ , where ψ(t) = − Φ(t) = exp − . (2.799)
0 dt τ τ
As a result of simple calculations for the Poisson case, we obtain the expression for
the velocity distribution function f (V ),which depends on two parameters R and τ ,
   
1 R 2 R
f (V ) = 3 exp − . (2.800)
V τ Vτ

For large values of V , one deals with the scaling f (V ) ∝ 1/V 3 . In addition, the ex-
ponential factor of this function is nonanalytical with V → 0 and therefore it cannot
be obtained by the asymptotic technique from the diffusive phase-space Fokker–
Planck equation.
In the framework of the scaling approach, it is important to consider the repre-
sentation of the probability Φ(t) in the power form [18–22]:
1
Φ(t) = . (2.801)
(1 + t/τ )γ
Here, γ < 1 is the characteristic exponent. In the continuous time random walk
approach, this corresponds to the consideration of memory effects. However, in our
case it is important that the mean waiting time t is an infinite value [18–22]. Upon
substituting (2.801) into (2.797), we obtain the velocity distribution function,
 γ
τ 1
f (V ) ≈ γ (γ + 1) 1−γ
. (2.802)
R V
 2 Correlations and Anomalous Transport Models 197

This corresponds to the relationship between both kinetic exponent β and waiting
time exponent γ : β = 1 − γ . Note that this distribution leads to the fractal char-
acter of collisionless particles phase space δN(V ) ∝ f (V )V ∝ V β with the di-
mensionality dF = β. Here, δN is the collisionless particles number. These scalings
characterize systems under the unconventional condition t = ∞, which differs sig-
nificantly from the conventional Poisson model, where the mean waiting time t is
finite.

2.22.5 The Arrhenius Law and Phase-Space Distribution

An interesting example of a long-tailed distribution is found by considering a particle

in a potential well of height E. Now suppose the random time the particle spends in
the well is governed by Arrhenius’s law:
 
E
τ = τ0 exp , (2.803)
kT

where 1/τ0 is the frequency, k is the Boltzmann constant, and T is the tempera-
ture. The barrier height E is a random variable governed by the probability distribu-
tion f (E),  
E
f (E) = exp − , (2.804)
E0
where E0 is the width of the barrier height distribution. This distribution of barrier
heights introduces a distribution of trapping times

ψ(τ ) dτ = f (E) dE. (2.805)

From (2.803), (2.804), and (2.805) we find the trapping times distribution
−γ
kT τ0 A
ψ(τ ) = , (2.806)
τ 1+γA

where γA = kT E0 . Here, we have an exponent of long-tailed probability distribution,

which could change continuously as the temperature parameter changes. If γA is
greater than unity, the first moment of the waiting time distribution will become fi-
nite. However, for 0 < β < 1, it will be infinite. In this case, the point set of jump
times will look like a randomized Carton set of dimension β, hence the name “frac-
tal time”. Actually the combination of both the Arrhenius law and the Boltzmann
distribution leads to a power law for escape probability. The exponent describing the
escape probability in the case considered above depends on the energetic parameters
of the model.
However for nonequilibrium states, the distribution function can differ signifi-
cantly from the Boltzmann one [200, 201]. Thus, nontrivial distribution functions
naturally lead to the necessity of considering relaxation laws different from the Ar-
rhenius law. As an example, let us obtain a relaxation law corresponding to a Levy-
198 O.G. Bakunin

like velocity distribution function:

  β
τ0 E0 βA E0 A
ψ(τ ) dτ = dτ = const · exp − dE. (2.807)
τ γA +1 E E βA +1
Simple calculations yield the relaxation law in the form

 βA  1
E γA
τ = exp . (2.808)
E0

Here, the case βA = 12 corresponds to the Levy distribution. The obtained expression
also includes two parameters, βA and γA , which characterize both the kinetics of
particles and trapping effects. This makes it possible to consider nontrivial transport
effects, where long-tailed distributions play an essential role [202].

2.23 Conclusion
Naturally, it is impossible to consider all the aspects of turbulent transport within the
framework of one article. Therefore, in this paper we focused on scaling arguments
that play an important role in obtaining estimates of transport effects. All necessary
calculations were considered in detail. In the framework of the mean field theory, of
course, many important problems cannot be considered; however, one of the main
tasks is to establish the character of the dependence of the transport coefficient on
such parameters as fluctuation amplitude, characteristic size, characteristic time, etc.
For these purposes, the considered approach is effective.

Acknowledgments The author thanks V. Shafranov, V. Kogan, E. Yurchenko, R. Balescu,

G. Zaslavsky, S. Shandarin, and T. Schep for the useful discussions and support. The author
is also indebted to A. Dnestrovsky, Y. Igitchanov, V. Ilgisonis, A. Kukushkin, A. Loskutov,
D. Morozov, V. Pustovitov, A. Runov, R. Shurigin and Hugo de Blank for their reviews and
for their invaluable help in proofreading the manuscript.

References
[1] J. Wesson, Tokamaks (Clarendon, Oxford, 1987)
[2] B.B. Kadomtsev, Tokamak Plasma: A Complex System (IOP, Bristol, 1992)
[3] W. Horton, Y.-H. Ichikawa, Chaos and Structures in Nonlinear Plasmas (World Scien-
tific, Singapore, 1994)
[4] J. Weiland, Collective Models in Inhomogeneous Plasma: Kinetic and Advanced Fluid
Theory (IOP, Bristol, 2000)
[5] M.N. Rosenbluth, R.Z. Sagdeev (eds.), Handbook of Plasma Physics (North-Holland,
Amsterdam, 1984)
[6] P.H. Diamond, Plasma Phys. Control. Fusion 47, R35 (2005)
[7] J.A. Krommes, Phys. Rep. 360, 1–352 (2002)
 2 Correlations and Anomalous Transport Models 199

[8] A. Yoshizawa, S.-I. Itoh, K. Itoh, N. Yokoi, Plasma Phys. Control. Fusion 43, R1 (2001)
[9] A. Brandenburg, Phys. Rep. 417, 1 (2005)
[10] B. Eliasson, Phys. Rep. 422, 225 (2006)
[11] P.W. Terry, Rev. Mod. Phys. 72, 109 (2000)
[12] B.D. Chandran, S.C. Cowley, Phys. Rev. Lett. 80, 3077 (1998)
[13] S. Chandrasekhar, Rev. Mod. Phys. 15, 1 (1943)
[14] N.G. van Kampen, Stochastic Processes in Physics and Chemistry (North-Holland, Am-
sterdam, 1981)
[15] Y.B. Zel’dovich, A.D. Myshkis, Principles of Mathematical Physics (Nauka, Moscow,
1973)
[16] P. Résibois, M. De Leener, Classical Kinetic Theory of Fluids (Wiley, New York, 1977)
[17] M.B. Isichenko, Rev. Mod. Phys. 64, 961 (1992)
[18] J.P. Bouchaud, A. Georges, Phys. Rep. 195, 132–292 (1990)
[19] J.W. Haus, K.W. Kehr, Phys. Rep. 150, 263 (1987)
[20] R. Balescu, Statistical Dynamics (Imperial College Press, London, 1997)
[21] M.F. Shlesinger, G.M. Zaslavsky, U. Frisch, Levy Flights and Related Topics in Physics.
Lecture Notes in Physics, vol. 450 (Springer, New York, 1995)
[22] R. Metzler, J. Klafter, Phys. Rep. 339, 1 (2000)
[23] A. Vedenov, E.P. Velikhov, R.Z. Sagdeev, Plasma Phys. Control. Nucl. Fusion Res. 2,
82 (1962)
[24] W.E. Drummond, D. Pines, Plasma Phys. Control. Nucl. Fusion Res. 3, 1049 (1962)
[25] A.M. Yaglom, Correlation Theory of Time-Independent Random Functions (Gidrome-
teoizdat, Leningrad, 1981)
[26] V.I. Klyatskin, Stochastic Equations by the Physicist’s Eyes (Fizmatlit, Moscow, 2001)
[27] A.S. Monin, A.M. Yaglom, Statistical Fluid Mechanics; Mechanics of Turbulence (MIT
Press, Cambridge, 1971)
[28] F.N. Frenkiel (ed.), Atmospheric Diffusion and Air Pollution (Academic Press, New
York, 1959)
[29] W. Frost, T.H. Moulden (eds.), Handbook of Turbulence (Plenum, New York, 1977)
[30] U. Frisch, Turbulence: The Legacy of A.N. Kolmogorov (Cambridge University Press,
Cambridge, 1995)
[31] G.I. Taylor, Proc. Lond. Math. Soc. Ser. 2 20, 196 (1921)
[32] H.K. Moffatt, J. Fluid Mech. 106, 27 (1981)
[33] H.K. Moffatt, Rep. Prog. Phys. 46, 621 (1983)
[34] S. Corrsin, Random Geometric Problems Suggested by Turbulence, in Statistical Mod-
els and Turbulence: Proceedings of a Symposium, ed. by M. Rosenblatt, C. van Atta.
San Diego CA, USA, 15–21 July 1971. Lecture Notes in Physics, vol. 12 (Springer,
Berlin, 1975), p. 300
[35] Y.B. Zel’dovich, Sov. Phys. Dokl. 27, 797 (1982)
[36] Y.B. Zel’dovich, S.A. Molchanov, A.A. Ruzmaı̌kin, D.D. Sokolov, Sov. Phys. Usp. 30,
353 (1987)
[37] A.A. Galeev, M.M. Kuznetsova, L.M. Zelenyi, Space Sci. Rev. 44, 1 (1986)
[38] J.A. Krommes, Prog. Theor. Phys. Suppl. 64, 137 (1978)
[39] J.A. Krommes, J. Plasma Phys. 30, 11 (1983)
[40] O.G. Bakunin, Plasma Phys. Control. Fusion 47, 1857 (2005)
[41] M. Avellaneda, A.J. Majda, Phys. Fluids 4, 41 (1992)
[42] O.G. Bakunin, J. Plasma Phys. 72, 647 (2006)
[43] G. Falkovich, K. Gawedzki, M. Vergassola, Rev. Mod. Phys. 73, 913 (2001)
[44] S.I. Vainshtein, Magnetic Fields in Space (Nauka, Moscow, 1983)
[45] L. Fradkin, Plasma Phys. Control. Fusion 33, 685 (1991)
200 O.G. Bakunin

[46] O.G. Bakunin, Plasma Phys. Rep. 29, 955 (2003)

[47] A. Fanjiang, J. Stat. Phys. 88, 1033 (1997)
[48] R.J. Bickerton, Plasma Phys. Control. Fusion 39, 339 (1997)
[49] D. Stauffer, Introduction to Percolation Theory (Taylor & Francis, London, 1985)
[50] H.E. Stanley, J. Stat. Phys. 34, 843 (1984)
[51] I.M. Sokolov, Sov. Phys. Usp. 29, 924 (1986)
[52] J.M. Ziman, Models of Disorder: The Theoretical Physics of Homogeneously Disor-
dered Systems (Cambridge University Press, Cambridge, 1979)
[53] D. Stauffer, Phys. Rep. 54, 1 (1979)
[54] R. Narayan, Phys. Rev. Lett. 42, 1247 (1999)
[55] H.E. Stanley, J. Stat. Phys. 34, 843 (1984)
[56] B.B. Mandelbrot, Fractals: Form, Chance, and Dimension (Freeman, San Francisco,
1977)
[57] J. Feder, Fractals (Plenum, New York, 1988)
[58] L.F. Richardson, Proc. R. Soc. Lond. Ser. A 110, 709 (1926)
[59] G.K. Batchelor, Proc. Camb. Philos. Soc. 48, 345 (1952)
[60] B.I. Davydov, C. R. Acad. Sci. USSR 2, 216 (1934)
[61] A. Khintchine, P. Lévy, C. R. Acad. Sci. (Paris) 202, 274 (1936)
[62] A.S. Monin, Dokl. Akad. Nauk SSSR 105, 256 (1955)
[63] A.N. Kolmogorov, Dokl. Akad. Nauk SSSR 30, 299 (1941)
[64] T.H. Dupree, Phys. Fluids 9, 1773 (1966)
[65] T.H. Dupree, Phys. Fluids 10, 1049 (1967)
[66] T.H. Dupree, Phys. Fluids 15, 334 (1972)
[67] B.B. Kadomtsev, O.P. Pogutse, in Plasma Physics and Controlled Nuclear Fusion Re-
search, Proceedings of the 7th International Conference, IAEA Vienna, 1978. Nuclear
Fusion Supplement, vol. 1 (1979), p. 649
[68] B.J. Berne, J. Chem. Phys. 56, 2164 (1972)
[69] G.G. Getmantsev, Sov. Astron. Rep. 6, 477 (1962)
[70] J.R. Jokipii, E.N. Parker, Astrophys. J. 155, 777 (1969)
[71] J.B. Taylor, B. McNamara, Phys. Fluids 14, 1492 (1971)
[72] Yu.A. Dreizin, A.M. Dykhne, Sov. Phys. JETP 36, 127 (1973)
[73] I.D. Howells, J. Fluid Mech. 9, 104 (1960)
[74] G.M. Zaslavskiı̌, B.V. Chirikov, Sov. Phys. Usp. 14, 549 (1972)
[75] A.B. Rechester, M.N. Rosenbluth, Phys. Rev. Lett. 40, 38 (1978)
[76] B.B. Mandelbrot, J. Fluid Mech. 72, 401 (1975)
[77] U. Frisch, P.-L. Sulem, M. Nelkin, J. Fluid Mech. 87, 719 (1978)
[78] H.G.E. Hentschel, I. Procaccia, Phys. Rev. A 29, 1461 (1984)
[79] M.E. Levinshtein, B.I. Shklovsky, M.S. Shur, A.L. Efros, J. Exp. Theor. Phys. 42, 112
(1976)
[80] M.V. Osipenko, O.P. Pogutse, N.V. Chudin, Sov. J. Plasma Phys. 13, 550 (1987)
[81] M.N. Rosenbluth, H.L. Berk, I. Doxas, W. Horton, Phys. Fluids 30, 2636 (1987)
[82] M.B. Isichenko, Y.L. Kalda, E.V. Tatarinova, O.V. Tel’kovskaya, V.V. Yankov, Sov.
Phys. JETP 69, 517 (1989)
[83] A.V. Gruzinov, M.B. Isichenko, Y.L. Kalda, Sov. Phys. JETP 70, 263 (1990)
[84] P.N. Yushmanov, Comments Plasma Phys. Control. Fusion 14, 313 (1992)
[85] M.B. Isichenko, J. Kalda, J. Nonlinear Sci. 1, 255 (1991)
[86] M.B. Isichenko, J. Kalda, J. Nonlinear Sci. 1, 375 (1991)
[87] B.Y. Balagurov, V.G. Vaks, Sov. Phys. JETP 38, 968 (1974)
[88] P. Langevin, C. R. Acad. Sci. (Paris) 146, 530 (1908)
[89] A. Einstein, Ann. Phys. (Leipzig) 17, 549 (1905)
 2 Correlations and Anomalous Transport Models 201

[90] J. Kampé de Feriet, Ann. Soc. Sci. Bruxelles 59, 145 (1939)
[91] A.M. Obukhov, C. R. Acad. Sci. URSS 32, 19 (1941)
[92] C. Cattaneo, Atti Semin. Mat. Fis. Univ. Modena 3, 83 (1949)
[93] S. Goldstein, Q. J. Mech. Appl. Math. 4, 129 (1951)
[94] R.W. Davies, Phys. Rev. 93, 1169 (1954)
[95] O.G. Bakunin, Phys. Usp. 46, 309 (2003)
[96] G.M. Zaslavsky, B.A. Niyazov, Phys. Rep. 283, 73 (1997)
[97] G.M. Zaslavsky, Chaos 4, 25 (1994)
[98] A. Okubo, J. Oceanol. Soc. Jpn. 20, 286 (1962)
[99] M. Smoluchowski, Phys. Z. 17, 557 (1916)
[100] S. Chapman, Philos. Trans. R. Soc. Ser. A 216, 279 (1916)
[101] A. Kolmogoroff, Math. Ann. 104, 415 (1931)
[102] A. Cauchy, C. R. Acad. Sci. (Paris) 37, 292 (1853)
[103] L.O. Silva, Astrophys. J. Lett. 596, L121 (2006)
[104] O.V. Tel’kovskaya, K.V. Chukbar, J. Exp. Theor. Phys. 85, 87 (1997)
[105] J. Weinstock, Phys. Fluids 19, 1702 (1976)
[106] R.H. Kraichnan, J. Fluid Mech. 81, 385 (1977)
[107] R.H. Kraichnan, Phys. Fluids 13, 22 (1970)
[108] T.S. Lundgren, Y.B. Pointin, Phys. Fluids 19, 355 (1976)
[109] J.S. Hay, F. Pasquill, Adv. Geophys. 6, 345 (1959)
[110] Y. Kaneda, T. Ishida, J. Fluid Mech. 402, 311 (2000)
[111] A.I. Smolyakov, P.N. Yushmanov, Nucl. Fusion 33, 383 (1993)
[112] C. Cambon, F.S. Godeferd, F.C.G.A. Nicolleau, J.C. Vassilicos, J. Fluid Mech. 499, 231
(2004)
[113] D.L. Koch, J.F. Brady, Phys. Fluids A 1, 47 (1990)
[114] E.B. Tatarinova, P.A. Kalugin, A.V. Sokol, Europhys. Lett. 14, 773 (1991)
[115] O.G. Bakunin, Phys. Usp. 46, 733 (2003)
[116] M.B. Isichenko et al., Phys. Fluids B 4, 3973 (1992)
[117] G.I. Taylor, Proc. R. Soc. Lond. Ser. A 219, 186 (1953)
[118] G.I. Taylor, Proc. R. Soc. Lond. Ser. A 223, 446 (1954)
[119] W.R. Young, S. Jones, Phys. Fluids A 3, 1087 (1991)
[120] G. Matheron, G. de Marsily, Water Resour. Res. 16, 901 (1980)
[121] J.-P. Bouchaud et al., Phys. Rev. Lett. 64, 2503 (1990)
[122] S. Redner, Physica D 38, 287 (1989)
[123] B.B. Kadomtsev, Cooperative effects in plasmas, in Reviews of Plasma Physics, vol. 22,
ed. by V.D. Shafranov (Kluwer Academic/Consultants Bureau, New York, 2001), p. 1
[124] A.S. Kingsep, Introduction to the Nonlinear Plasma Physics (Mosk. Fiz.-Tekh. Inst.,
Moscow, 1996)
[125] V.N. Tsytovich, Theory of Turbulent Plasma (Consultants Bureau, New York, 1977)
[126] M.N. Rosenbluth, R.Z. Sagdeev, L.B. Taylor, Nucl. Fusion 6, 297 (1966)
[127] R.W. Flynn, Phys. Fluids 14, 956 (1971)
[128] K.N. Graham, J.A. Fejer, Phys. Fluids 19, 1054 (1976)
[129] F. Doveil, D. Grésillon, Phys. Fluids 25, 1396 (1982)
[130] O. Ishihara, A. Hirose, Comments Plasma Phys. Control. Fusion 8, 229 (1984)
[131] A. Salat, Z. Naturforsch. A 38, 1189 (1983)
[132] O. Ishihara, A. Hirose, Phys. Fluids 28, 2159 (1985)
[133] A. Salat, Phys. Fluids 31, 1499 (1988)
[134] O. Ishihara, H. Xia, A. Hirose, Phys. Fluids B 4, 349 (1992)
[135] R. Balescu, H.-D. Wang, J.H. Misguich, Phys. Plasmas 1, 3826 (1994)
202 O.G. Bakunin

[136] D. Bohm, E. Burhop, in The Characteristics of Electrical Discharges in Magnetic

Fields, ed. by A. Guthrie, R.K. Wakerling (McGraw-Hill, New York, 1949)
[137] V.S. Ptuskin, Astrophys. Space Sci. 61, 251 (1979)
[138] T.H. Stix, Nucl. Fusion 18, 353 (1978)
[139] A.B. Rechester, M.N. Rosenbluth, R.B. White, Phys. Rev. Lett. 42, 1247 (1979)
[140] B.B. Mandelbrot, The Fractal Geometry of Nature (Freeman, San Francisco, 1982)
[141] L.F. Richardson, in General System Yearbook of the Society for the Advancements of
General System Theory, vol. 6 (The Society for General Systems Research, Ann Arbor,
1961), p. 139
[142] J. Klafter, A. Blumen, M.F. Shlesinger, Phys. Rev. A 35, 3081 (1987)
[143] W.D. McComb, The Physics of Fluid Turbulence (Clarendon, Oxford, 1994)
[144] A. Bunde, J.F. Gouyet, J. Phys. A: Math. Gen. 18, L285 (1985)
[145] H. Saleur, B. Duplantier, Phys. Rev. Lett. 58, 2325 (1987)
[146] B. Sapoval, B. Rosso, J. Gouyet, J. Phys. Lett. (Paris) 46, L149 (1985)
[147] S.A. Trugman, Phys. Rev. B 27, 7539 (1983)
[148] O.G. Bakunin, Chaos Solitons Fractals 23, 1703 (2005)
[149] O.G. Bakunin, Physica A 345, 1 (2005)
[150] M.B. Isichenko, Plasma Phys. Control. Fusion 33, 795 (1991)
[151] M.B. Isichenko, Plasma Phys. Control. Fusion 33, 809 (1991)
[152] M. Ottaviani, Europhys. Lett. 20, 111 (1992)
[153] J.-D. Reuss, J.H. Misguich, Phys. Rev. E 54, 1857 (1996)
[154] G. Zimbardo, P. Veltri, P. Pommois, Phys. Rev. E 61, 1940 (2000)
[155] Y.B. Zeldovich, JETP Lett. 38, 57 (1983)
[156] O.G. Bakunin, Plasma Phys. Control. Fusion 45, 1909 (2003)
[157] O.G. Bakunin, J. Plasma Phys. 71, 435 (2005)
[158] M. Vlad, F. Spineanu, J.H. Misguich, R. Balescu, Phys. Rev. E 58, 7359 (1998)
[159] M. Matsushita, S. Ouchi, K. Honda, J. Phys. Soc. Jpn. 60, 2109 (1991)
[160] S. Isogami, M. Matsushita, J. Phys. Soc. Jpn. 61, 1445 (1992)
[161] J. Kondev, C.L. Henley, Phys. Rev. Lett. 74, 4580 (1995)
[162] J. Kondev, C.L. Henley, D.G. Salinas, Phys. Rev. E 61, 104 (2000)
[163] J. Kondev, G. Huber, Phys. Rev. Lett. 86, 5890 (2001)
[164] J. Kalda, Phys. Rev. E 64, 020101 (2001)
[165] J. Kalda, Phys. Rev. Lett. 90, 118501 (2003)
[166] O.G. Bakunin, Physica A 351, 241 (2005)
[167] O.G. Bakunin, T.J. Schep, Phys. Lett. A 322, 105 (2004)
[168] O.G. Bakunin, Physica A 347, 289 (2004)
[169] V.E. Kravtsov, I.V. Lerner, V.I. Yudson, Sov. Phys. JETP 64, 336 (1986)
[170] O.G. Bakunin, Physica A 348, 245 (2005)
[171] I.A. Lubashevskiı̌, A.A. Zemlyanov, J. Exp. Theor. Phys. 87, 700 (1998)
[172] P.S. Iroscnikov, Astron. J. 40, 742 (1963)
[173] R.H. Kraichnan, Phys. Fluids 8, 1385 (1965)
[174] E.W. Montroll, G.H. Weiss, J. Math. Phys. 6, 167 (1965)
[175] H. Scher, E.W. Montroll, Phys. Rev. B 12, 2455 (1975)
[176] P. Goldreich, S. Sridhar, Astrophys. J. 485, 680 (1997)
[177] A. Compte, J. Camacho, Phys. Rev. E 56, 5445 (1997)
[178] M.F. Shlesinger, B.J. West, J. Klafter, Phys. Rev. Lett. 58, 1100 (1987)
[179] R. Balescu, Plasma Phys. Control. Fusion 42, B1 (2000)
[180] R. Balescu, Phys. Rev. E 51, 4807 (1995)
[181] O.G. Bakunin, Rep. Prog. Phys. 67, 965 (2004)
[182] K.V. Chukbar, J. Exp. Theor. Phys. 82, 719 (1996)
 2 Correlations and Anomalous Transport Models 203

[183] O. Cardoso, P. Tabeling, Europhys. Lett. 7, 225 (1988)

[184] O. Cardoso, P. Tabeling, Eur. J. Mech. B: Fluids 8, 459 (1989)
[185] T.H. Solomon, E.R. Weeks, H.L. Swinney, Physica D 76, 70 (1994)
[186] E.R. Weeks, J.S. Urbach, H.L. Swinney, Physica D 97, 291 (1996)
[187] J.-D. Reuss, F. Spineanu, J.H. Misguich, J. Plasma Phys. 59, 707 (1998)
[188] M. Vlad, F. Spineanu, J.H. Misguich, R. Balescu, Phys. Rev. E 63, 066304 (2001)
[189] M. Vlad, F. Spineanu, J.H. Misguich, R. Balescu, Phys. Rev. E 66, 038302 (2002)
[190] H.A. Kramers, Physica 7, 284 (1940)
[191] G.M. Zaslavsky, Chaos 4, 25 (1994)
[192] A.V. Chechkin, V.Y. Gonchar, J. Exp. Theor. Phys. 91, 635 (2000)
[193] O.G. Bakunin, Plasma Phys. Rep. 29, 785 (2003)
[194] O.G. Bakunin, Plasma Phys. Rep. 30, 338 (2004)
[195] G.M. Zaslavsky, M. Edelman, Chaos 10, 135 (2000)
[196] C.F.F. Karney, Physica D 8, 360 (1983)
[197] O.G. Bakunin, S.I. Krasheninnikov, Plasma Phys. Rep. 21, 502 (1995)
[198] O.G. Bakunin, S.I. Krasheninnikov, Plasma Phys. Rep. 16, 501 (1990)
[199] O.G. Bakunin, Phys. Lett. A 330, 22 (2004)
[200] A.S. Mikhailov, Phys. Rep. 425, 79 (2006)
[201] T.M. Squires, Rev. Mod. Phys. 77, 977 (2005)
[202] O.G. Bakunin, Physica A 346, 284 (2005)