Volume 3
Number 11
Electronic Journal of Theoretical Physics
ISSN 1729-5254
EJTP
Editors
Ammar Sakaji
http://www.ejtp.com
Ignazio Licata
June, 2006
E-mail:info@ejtp.com
Volume 3
Number 11
Electronic Journal of Theoretical Physics
ISSN 1729-5254
EJTP
Editors
Ammar Sakaji
http://www.ejtp.com
Ignazio Licata
June, 2006
E-mail:info@ejtp.com
Editor in Chief
A. J. Sakaji
EJTP Publisher
P. O. Box 48210
Abu Dhabi, UAE
info@ejtp.com
Info@ejtp.info
Editorial Board
Co-Editor
Ignazio Licata,
Foundations of Quantum Mechanics
Complex System & Computation in Physics and Biology
IxtuCyber for Complex Systems
Sicily – Italy
editor@ejtp.info
ignazio.licata@ejtp.info
ignazio.licata@ixtucyber.org
Wai-ning Mei
Condensed matter Theory
Physics Department
University of Nebraska at Omaha,
Omaha, Nebraska, USA
e-mail: wmei@mail.unomaha.edu
physmei@unomaha.edu
Tepper L. Gill
F.K. Diakonos
Mathematical Physics, Quantum Field Theory
Department of Electrical and Computer Engineering
Howard University, Washington, DC, USA
e-mail: tgill@Howard.edu
Statistical Physics
Physics Department, University of Athens
Panepistimiopolis GR 5784 Zographos, Athens, Greece
e-mail: fdiakono@cc.uoa.gr
tgill@ejtp.info
Jorge A. Franco Rodríguez
General Theory of Relativity
Av. Libertador Edificio Zulia P12 123
Caracas 1050 Venezuela
e-mail: jorafrar301@cantv.net
jorgeafr@ejtp.info
Nicola Yordanov
Physical Chemistry
Bulgarian Academy of Sciences,
BG-1113 Sofia, Bulgaria
Telephone: (+359 2) 724917 , (+359 2) 9792546
e-mail: ndyepr@ic.bas.bg
J. A. Maki
Applied Mathematics
School of Mathematics
University of East Anglia Norwich NR4 7TJ UK
e-mail: jam@cmp.uea.ac.uk
maki@ejtp.info
S.I. Themelis
Atomic, Molecular & Optical Physics
Foundation for Research and Technology - Hellas
P.O. Box 1527, GR-711 10 Heraklion, Greece
e-mail: stheme@iesl.forth.gr
ndyepr[AT]bas.bg
T. A. Hawary
Mathematics
Department of Mathematics
Mu'tah University
P.O.Box 6
Karak- Jordan
e-mail: drtalal@yahoo.com
Arbab Ibrahim
Theoretical Astrophysics and Cosmology
Department of Physics,
Faculty of Science,
University of Khartoum,
P.O. Box 321, Khartoum 11115,
Sudan
e-mail: aiarbab@uofk.edu
arbab_ibrahim@ejtp.info
Sergey Danilkin
Instrument Scientist, The Bragg Institute
Australian Nuclear Science and Technology Organization
PMB 1, Menai NSW 2234
Australia
Tel: +61 2 9717 3338
Fax: +61 2 9717 3606
e-mail: s.danilkin@ansto.gov.au
Robert V. Gentry
The Orion Foundation
P. O. Box 12067
Knoxville, TN 37912-0067
USA
e-mail: gentryrv[@orionfdn.org
rvgentry@ejtp.info
Attilio Maccari
Beny Neta
Nonlinear phenomena, chaos and solitons in classic and quantum
physics
Technical Institute "G. Cardano"
Via Alfredo Casella 3
00013 Mentana RM - ITALY
Applied Mathematics
Department of Mathematics
Naval Postgraduate School
1141 Cunningham Road
Monterey, CA 93943, USA
e-mail: solitone@yahoo.it
e-mail: byneta@gmail.com
Haret C. Rosu
A. Abdelkader
Advanced Materials Division
Institute for Scientific and Technological Research (IPICyT)
Camino a la Presa San José 2055
Col. Lomas 4a. sección, C.P. 78216
San Luis Potosí, San Luis Potosí, México
Experimental Physics
Physics Department, AjmanUniversity
Ajman-UAE
e-mail: atef28@gmail.com
atef@ejtp.info
e-mail: hcr@titan.ipicyt.edu.mx
Leonardo Chiatti
Medical Physics Laboratory ASL VT
Via S. Lorenzo 101, 01100 Viterbo (Italy)
Tel : (0039) 0761 236903
Fax (0039) 0761 237904
e-mail: fisica1.san@asl.vt.it
Zdenek Stuchlik
Relativistic Astrophysics
Department of Physics, Faculty of Philosophy and Science,
Silesian University, Bezru covo n´am. 13, 746 01 Opava,
Czech Republic
e-mail: Zdenek.Stuchlik@fpf.slu.cz
chiatti@ejtp.info
Copyright © 2003-2006 Electronic Journal of Theoretical Physics (EJTP)
All rights reserved
Table of Contents
No
1
Articles
Non-Minimal Coupling Effects of the Ultra-Light
Particles on Photons Velocities in the Radiation
Dominated Era of the Universe.
Page
1
El-Nabulsi Ahmad Rami
2
A Toy Model of Financial Markets
11
J. P. Singh and S. Prabakaran
3
Rayleigh process and matrix elements for the onedimensional harmonic oscillator
29
J.H. Caltenco, J.L. López-Bonilla, and J. Morales
4
Identical synchronization in chaotic jerk dynamical
systems
33
Vinod Patidar and K. K. Sud
5
Second Order Perturbation of Heisenberg Hamiltonian
for Non-Oriented Ultra-Thin Ferromagnetic Films
71
P. Samarasekara
6
Frameable Processes with Stochastic Dynamics
85
Enrico Capobianco
7
Ab-initio Calculations for Forbidden M1/E2 Decay
Rates in Ti XIX ion
111
A. Farrag
8
Some Properties of Generalized Hypergeometric
Thermal Coherent States
123
Dusan Popov
9
Space-Filling Curves for Quantum Control Parameters
133
Fariel Shafee
10
The Spectrum of the Lagrange Velocity
Autocorrelation Function in Confined Anisotropic
Liquids
143
Sakhnenko Elena I and Zatovsky Alexander V.
11
On the Quantum Correction of Black Hole
Thermodynamics
151
Kourosh Nozari and S. Hamid Mehdipour
12
A Graphic Representation of States for Quantum
Copying
Sara Felloni and Giuliano Strini
159
EJTP 10 (2006) 1–10
Electronic Journal of Theoretical Physics
Non-Minimal Coupling Effects of the Ultra-Light
Particles on Photons Velocities in the Radiation
Dominated Era of the Universe
El-Nabulsi Ahmad Rami
∗
Plasma Application Laboratory, Department of Nuclear and
Energy Engineering and Faculty of Mechanical, Energy and Production Engineering,
Cheju National University, Ara-dong 1, Jeju 690-756,
Korea
Received 6 September 2005 , Accepted 27 November 2005, Published 25 May 2006
Abstract: The effect of the ultra-light masses of the order of the Hubble constant, implemented
in Einstein’s field equations from non-minimal coupling and supergravities arguments, on
photons velocities in the radiation dominated epoch of the Universe within the framework
of non-minimal interaction of electromagnetic fields with gravity is developed and discussed in
details.
c Electronic Journal of Theoretical Physics. All rights reserved.
Keywords: Non-minimal coupling, ultra-light masses, effective cosmological constant, photons
velocity.
PACS (2006): 04.40 Nr, 98.80 -k, 95.30 SI
1.
Introduction
A standard result of Einstein’s gravity is that massless particles, in particular photons,
move at light celerity ‘c’. A question worth examining is whether the velocity of massless
photons is shifted when these later propagate in exotic background filled of ultra-light
particles (ULP) of tiny masses in the order of the Hubble constant (m ≈ H). This could
have important cosmological and astrophysical implications. In fact, the possibility of
shifting photon propagation (SPP) in gravitational fields (or non-trivial topologies) is
an interesting prediction of quantum field theory in curved space-time. It appears that
photon propagation may depend on their direction and polarisation, travel with speeds
exceeding the normal speed of light ‘c’ [1]. It is a quantum effect induced by vacuum
∗
atomicnuclearengineering@yahoo.com
2
Electronic Journal of Theoretical Physics 10 (2006) 1–10
polarisation (allowing the photons to exist as a virtual e+ e− pair so that at the quantum
level it is characterized by the Compton wavelength of the electron) and implies that
the Principle of Equivalence does not hold for interacting quantum field theories such
as QED. The propagation of photon in Schwarzschild, Robertson-Walker, gravitational
wave, de Sitter backgrounds, charged black hole were done and remarkable results were
discovered [2]. In each case (except the totally isotropic de Sitter space-time) it was
possible to find directions and polarisations for which the photon velocity exceeds ‘c’.
Generalization to neutrino propagation in a Robertson-Walker metric using the WeinbergSalam model was done in [3]. In a gravitational field, the photon propagation is sensitive
to an anisotropic space-time curvature and may depend on this later [4,5]. Recently,
a series of papers has appeared in which the light velocity varies in the early Universe
and this solves the horizon, monopole and the flatness problems in standard cosmology
[6,7,8,9]. In this work, we will investigate further the consequences of non-minimal
coupling on light velocity in the presence of ultra-light particles.
2.
Non-minimal Coupling, Supergravities Arguments and Einstein Fields Equations
We start with the non-minimal interaction of electromagnetic fields with gravity in the
√
following form L̃ = gξRμν Fαμ F νμ , ξ being the coupling constant, Rμν the Riemann
tensor, g the metric scalar and Fμν = ∂μ Aν − ∂ν Aμ is the electromagnetic strength (Aμ is
the vector potential ). Terms like RF 2 are neglected here. In this way, the field equations
read [10,11,12]:
Dν [F μν − 2ξ (Rαμ F αν − Rαν F αμ )] = 0
where Dν is the covariant derivative. Let us now restrict our attention to the form of the
Riemann tensor we will choose for this work. In a recent paper [13], we introduced, for
some scalar field , a non-minimal coupling between the scalar curvature and the density of
√
the scalar field in the following form L = −ξ gRφ∗ φ, ξ = 1/6 . R is the scalar curvature
and is the complex conjugate of . From a view point of quantum field theory in curved
space-time, it is natural to consider such a non-minimal coupling. In fact, the conformal
case results in an extension of the property of conformal invariance for massless fields,
which is attractive from physical point of view. This parameter describes the strength
of the coupling between the curvature of spacetime and the inflation. Minimal coupling
corresponds to ξ = 0. It was shown that in this case and for a particular scalar negative
complex potential field V (φφ∗ ) = 3/4m2 (ωφ2 φ∗2 − 1),ω being a tiny parameter inspired
from supergravity inflation theories, ultra-light masses ‘m’ are implemented naturally in
Einstein field equations (EFE), leading to a cosmological constant ‘Λ’ in accord with
observations2 . In matter-free background, the scalar curvature was found to be =4Λ̄
2
It has been argued that a non-minimal coupling term-generated by quantum corrections-is to be expected
whenever the space-time curvature is large; in most theories that describe inflationary scenarios, it turns
out that a value of ξ different from zero is unavoidable. As a matter of fact, it seems sensible to consider
an explicit non-minimal coupling in the supergravities inflationary paradigm.
Electronic Journal of Theoretical Physics 10 (2006) 1–10
3
where Λ̄ = Λ−3/4m2 is the effective cosmological constant 3 (in natural units, m ≈ H/c2
where ‘’ is the Planck constant and ‘c’ being the celerity of light). As a result, one
candidate Einstein field equations is( = c = 1):
1
Gμν ≡ Rμν − gμν R = −Λ̄gμν
2
(1)
Λ̄ = Λ − 3m2 /4 ≡ Λ1 + Λ2 is the effective cosmological constant (Λ2 = −3λ̄−2
C where
λC = /mc is the Compton wavelength in natural units). Remark that for Λ1 = 0, the
scalar curvature is negative and the space-time is not Minkowskian. Other field equations
exist also but correspond only for p = ρ/3 (radiation era):
1
3m2
Rμν − gμν R + Λgμν = −8πG p + ρ +
(2a)
uμ uν + pgμν
2
8πG
Rμν
1
− gμν R = −8πG
2
=
≡
≡
≡
3m2
p+ρ+
8πG
Λ
uμ uν + p +
8πG
Λ
gμν
−8πG (p + ρ̄) uμ uν + p +
8πG
Λ
gμν + tμν (m)
−8πG Tμν (p, ρ) +
8πG
−8πG [Tμν (p, ρ) + Tμν (Λ, m)]
−8πG
Tμν
gμν
(2b)
(2c)
(2d)
(2e)
(2f)
p and ρ are the pressure and density of matter,ρ̄ = ρ + 3m2 /8πG, tμν = 3m2 /8πGuμ uν
and:
(3)
Tμν (p, ρ) = (p + ρ) uμ uν + pgμν
1 2
Λ
gμν =
(4)
3m uμ uν + Λgμν
8πG
8πG
Contracting equations (2) with g μν using g μν uμ uν = −1 yields of course. In this way,
the ultra-light masses and the cosmological constant are parts of the matter contents of
the Universe rather than geometrical entities. The radiative field equations (RFE) (2a,b,c,d,e,f) are identical to that of Einstein standard ones but with an additional energy
density ρm = 3m2 /8πG(m ≤ H). One can also refer to equation (4) as the stress-energy
tensor of vacuum and light particles, which is a ”microscopic stress-energy tensor ”. In
fact, the conservation law holds and we have:
Tμν ≡ ∇ν Tμν (p, ρ) + ∇ν Tμν (Λ, m) = 0
(5)
∇ν
Tμν (Λ, m) = tμν +
When Tμν (p, ρ) = 0, the microscopic stress-energy tensor will behave as the macroscopic
one if we assume that:
P (Λ, m) ≡ PΛ = Λ/8πG
(6)
3
In [13], 8πG ≡ κ was set equal to unity.
4
Electronic Journal of Theoretical Physics 10 (2006) 1–10
3m2 − Λ
(7)
8πG
That is, in the microscopic version, if Λ = 0, P = 0 but the density is positive. While
for Λ > 3m2 , the pressure is positive and the density is negative. If 0 < Λ < 3m2 , then
both the pressure and the density are positive.
Before treating the non-minimal coupling scenario, we will discuss briefly the implications of equations (2) in standard cosmology. For this, we consider a homogenous and
isotropic Universe in the radiation dominated epoch described by Friedman-RobertsonWalker line element with scale factor a (t)[14]. The radiative field equations read:
ρ (Λ, m) =
k
8πGρ Λ
ȧ2
+ + m2
+ 2 =
2
a
a
3
3
(8)
ä
8πGρ Λ m2
=−
+ −
(9)
a
3
3
2
k = −1, 0, +1 is the curvature constant for open, flat or closed space-time and dim (Λ) =
dim (m2 ) = length−2 . If the cosmological constant and the ultra-light masses are assumed
to be constant with time, then from the energy conservation law: ρ ∝ a−4 . For zero
density, 2Λ > (<) 3m2 and the acceleration of the Universe accelerate (decelerate) with
time(ä > (<) 0). Combining equation (8) and (9), we can eliminate the density factor
and obtains:
2
3m2
k
2Λ m2
ä ȧ2
2
+ 2+ 2 =
+
= Λ 1+
(10)
≡ Λ̂
a a
a
3
2
3
4Λ
3
where Λ̂ = Λ (1 + 3m2 /4Λ).
If we restrict ourselves to spatially flat universes (k = 0) and we use the definition
da
H ≡ ȧ/a = H0 adT
where time is assumed to be measured in units of Hubble times
T ≡ H0 t, then:
H2
dH
2Λ̂
+2
=
(11)
dT
H0
3H0
Assuming that Λ̂ ∝ T −α , that is the cosmological constant and the ultra-light masses
decrease with time, this model was found to be singular but can significantly be older
than models with constant Λ and m2 [15,16,17,18,19]. For m2 << Λ, Λ̂ → Λ, while
for m2 >> Λ, Λ̂ → 3/4m2 . If for instance, we assume that m2 = β/a2 and Λ = δ ȧ2 /a2 +
ηä/a, β, δ, η are constants [20,21], than from equation (10)4 :
ä
2δ
2η
ȧ2
β 1
=0
(12)
1−
+ 2 1−
+ k−
a
3
a
3
2 a2
which gives:
ȧ2 =
4
2(3−2δ)
3 (β − 2k)
+ Da− 3−2η , D = const.
2 (3 − 2δ)
(13)
The fact that the two terms Λ and m2 play the role of two cosmological constant in the theory, we
have the freedom to choose Λ = δ1 ȧ2 a2 + η1 ä/a + β1 a2 and m2 = δ2 ȧ2 a2 + η2 ä/a + β2 a2 where
δ1,2 , η1,2 , β1,2 are constants. In this work, we simplified our assumptions just to have at the beginning a
simple idea about the effects of the ultra-light masses in the theory.
Electronic Journal of Theoretical Physics 10 (2006) 1–10
5
For D = 0 which corresponds to singular solutions, one finds for flat space-time(k = 0):
a=
3β
t
2 (3 − 2δ)
(14)
2 (3 − 2δ)
3t2
(15)
where δ < 3/2, β > 0. In this way :
m2 =
Λ=
δ
t2
(16)
From equation (8), we find:
3 (δ − 1)
(17)
8πGt2
with 1 < δ < 3/2. In this way, we don’t have an inflationary phase and no horizon
problem appears. From the above equations, we see that the ultra-light masses, the
cosmological constant and the density are independent of the value of β and η. The
Hubble parameter is H = ȧ/a = 1/t and the density matter of the Universe is given by
Ωr = ρ/ρc = δ − 1 < 1/2 where ρc = 3H 2 /8πG is the critical density. The deceleration
parameter is q ≡ −äa/ȧ2 = 0. The density parameter due to vacuum contribution is
ΩΛ = Λ/3H 2 = δ/3 and that due to ultra-light particles contributions is Ωm = m2 /H 2 =
2 (3 − 2δ)/3. In this way ΩT otal = Ωr + Ωm + ΩΛ = 1 as required by inflation [22]. The
ultra-light particles than contribute to the total energy density and their masses decrease
as inverse to time. Note from equations (15) and (16) that Λ = 3δm2 /2 (3 − 2δ) < 9m2 /4.
Finally, note that when the ‘Λ’ and ‘m2 ’ terms dominate the dynamics of equation (8)
with the assumption that the Universe undergoes a long period of evolution during which
the celerity of light changes as c = c0 an , c0 , n =constants [8]:
Λc2 (t)
ȧ2
Λ
2 2
2
2n
=
+ m c (t) →
+ m c2n
(18)
0 a
a2
3
3
ρ=
So at large times, we have a ∝ t−1/n and it was found in [8] that for negative ”n”, there
is a solution to the quasi-lambda problem.
In order to have a very simple idea about the role of the ultra-light masses in the
theory, we suppose that the space-time is flat, that is k = 0 with the following behavior of
the ultra-light masses m2 = β/a2 and the cosmological constant Λ = δ/a2 , β, η=constants
(see footnote 4 ) [23,24,25]. In this case, when ‘Λ’ and ‘m2 ’ terms dominate at large
times the dynamics of equation (8):
Λc2 (t)
δ 2 2n−2
ȧ2
2 2
=
(19)
+ m c (t) = β +
ca
a2
3
3 0
That is a ∝ t−1/n−1 and from [8,26], it is required that n < 0 and c = c0 t−n/n−1 . In
summary, m2 ∝ t2/n−1 and as a result mc ∝ 1/t. Another way to study shifting and
time-varying photons velocities is by using the non-minimal coupling of electromagnetic
fields and gravity.
6
3.
Electronic Journal of Theoretical Physics 10 (2006) 1–10
Varying Photons Velocities From Non-Minimal Coupling
Following [10,11], we admit the existence of a surface S represented by φ (x) = 0. The
wavenumber of the photon trajectories is given by the gradient of its phase kλ = ∇λ φ
where the Faraday tensor vanishes at its hypersurface, that is (Fμν )S = 0. Its derivative
defines a function φμν such that:
(∂λ Fμν )S = (Dλ Fμν )S = kλ φμν
(20)
As a consequence, equation (1) takes the form:
[φμν − 2ξ (Rαμ φαν − Rαν φαμ )] kν = 0
(21)
In the radiation dominated era, it follows that:
4ρ
ρ ρm Λ
μν
αν
μ
+ ρ m u uν +
+
−
φ kν − 2ξ φ . (−χ)
δνμ
3
3
2
χ
ρ ρm Λ
4ρ
ν
αμ
+ ρm u uα +
+
−
kν = 0
+ −φ . (−χ)
δαν
3
3
2
χ
(22)
where χ = 8πG ( = c = 1). For simplicity, we let k0 = kμ uμ and we use the antisymmetric fact of (φμν − φνμ = 2φμν ) as well as Maxwell equations:
φμν kλ + ϕνλ kμ + φλμ kν = 0
(23)
By contracting by k λ the last equation, equation (23) reduces to:
φμν kν =
−2ξχ 4ρ
+ ρm
3
k0 φμν uν ≡ (N ) k0 φμν uν
ρ
ρm
Λ
1 + 4χ 3 + 2 − χ ξ
(24)
Replacing (24) into (23), then:
φμν k 2 + N (−φμν kλ − ϕλμ kν − kν φμλ ) k0 uλ = 0
(25)
The antisymmetric of φμν eliminates all the terms in the parentheses of equation (25) and
we are left with:
(26)
φμν k 2 − N k02 = 0 ⇒ k 2 − N k02 = 0, ∀φμν
The effective photons velocity, in case allρ, ρm , Λ = 0, ∀ξ is then given by [6]:
ρ
i
1 − 4ξΛ − 4ξχ 3
|ki k |
2
v =
=
|1
+
N
|
=
k02
1 + 4ξχ ρ + ρm − Λ
3
3
(27)
χ
and the light velocities is not equal to ‘c’. Adopting equations (15), (16) and (17),
equation (27) takes the form in normal units:
1 − 4 ξ (2δ − 1)
t2
(28)
v2 =
1 + 4 tξ2 (5 − 4δ)
Electronic Journal of Theoretical Physics 10 (2006) 1–10
7
with 1 < δ < 3/2. As a result, for ξ > (<) 0, vphotons < (>) c (light celerity). Adopting
the fact c = c0 an with n < 0, than the photons velocities decreases with time whatever
is the sign of the coupling constant.
An interesting case is when the background is ‘free from matter ’ . From (27) we get:
1
−
4ξΛ
2
(29)
v =
1 + 4ξ (m2 − Λ)
If m2 = 0, than v 2 = 1 which is light celerity in units ( = c = 1). Assuming m2 = β/t2 ,
4Λ = 3m2 or Λ̄ = 0 and as a result R = 0. In this case, equation (28) gives:
3β
2
2
1
−
ξ
1
−
3m
ξ
t
−
3ξβ
2
t
=
v 2 =
=
(30)
1 + m2 ξ 1 + ξ tβ2 t2 + ξβ
Again, if ξ > (<) 0, vphotons < (>) c (light celerity). As a result, the velocity of photons is
affected and shifted by the presence of the ultra-light tiny masses and depends on the sign
of the coupling constant. It doesn’t correspond in fact to null geodesics as in the standard
case. Positive coupling constant corresponds to friction and negative one corresponds to
superluminal case [27,28]. If we adopt the fact c = c0 an , then the photons velocities not
only is shifted but also decrease with time if n < 0 and increase if n > 0.
The constancy of the speed of light is not preserved in this analysis. It depends on
how is filled the background space and how is used a coupling constant different of zero
that modifies presentation of the Einstein’s Field Equations (EFT ), with an additional
term. It is important to notice that the environment where speed of light reaches its
maximum value is the lightest one: the empty space, all because of the constancy of the
speed of light law, which in time, originates the fourth time-coordinate. In our case, the
red-shift coefficient ‘z’ varies with time according to cz = Hr combined to equations (29)
or (30) for a matter free background. ‘r’ is supposed to be the distance form the galaxy
to the earth [14]. If the coupling constant is assume to be positive, one can than have a
cosmological model based on interpretation of the red shift by decrease of the light speed
with time everywhere in the universe beginning with a certain moment of time in the
past. Of course, the agreement with the fundamental physics laws will be completed by
introducing in a future work the evolution of other fundamental constants synchronously
with the variation of the light speed [29].
Finally, we note that recently, growing amount of astrophysical data show important
evidence for statistical and apparent physical association between low-redshift galaxies
and high-redshift quasi-stellar objects suggesting noncosmological origin of their redshift
and as a result failure of classical quasar explanation [30]. The author found analytical
solution of Einstein equations describing bubbles made from axions with periodic interaction potential considered as one of the leading dark matter candidate. Remember that
in our model [13], the ultra-light masses implemented in Einstein field equations enabled
us to solve the ‘missing mass problem’ and as result considered as dark matter candidate.
In Minkowski space, objects at constant proper distance with respect to an observer have
zero redshift. However, in an expanding universe special relativistic concepts do not generally apply. In fact, a galaxy with zero total velocity does not have zero redshift even
8
Electronic Journal of Theoretical Physics 10 (2006) 1–10
in the empty universe case. This demonstrates that cosmological redshifts are not special relativity Doppler shifts [31,32]. It was also proved that Minkowski coordinate and
the Robertson-Walker coordinates (FRW universe) are interchangeable descriptions for
an empty universe. However, velocities in the Minkowski universe are not equivalent to
velocities in the FRW universe because of the different definitions of time and distance in
these two models. A coordinate transform relates velocities in the Minkowski universe to
velocities in the FRW universe. Superluminal recession velocities in the FRW universe
do not violate special relativity because they are not in the observer’s inertial frame.
4.
Conclusions:
In this work, we used the Einstein’s field equations with effective cosmological constant
inspired from non-minimal coupling and supergravities arguments to study the consequences of non-minimal coupling between electromagnetic fields and gravity on light
velocity in the presence of ultra-light particles at the radiation dominated epoch of the
Universe. We showed that if the cosmological constant and the ultra-light square masses
varies as Λ = δ ȧ2 /a2 +ηä/a and m2 = β/a2 , then at singular solutions and for a flat spacetime, Λ, m2 and ρ decreases with time as 1/t2 , the Hubble parameter vary as H = 1/t,
the deceleration parameter is zero and ΩT otal = Ωr +Ωm +ΩΛ = 1 as required by inflation.
As a result, the ultra-light particles than contribute to the total energy density and their
masses decrease as inverse to time. When the ‘Λ’ and ‘m2 ’ terms dominate the dynamics
of our field equations with the assumption that the Universe undergoes a long period of
evolution during which the celerity of light changes as c = c0 an , it was found that at large
times a ∝ t−1/n and that for negative ‘n’, there is a solution to the quasi-lambda problem.
Finally, we studied varying light velocities from non-minimal coupling. We found that
photons velocities depends on the coupling constant and only on δ ∈ (1, 3/2)in a way
that ξ > (<) 0, vphotons < (>) c with m2 > 0.
The model described in this paper could have important implications in various systems, in particular cosmological scenarios, black hole physics and quantum interactions
[33,34,35,36,37]. It is important to discuss the impact of the assumptions established to
reach the model and the fact that a non-Minkowskian space is necessary to obtain speeds
greater than that of light, and that photon could be represented by a positron-electron
pair for characterizing it with the electron’s Compton wavelength.
To further investigate all these issues, further studies will be necessary and work is in
progress.
Acknowledgments:
The author is grateful for the referees for their useful comments and suggestions.
Electronic Journal of Theoretical Physics 10 (2006) 1–10
9
References
[1] I.T. Drummond and S.J. Hathrell, Phys. Rev. D 22, 343 (1980).
[2] R.D. Daniels and G.M. Shore, Nucl. Phys. B 425,634 (1994).
[3] Y. Ohkuwa, Prog. Theor. Phys. 65, 1058 (1981).
[4] R. G. Cai, Nucl. Phys. B 524,639 (1998).
[5] G. M. Shore, Nucl. Phys. B 460, 379 (1996).
[6] J. D. Barrow, astro-ph/9811022.
[7] A. Albrecht and J. Maguiejo, Phys. Rev D 59, 043516 (1999) 043516.
[8] J. D. Barrow and J. Maguiejo, Phys. Lett. B 443, 104-110 (1998).
[9] P. Teyssandier, Annales de la Fondation Louis de Broglie, Volume 29 no 1-2, (2004)
(gr-qc/0303081 ).
[10] M. Novello, L. A. R Olivera and J. M. Salim J.M., Class. Quant. Grav. 7, 51 (1990).
[11] M. Novello and S. D. Jorda, Mod. Phys. Lett. A, vol 4, n19, 1809, (1989).
[12] 1-I. L. Buchbinder, S. D. Odintsov and I. L. Shapiro, Effective Action in Quantum
Gravity, IOP-Publishing Ltd, (1982).
[13] R. A. El-Nabulsi, Phys. Letts. B, Vol. 619, 26 (2005).
[14] S. Weinberg, Gravitation and Cosmology, NY, Wiley (1972).
[15] J. M. Overduin and F. I. Cooperstock, astro-ph/9805260.
[16] J. W. Moffat, Int. J. Mod. Phys. D2, 351 (1993).
[17] J. W. Moffat, Found. Of Phys., 23, 411 (1993).
[18] M. A. Clayton and J. W. Moffat, Phys. Lett.B 460, 263 (1999).
[19] M. A. Clayton and J. W. Moffat, Phys. Lett.B 477, 269 (2000).
[20] A. S. Al-Rawaf, Mod. Phys. Lett. A 13, 429 (1998).
[21] A. I. Arbab, Spacetime and Substance 1(6), 39 (2001).
[22] A. H. Guth, Phys. Rev. D 23, 347 (1981).
[23] M. S. Berman, Nuovo Cimento 74B, 182 (1983).
[24] M. S. Berman, Gen. Rel. Grav. 23, 465 (1991).
[25] M. S. Berman and F. M. Gomide, Gen. Rel. Grav. 20, 191 (1988).
[26] M. S. Berman and L. A. Trevisan, gr-qc/0112011.
[27] D. Syer, astro-ph/9404063.
[28] T. Chiba and K. Kohri, Prog. Theor. Phys. 107, 631 (2002).
[29] V. S. Troitskii, Astrophysics and Space Science 139, 389 (1987).
[30] A. A. Svidzinsky, astro-ph/0409064.
10
Electronic Journal of Theoretical Physics 10 (2006) 1–10
[31] Davis T. M., Lineweaver C. H., 2001, in Durrer R., Garcia-Bellido J., Shaposhnikov
M., eds, Cosmology and Particle Physics 2000, American Institute of Physics conference
proceedings, Volume 555. New York, p. 348.
[32] Davis T. M., Lineweaver C. H., Webb J. K., Am. J. Phys., 71, 358 (2003)
[33] C. F. E. Holzhey and F. Wilzeck, Nucl. Phys. B 380, 447 (1992).
[34] R. A. El-Nabulsi, Chinese Phys. Lett. 23 No 5, 1124-1127 (2006).
[35] R. A. El-Nabulsi, Elec. J. Theor. Phys. 7, 27-46 (2005).
[36] R. A. El-Nabulsi, Elec. J. Theor. Phys. 9, 1-5 (2006).
[37] R. A. El-Nabulsi, Supergravity, Non-mimimal Coupling and Eternal Accelerated
Expansion: Some important Cosmological Features of Ultra-Light Particles and
Induced Cosmological Constant, Submitted to Elec. J. Theor. Phys.
EJTP 3, No. 11 (2006) 11–27
Electronic Journal of Theoretical Physics
A Toy Model of Financial Markets
J. P. Singh and S. Prabakaran
∗ †
Department of Management Studies
Indian Institute of Technology Roorkee
Roorkee 247667, India
Received 27 October 2005 , Accepted 9 January 2006, Published 25 June 2006
Abstract: Several techniques of fundamental physics like quantum mechanics, field theory
and related tools of non-commutative probability, gauge theory, path integral etc. are being
applied for pricing of contemporary financial products and for explaining various phenomena
of financial markets like stock price patterns, critical crashes etc.. In this paper, we apply the
well entrenched methods of quantum mechanics and quantum field theory to the modeling of
the financial markets and the behaviour of stock prices. After defining the various constituents
of the model including creation & annihilation operators and buying & selling operators for
securities, we examine the time evolution of the financial markets and obtain the Hamiltonian
for the trading activities of the market. We finally obtain the probability distribution of stock
prices in terms of the propagators of the evolution equations.
c Electronic Journal of Theoretical Physics. All rights reserved.
Keywords: Economic Physics , Financial Markets , Stock Prices, Quantum Models
PACS (2006): 89.65.Gh , 03.65.-w, 02.50.-r, 05.30.-d, 02.30.Tb
1.
Introductionn
The specialty of “physics” is the study of interactions between the various manifestations of matter and its constituents. The development of this subject over the last
several centuries has led to a gradual refining of our understanding of natural phenomena.
Accompanying this has been a spectacular evolution of sophisticated mathematical tools
for the modeling of complex systems. These analytical tools are versatile enough to find
application not only in point processes involving particles but also aggregates thereof
leading to field theoretic generalizations and condensed matter physics.
Furthermore, with the rapid advancements in the evolution and study of disordered
∗
†
jatinfdm@iitr.ernet.in
Jatinder pal2000@yahoo.com
12
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
systems and the associated phenomena of nonlinearity, chaos, self organized criticality
etc., the importance of generalizations of the extant mathematical apparatus to enhance
its domain of applicability to such disordered systems is cardinal to the further development of science.
A considerable amount of work has already been done and success achieved in the
broad areas of q-deformed harmonic oscillators [1], representations of q-deformed rotation
and Lorentz groups [2-3]. q-deformed quantum stochastic processes have also been studied
with realization of q-white noise on bialgebras [4]. Deformations of the Fokker Planck’s
equation [5], Langevin equation [6] and Levy processes [7-8] have also been analysed and
results reported.
Though at a nascent stage, the winds of convergence of physics and finance are unmistakably perceptible with several concepts of fundamental physics like quantum mechanics,
field theory and related tools of non-commutative probability, gauge theory, path integral
etc. being applied for pricing of contemporary financial products and for explaining various phenomena of financial markets like stock price patterns, critical crashes etc. [8-19].
The origin of the association between physics and finance, though, can be traced way back
to the seminal works of Pareto [20] and Batchlier [21], the former being instrumental in
establishing empirically that the distribution of wealth in several nations follows a power
law with an exponent of 1.5, while the latter pioneered the modeling of speculative prices
by the random walk and Brownian motion. The cardinal contribution of physicists to the
world of finance came from Fischer Black & Myron Scholes through the option pricing
formula [22] which bears their epitaph and which won them the Nobel Prize for economics
in 1997 together with Robert Merton [23]. They obtained closed form expressions for the
pricing of financial derivatives by converting the problem to a heat equation and then
solving it for specific boundary conditions.
The theory of stochastic processes constitutes the “golden thread” that unites the
disciplines of physics and finance. Modeling of non relativistic quantum mechanics as
energy conserving diffusion processes is, by now, well known [24]. Unification of the
general theory of relativity and quantum mechanics to enable a consistent theory of
quantum gravity has also been attempted on “stochastic spaces” [25]. Time evolution of
stock prices has been, by suitable algebraic manipulations, shown to be equivalent to a
diffusion process [26].
Contemporary empirical research into the behavior of stock market price /return patterns has found significant evidence that financial markets exhibit the phenomenon of
anomalous diffusion, primarily superdiffusion, wherein the variance evolves with time
according to a power law tα with α > 1.0. The standard technique for the study of superdiffusive processes is through a stochastic process that evolves according to a Langevin
equation and whose probability distribution function satisfies a nonlinear Fokker Planck
equation [27].
There is an intricate yet natural relationship between the power law tails observed in
stock market data and probability distributions that emanate as the solution of the Fokker
Planck equation. The Fokker Planck equation is known to describe anomalous diffusion
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
13
under time evolution. Empirical results [28-31] establish that temporal changes of several
financial market indices have variances that that are shown to undergo anomalous super
diffusion under time evolution.
One of the most exhaustive set of studies on stock market data in varying dimensions
has been reported in [32-36]. In [36], a phenomenological study was conducted of stock
price fluctuations of individual companies using data from two different databases covering three major US stock markets. The probability distributions of returns over varying
timescales ranging from 5 min. to 4 years were examined. It was observed that for
timescales from 5 minutes upto 16 days the tails of the distributions were well described
by a power law decay. For larger timescales results consistent with a gradual convergence
to Gaussian behaviour was observed. In another study [32] the probability distributions
of the returns on the S & P 500 were computed over varying timescales. It was, again,
seen that the distributions were consistent with an asymptotic power law behaviour with
a slow convergence to Gaussian behaviour. Similar findings were obtained on the analysis
of the NIKKEI and the Hang –Sang indices [32].
Stock market phenomena are assumed to result from complicated interactions among
many degrees of freedom, and thus they were analyzed as random processes and one could
go to the extent of saying that the Efficient Market Hypothesis [37-38] was formulated
with one primary objective – to create a scenario which would justify the use of stochastic
calculus [39] for the modeling of capital markets.
The Efficient Market Hypothesis contemplates a market where all assets are fairly
priced according to the information available and neither buyers nor sellers enjoy any
advantage. Market prices are believed to reflect all public information, both fundamental
and price history and prices move only as sequel to new information entering the market.
Further, the presence of large number of investors is believed to ensure that all prices
are fair. Memory effects, if any at all, are assumed to be extremely short ranging and
dissipate rapidly. Feedback effects on prices are, thus, assumed to be marginal. The
investor community is assumed rational as benchmarked by the traditional concepts of
risk and return.
An immediate corollary to the Efficient Market Hypothesis is the independence of
single period returns, so that they can be modeled as a random walk and the defining
probability distribution, in the limit of the number of observations being large, would be
Gaussian.
Anomalous diffusion is a hallmark of several intensively studied physical systems. It
is observed, for example, in the chaotic dynamics of fluid in rapidly rotating annulus [40],
conservative motion in a periodic potential [41], transport of fluid in a porous media [42],
percolation of gases in porous media [43], crystal growth spreading of thin films under
gravity [44], radiative heat transfer [45], systems exhibiting surface to surface growth [46]
and so on.
Several analogies between physical systems and financial processes have been explored
in the last decade, some of which have already been mentioned above. Perhaps, the most
striking one is that between financial crashes witnessed in stock markets and critical
14
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
phenomena like phase transitions is discussed here to place the main theme of this paper
in its proper perspective.
Stock market crashes are believed to exhibit log periodic oscillations which are characteristic of systems exhibiting discrete scale invariance i.e. invariance through rescaling
by integral powers of some length scale like the Serpinski triangle and other similar fractal
shapes. In the years preceding the infamous crash of October 19, 1987, the S & P market
index was seen to fit the following expression exceedingly precisely [47-48],
(S & P )t = Ω + Γ (tc − t)γ {1 + Ξ cos [θ ln (tc − t) + φ]}
Physicists working in solid state and condensed matter physics would immediately
recognize the analogy of the above expression with the one obtained for critical phenomenon in spin model of ferromagnetism [49]. We briefly elucidate the salient features
of this model. Crystalline solids comprise of atoms arranged in a lattice. Each such
atom generates a magnetic field parallel to the direction of the atom’s spin. In the case
of substances that do not exhibit ferromagnetic character, these spin directions are randomly oriented so that the aggregate magnetic field vanishes. However, in ferromagnetic
substances these spins are polarized in a particular direction resulting in a nonzero aggregate field. Ferromagnetic substances usually exhibit two distinct phases. one in which
the spins orient themselves in a particular direction resulting in an aggregate magnetic
moment at temperatures below a well defined critical temperature tc and the other where
the spins are disoriented with a zero aggregate moment above the critical temperature.
At temperatures below tc , the coupling force between neighboring atoms predominates
resulting in an alignment of spins whereas above tc the additional energy manifests itself
in disorienting (randomizing) the spins.
Renormalization group theory enables us to group these atoms in blocks of spins
whose composite spins are equal to the algebraic sum of the spins of the atoms constituting the block. It then provides that a model involving interactions between these
composite spins of a block can be constructed that replicates the macroscopic properties of the block and yet cannot depend on the size of the block. That is, the system would exhibit a scaling symmetry, which is discrete, if we allow for the finite size
of the atom and continuous otherwise. The magnetic susceptibility of such a mag
, where the symbols have their usual meannetic substance defined by χ (T ) = ∂M
∂B B=0
ing, obeys a power law of the form χ (T ) = Re (T − Tc )α+iβ or equivalently χ (T ) =
(T − Tc )α {1 + β cos [ln (T − Tc )] + O (β 2 )}which is reminiscent of op cited expression for
log periodic oscillations in financial crashes.
Furthermore, the access to enhanced computing power during the last decade has
enabled analysts to try refined methods like the phase space reconstruction methods
for determining the Lyapunov Exponents [50] of stock market price data, besides doing
Rescaled Range Analysis [51] etc. A set of several studies has indicated the existence
of strong evidence that the stock market shows chaotic behavior with fractal return
structures and positive Lyapunov exponents. Results of these studies have unambiguously
established the existence of significant nonlinearities and chaotic behavior in these time
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
15
series [52-55].
In this paper, we attempt one such model. The objective is to apply the well entrenched methods of quantum mechanics and quantum field theory to the study of the
financial markets and the behaviour of stock prices. Section 2, which forms the essence of
this paper, arrives at various results for financial markets by modeling them as quantum
Hamiltonian systems. The probability distribution for stock prices in efficient markets is
also obtained. Section 3 concludes.
1.1 Quantum Model of Financial Markets
We consider an “isolated” financial market comprising of n investors and m type of
securities. The market is “isolated” in the sense that new types of securities are neither created nor are existing ones destroyed. Further, the number of investors is also
constant. The investor i, i = 1, 2, 3.......n is assumed to possess a cash balance of
xi , i = 1, 2, 3.......n (which may be negative, representing borrowings) and yij (z) , i =
1, 2, 3.....n; j = 1, 2, 3....m units of security j at a unit price of z . Obviously, yij ≥ 0, ∀i, j.
Towards constructing a basis for our Hilbert space representing the financial market,
we define a pure state of the system as
|Ψi = |{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n}
(1)
Thus, a pure state represents a state of the market where the entire holdings of cash
and securities of every investor are known with certainty. This represents a complete
measurement of the market and hence, is in conformity with the standard definition of
“pure state” of a system.
A basis for our Hilbert space may then be constituted by the set of all the pure states
of the type (1) i.e.
Ψ = {|{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n}}
(2)
The elements of this basis set Ψ satisfy the orthogonality condition Ψi | Ψj = δij with
respect to the scalar product defined in the sequel. The orthogonality condition makes
sense in the financial world – it implies that if a market is in a pure state |Ψi then it
cannot be in any other pure state.
However, a complete measurement of the market is, obviously, not practicable in real
life. At any point in time, we are likely to have certain information only about a fraction
of the market constituents. Hence, the instantaneous state of the market |ψ (t) may be
represented by a linear combination of the pure states |Ψl (t) i.e.
|ψ (t) =
Cl |Ψl (t)
(3)
l
We endow our Hilbert space H with the scalar product
ψ (t) | ξ (t) =
Cl∗ Dm Ψl (t) | Ψm (t) =
Cl∗ Dm δlm =
Cl∗ Dl
l,m
l,m
l
(4)
16
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
where we have assumed the orthogonality of the pure states.
The components of the state space vector |ψ (t)are given byCl = Ψl (t) | ψ (t) and
are related to the probability of finding the market in the pure state |Ψl (t).
Since our basis comprises of all possible measurable pure states, the completeness of
the basis is ensured so that
I=
|Ψl (t) Ψl (t)|
(5)
l
In analogy with the no particle state or ground state in quantum mechanics, we can
define a ground state of our financial market as
|0 = |xi = 0, yij (z) = 0∀i, j, z
(6)
i.e. the ground state is the market state in which no investor has any cash balances nor
any securities. This state is, obviously, a pure state being fully measurable and would
also not evolve in time since no trade can take place in this market.
We define the cash and security coordinate operators x̂i & ŷij (z) by their action on
the basis state (1) to provide respectively the balances of cash and the j th security (at
price z) with the ith investor as the eigenvalues i.e.
x̂i |{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n} = xi |{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n}
(7)
ŷij (z) |{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n} = yij (z) |{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n}
(8)
A cash translation operator T̂i (z) is also defined by the following
T̂i (z ′ ) |{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n} = |{xi + z ′ , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n} (9)
i.e. it transfers an amount of cash z to the ith investor.
The operator T̂i (z) obviously satisfies the following properties
T̂i (z1 ) T̂i (z2 ) = T̂i (z1 + z2 )
(10)
T̂i (0) = Iˆ
(11)
T̂i (z) , x̂j = T̂i (z) x̂j − x̂j T̂i (z) = −zδij T̂i (z)
(12)
T̂i↑ (z) = T̂i (−z)
(13)
Towards obtaining
an explicit representation of the cash translation operator, we assume
dT̂i (z)
p̂i = dz as the generator of infinitesimal cash translations dz to the investor i.
z=0
Expanding T̂i (z) as a Taylor’s series and using eqs. (10), (11) we have
T̂i (0) +
T̂i (dz) − 1 T̂i (z)
dT̂i (z)
T̂i (z + dz) − T̂i (z)
= lim
= lim
= lim
dz→0
dz→0
dz→0
dz
dz
dz
dT̂i (z)
dz z=0
dz
dz... − 1 T̂i (z)
= p̂i T̂i (z)
(14)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
17
with the solution T̂i (z) = ezp̂i . Furthermore, we have (suppressing the yij indices for the
sake of brevity)
|{xi + dz, i = 1, 2, ...n} = T̂i (dz) |{xi , i = 1, 2, ...n} = T̂i (0) +
dT̂i (z)
dz
= [I + p̂i dz...] |{xi , i = 1, 2, ...n}
z=0
dz... |{xi , i = 1, 2, ...n}
(15)
Hence,
∂{xi ,i=1,2,...n}|ψ
∂xi
=
p̂↑i
= lim
dz→0
{xi +dz,i=1,2,...n}|ψ−{xi ,i=1,2,...n}|ψ
dz
{xi , i = 1, 2, ...n} | ψ = −p̂i {xi , i = 1, 2, ...n} | ψ ⇒ p̂i =
∂
−z ∂x
so that T̂i (z) = e
i
(16)
− ∂x∂ i
. The following commutation rule holds between x̂i and p̂i
[x̂i , p̂j ] = δij
(17)
The condition of an isolated market ensures that the basis and hence the Hilbert space
does not depend on time. This implies that the temporal evolution of the system is
unitary.
Creation & Annihilation Operators for Securities
We define âij (z)as the annihilation operator of the security j from the portfolio of
investor i for a price z i.e. when operator âij (z) acts on a state, the number of units of
security j is reduced by one from the portfolio of investor i for a price z.Similarly, we
define creation operators â↑ij (z) as the adjoint of the annihilation operators that increase
the number of units of security j in the portfolio of investor i for a price z.The precise
action of these operators on a state vector is defined by the following
âij (z) |{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n} =
y ij (z).z |{xi , {yij (z) − 1, j = 1, 2, ...m} , i = 1, 2, ...n}
(18)
and
â↑ij (z) |{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n} =
(yij (z) + 1)z |{xi , {yij (z) + 1, j = 1, 2, ...m} , i = 1, 2, ...n}
(19)
where the factor ‘z’ has been introduced in the eigenvalues to ensure “scale invariance”
of the theory.
These operators satisfy the following commutation relations:
↑
′
âij (z), âij (z ) = zδzz′ δik δjl
(20)
and
′
â↑ij (z), â↑ij (z ′ )
[âij (z), âkl (z )] =
=0
T̂i (z) , âjk (z ′ ) = T̂i (z) , â↑jk (z ′ ) = 0
T̂i↑ (z) , âjk (z ′ ) = T̂i↑ (z) , â↑jk (z ′ ) = 0
(21)
(22)
(23)
18
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
Further more
â↑ij (z) âij (z) |{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n} = zyij (z) |{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n}
(24)
which implies that the number operator would be
â↑ij (z) âij (z)
ŷij (z) =
z
(25)
Using the aforesaid operators we can construct an arbitrary basis state from the ground
state as follows
|{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n} =
n
T̂i (xi )
i=1
m
j=1 {z,yij (z)∈N}
yij (z)
â↑ij (z)
|0
(26)
Buying and selling operators
The buying (selling) operation of a security is, in each case, a composite operation
consisting of the following:i. the creation (annihilation) of a security at the relevant price z; and
ii. the decrease (increase) in the cash balance by z of the investor undertaking the
trade.
Hence we can define the buying (selling) operator as composite of the cash translation
operator and the creation (annihilation) operators for securities as follows:b̂↑ij (z) = â↑ij (z) T̂i↑ (z) = â↑ij (z) T̂i (−z)
(27)
for the “buying” operation and
b̂ij (z) = âij (z) T̂i (z)
(28)
for the “selling” operation. These operators satisfy the following commutation rules
b̂ij (z) , b̂↑kl (z ′ ) = zδzz′ δik δjl
↑
↑
′
b̂ij (z) , b̂kl (z ) = b̂ij (z) , bkl (z ) = 0
′
b̂ij (z) , T̂k (z ′ ) = b̂↑ij (z) , T̂k (z ′ ) = 0
↑
b̂ij (z) , x̂k = zδik b̂↑ij (z)
b̂ij (z) , x̂k = −zδik b̂ij (z)
(29)
(30)
(31)
(32)
(33)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
19
1.2 Temporal Evolution of Financial Markets
In analogy with quantum mechanics, we mandate that the state of the market at a given
instant of time ‘t’, is represented by a vector in the Hilbert space H whose components
determine the statistical nature of the market. Hence the temporal evolution of the
market in essentially determined by the evolution of this vector with the flow of time. In
the Schrödinger picture, the time evolution of a system can be characterized through a
∧
unitary evolution operator U (t, t0 ) in H, that acts on the initial state |ψ (t0 ) to transform
it to |ψ (t) i.e
|ψ (t) = Û (t, t0 ) |ψ (t0 )
(34)
The assumption of the market being isolated and hence
Ψ = {|{xi , {yij (z) , j = 1, 2, ...m} , i = 1, 2, ...n}} being a complete basis at all times, and
|Cl (t)|2 = 1, ∀t together with the group property
the conservation of probability i.e.
∧
l
of U (t, t0 ) implies that the temporal evolution is unitary i.e.
U (t, t0 ) U ↑ (t, t0 ) = U ↑ (t, t0 ) U (t, t0 ) = 1
(35)
∂
Furthermore Û (t0 , t0 ) = 1. Defining the Hamiltonian Ĥ (t) = i ∂t
Û (t, t0 )
t=t0
as the in-
finitesimal generator of time translations (evolution) we obtain, through a Taylor’s ex
Û (t, t0 ) δt + ...or
pansion up to first order Û (t + δt, t0 ) = Û (t, t0 ) + ∂∂tÛ (t + δt, t)
δt=0
∂ Û (t, t0 )
Û (t + δt, t0 ) − Û (t, t0 )
= lim
= −iĤ (t) Û (t, t0 )
δt→0
∂t
δt
−
4t
(36)
Ĥ(t)dt
where time ordering of the operators
with the immediate solution Û (t, t0 ) = e t0
constituting the Hamiltonian is assumed.
Before progressing further with the development of the model, some observations are in
order about the theory developed thus far.
(1) In standard quantum mechanics, Ĥ
(t) is usually a bounded operator and hence the
4
− tt Ĥ(t)dt
exponential series in Û (t, t0 ) = e 0
converges so that its approximation to first
∂|ψ(t)
order is acceptable giving i ∂t = Ĥ (t) |ψ (t) which is the Schrödinger equation
of wave mechanics. This may not always be the case in financial markets.
(2) Since time evolution of financial market, essentially, occurs through trades in securities, it is appropriate to infer that the Hamiltonian represents the trading activities
of the market.
(3) In order that the evolution operator Û (t, t0 ) is well defined, we mandate that the
Hilbert space His so constructed that the kernel of Û (t, t0 ) is empty.
1.3 Modeling Time Value of Money
Time value of money and interest rate instruments are classically modeled through
the
4
dB(t)
r(t)dt
first order differential equation dt = r (t) B (t)with the solution B (t) = B (0) e
.
20
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
A possible candidate for the Hamiltonian function H(in the classical picture) that would
generate this temporal development as the equations of motion is
H (x, p; t) =
n
Hi (xi , pi ; t) =
i=1
n
ri (t) xi (t) pi (t)
(37)
i=1
This Hamiltonian leads to the following equations of motion
dpi (t)
∂Hi (xi , pi ; t)
= −ri (t) pi (t)
=−
dt
∂xi
dxi (t)
∂Hi (xi , pi ; t)
= ri (t) xi (t) ,
=
dt
∂pi
(38)
While the interpretation of first of these equations is straightforward being the growth
of cash reserves of the ith investor with the instantaneous rate ri (t), the implications of
second equation are more subtle. To provide a financial logic to this equation, we note
that pi is the infinitesimal generator of cash translations in the classical picture and hence
dpi (t)
i ,pi ;t)
= −ri (t) pi (t) represents the rate of change of the cash translations
= − ∂Hi (x
dt
∂xi
generator which, given a fixed rate of growth of cash, would decrease with the amount of
cash translations.
Using the Weyl formalism for transformation from the classical to the quantum picture, we require that the quantum mechanical analog of H (x, p; t) be Hermitian and
symmetric in its component operators. Hence, we postulate the ansatz
n
n
n
iri (t)
1ˆ
Ĥ (x̂ (t) , p̂ (t) ; t) =
(x̂i p̂i + p̂i x̂i ) =
Ĥ (x̂i (t) , p̂i (t) ; t) =
iri (t) x̂i p̂i + I
2
2
i=1
i=1
i=1
(39)
for the quantum mechanical Hamiltonian representing the time value of money, so that
the time development operator is
Û (t, t0 ) = e
−i
4t
n 4t
2
Ĥ(t)dt
=e
t0
i=1 t0
ri (t)xi (p̂i + 12 Iˆ)dt
(40)
which may be evaluated using standard methods like Green’s functions and Feynmann
propagator theory.
1.4 Representation of Trading Activity
Let us consider a deal in which an investor ‘i’ buys a security ‘j’ at a price of ‘z’ units
and immediately thereafter sells the same security to another investor ‘k’ at a price of
‘z ′ ’ units and credits/debits the difference z ′ − z to his cash account. The composite
transaction will, in our operator formalism, take the form b̂ij (z ′ ) b̂↑ij (z). In analogy with
this argument, we can represent the Hamiltonian for trading activity of the market as
∞
∞
dz dz ′
HT r (t) =
hijkl (ξ, t) b̂↑ij (z) b̂kl (z ′ )
′
z
z
i,j,k,l
0
′
0
where ξ = ln zz ensures that the amplitudes are scale invariant.
(41)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
21
1.5 Probability Distribution of Stock Prices
We now derive the probability distribution of stock prices in market equilibrium and show
that the prices follow a lognormal distribution, thereby vindicating the efficacy of this
model.
For this purpose, we assume that an investor i = α buys one unit of a security j = β
at time t = ti for a price z. We need to ascertain the probability PT (z ′ |z )i.e. the
probability of the security j = β having a price z ′ at time tf = ti + T .We assume that
during the period tf − ti , investor α holds exactly one unit of β and that before ti and
after tf ,α
holdsno unit of β.
z
(ti ) be the state that represents investor α holding exactly one unit of β at a
Let ψαβ
z
↑
price z at time ti in the Hilbert space H. Hence, we have ψαβ (ti ) = b̂αβ (z) ψαβ (ti ) where
ψ
(t
)
αβ i is the state that represents investor α not holding any unit of β. This also im
plies that b̂αβ (z) ψαβ (ti ) = 0.Let us assume that the final state corresponding to the
z
z
initial state ψαβ (ti ) is represented by ψαβ (tf ) so that
tf
4
−i Ĥdt
z
z
↑
ψαβ (tf ) = Û (ti , tf ) ψαβ (ti ) = e ti
b̂αβ (z) ψαβ (ti )
(42)
z′
The amplitudes
of ψαβ (tf ) are determined in the usual way by taking scalar product
z
z′
(tf ) and we have, for the matrix elements of the propagator
(tf ) ψαβ
ψαβ
tf
4
−i Ĥdt
′
′
ti
G (z , tf ; z, ti ) = ψαβ (tf ) b̂αβ (z ) e
b̂↑αβ (z) ψαβ (ti )
(43)
In this case, the trading Hamiltonian will contain creation and annihilation operators
relating to the investor α and those relating to the securityβ i.e., it will be of the form
∞
∞
dz dz ′
ĤT r (t) =
h
(ξ, t) b̂↑αβ (z) b̂kl (z ′ )
′ αβkl
z
z
k,l
0
(44)
0
We further make the assumption that the amplitudes can be approximated by their first
two moments about ξ = 0, being sharply peaked about z ′ = zsince, in the timescales
being considered, most trades would occur around z. Hence, we have
hαβkl ∼ Ωαβkl (t) − iξ −1 Ξαβkl (t) δ (ξ)
(45)
′
′
−1
′
−1
= zz − 1
= z′z−z to first order and
Noting that ξ = ln zz , we have ξ −1 = ln zz
−1
′
d ln zz
′
δ (ξ) = δ ln zz = δ (z ′ − z)
= z ′ δ (z ′ − z). Using these results and eqs. (44)
dz ′
& (45), we obtain
∞
∞
dz dz ′
z
′
ĤT r (t) ∼
Ξαβkl (t) b̂↑αβ (z) b̂kl (z ′ )
zδ (z − z) Ωαβkl (t) − i ′
′
z
z
z
−
z
k,l
0
0
22
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
dz ↑
∂
b̂αβ (z) Ωαβkl (t) + izΞαβkl (t)
b̂kl (z ′ )
=
z
∂z
k,l
∞
(46)
0
∂
and hence it can be
We note that this expression for the Hamiltonian is linear in ∂z
diagonalized in the “momentum space” through a Fourier transformation and we have
∞ ∞ ′ ∞
1 dz
dz
ĤT r (t) =
dpb̂↑αβ (z) [Ωαβkl (t) + iΞαβkl (t) p] b̂kl (z ′ ) eipξ
(47)
′
2π k,l
z
z
−∞
0
0
The assumption of market equilibrium implies that the Hamiltonian should be independent of time over the relevant timescales that would be much smaller than those
determining aggregate market behaviour so that we may write eq. (43) as
(48)
G (z ′ , tf ; z, ti ) = ψαβ (ti ) b̂αβ (z ′ ) e−iĤ(ti )T b̂↑αβ (z) ψαβ (ti )
Because of the Hamiltonian being diagonal in momentum space, it is more convenient to
work in momentum space for evaluating the propagators and we have, for the equivalent
of eq. (48) in momentum space as
⎡
⎤
⎤
⎡∞
∞ dz ′ ′
′
dz −ipln(z/κ) ↑
e
G̃ (p′ , p; T, ti ) = ψαβ (ti ) ⎣
eip ln(z /κ) b̂αβ (z ′ )⎦ e−iĤ(ti )T ⎣
b̂αβ (z)⎦ b̂↑αβ (z) ψαβ (ti )
′
z
z
0
0
(49)
To solve the problem further, we make use of second order perturbation theory. The first
step is to split the Hamiltonian into components as follows
∞
∞
dz ↑
dz ↑
Ĥ (t) =
b̂ (z) [Ωαβαl (t) + Ξαβαl (t) p] b̂αβ (z) +
b̂ (z) [Ωαβkl (t) + iΞαβkl (t) p] b̂kl (z ′ )
z αβ
z αβ
l
k,l,k=α 0
0
(50)
Let Ei be the energy eigenstate of the unperturbed Hamiltonian i.e. of the state of the
market before the purchase of security β by the investor α , then the energy eigenstate
of the Hamiltonian Ĥ (ti )i.e. after the purchase of security β by the investor α will be of
the form Ep = Ei + [Ωαβαl (ti ) + iΞαβαl (ti ) p] − ip2 σ 2 where the second term represents
l
the impact on the energy eigenstates of the transactions involving investor α or security
β and the last term is the second order perturbation term due to the overall fluctuations
of the market. Substituting this value of Ep in eq. (49) and noting that the Hamiltonian
and hence the propagator G̃ (p′ , p; T, ti )is also diagonal in “momentum space”, we have
G̃ (p′ , p; T, ti ) ∼ 2πδ (p′ − p) e−iT Ep = 2πδ (p′ − p) e−iT [Ei +Ω(ti )+iΞ(ti )p−ip
2 σ2
where
Ωαβαl (ti ) = Ω (ti ) ,
l
]
(51)
Ξαβαl (ti ) = Ξ (ti ) .
l
Inverting back to “coordinate space”, we obtain
′ z +ΞT 2
∞
/ )
)
1
eiT (Ei +Ω) (ln(z 4σ
2T
−iT [Ei +Ω(ti )+iΞ(ti )p−ip2 σ 2 ]−ip ln z ′ /z
′
√
G (z , tf ; z, ti ) =
e
∼
dpe
2π
2σ πT
−∞
(52)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
23
The probability PT (z ′ |z )i.e. the probability of the security j = β having a price z ′ at
time tf = ti + T will then be proportional to the square of the above amplitude and
hence, we finally obtain
2
PT (z ′ |z ) α |G (z ′ , tf ; z, ti )| = 4πσ 2 T
−1
e
(In(z′ /z)+ΞT )2
2σ 2 T
(53)
which agrees perfectly with the standard stochastic theory of finance wherein stock returns are modeled extensively through lognormal distributions.
2.
Conclusions
The following interesting observations emanate from the above analysis:-
(1) Eq. (53), on comparison with the standard expression for probability distribution
of stock price in the conventional stochastic calculus based approach to the Black Scholes
formula, identifies Ξ with the expected return on stock. This return is independent of
the eigenvalue Ei and hence, the state of the market. A financial interpretation of this
could be that the stock returns are dependent on the performance of the company and
independent of market dynamics.
(2) Independence of stock returns of the market dynamics would, however, mandate
that the stock volatility measured by the standard deviation σ is related to the stock
market dynamics which seems justified since higher trading volumes would imply greater
volatility and vice versa.
(3) If we define the uncertainty of measurement of a random variate by its standard
deviation, then, from eqs. (52) & (53), we have the uncertainty
for the stock price process
√
zand its Fourier conjugate p, after a time T , as σz = σ T and σp = 2σ1√T so that σz σp = 21
as it should be, since the distribution of z is assumed Gaussian in the aforesaid analysis.
(4) Furthermore, σσpz = 2σ 2 T which enables the identification of σ 2 as the reciprocal
of the mass and hence, the inertia of the stock price process. It is intriguing to note that
the same analogy follows through another completely independent analysis i.e. the Black
= −rSt ∂C
−
Scholes equation for the option price in its standard form is given by ∂C
∂t
∂S
2
1 2 2∂ C
σ St ∂S 2 + rC, where C (St , t) denotes the instantaneous price of a call option with
2
exercise price E at any time t before maturity when the price per unit of the underlying
2
= −r ∂C
− 21 σ 2 ∂∂xC2 + 12 σ 2 ∂C
+ rC
is St . Making the substitution St = ex we obtain ∂C
∂t
∂x
∂x
which, when compared with the standard quantum mechanical Hamiltonian in one degree
of freedom identifies σ 2 as the reciprocal of the mass of the underlying system.
Contemporary quantitative finance is dominated by stochastic modeling of market
behaviour. These models are essentially in the nature of tools of data analysis that aim
to predict future events by applying probabilistic methods to historical data. Empirical
evidence testifies that probability distributions of stock returns are negatively skewed,
have fat tails and show leptokurtosis [56]. Some of the features of empirical distributions
of stock prices are modeled through Levy distributions [57-60], stochastic volatility [61]
or cumulant expansions around the lognormal case. Each of these models, however,
24
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
attempts to empirically attune the model parameters to fit observed data and hence, is
equivalent to interpolating or extrapolating observed data in one form or the other.
Hence, stochastic models fail to take cognizance of causal factors that get submerged
in the superficial patterns exhibited by the avalanche of data being analysed. In actual
fact, every new price determination of a security and hence, the fluctuation of prices is
attributable to a new trade in the relevant security at that price. The “trading process”
therefore manifests itself as a price history of a security. The fundamental limitation of
stochastic tools in simulating extended memory effects is circumvented by this approach.
An attempt has also been made through this “toy model” to establish that a quantum
mechanical version of financial markets results in a temporal evolution of the probability distribution analogous to that of simple stochastic systems. Stochastic models also
lack ability to accommodate collective effects like phase coherency in lasers that could,
possibly, be built into this quantized description.
It need be emphasized here that the above is purely a phenomenological model for
modeling stock behavior. It is fair to say that the current stage of research in financial
processes is dominated by the postulation of phenomenological models that attempt to
explain a limited set of market behavior. There is a strong reason for this. A financial
market consists of a huge number of market players. Each of them is endowed with
his own set of beliefs about rational behavior and it is this set of beliefs that govern
his actions. The market, therefore, invariably generates a heterogeneous response to
any stimulus. Furthermore, “rationality” mandates that every market player should have
knowledge and understanding about the “rationality” of all other players and should take
full cognizance in modeling his response to the market. This logic would extend to each
and every market player so that we have a situation where every market player should have
knowledge about the beliefs of every other player who should have knowledge of beliefs
of every other player and so on. We, thus, end up with an infinitely complicated problem
that would defy a solution even with the most sophisticated mathematical procedures.
Additionally, unlike as there is in physics, financial economics does not possess a basic
set of postulates like General Relativity and Quantum Mechanics that find homogeneous
applicability to all systems in their domain of validity.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
25
References
[1] A. J. Macfarlane, J.Phys., A 22, (1989), 4581;
[2] L.C. Biedenharn, J.Phys., A 22, (1989), L873;
[3] S. Zakrzewski, J.Phys., A 31, (1998), 2929 and references therein; Shahn Majid, J.
Math. Phy., 34, (1993), 2045;
[4] Michael Schurmann, Comm. Math. Phy., 140, (1991), 589;
[5] C. Blecken and K.A. Muttalib, J.Phys., A 31, (1998), 2123;
[6] J. P. Singh, Ind. J. Phys., 76, (2002), 285;
[7] U. Franz and R. Schott, J. Phys., A 31, (1998), 1395;
[8] V.I. Man’ko et al, Phy. Lett., A 176, (1993), 173; V.I. Man’ko and R.Vileta Mendes,
J.Phys., A 31, (1998), 6037;
[9] W. Paul & J. Nagel, Stochastic Processes, Springer, (1999);
[10] J. Voit, The Statistical Mechanics of Financial Markets, Springer, (2001);
[11] Jean-Philippe Bouchard & Marc Potters, Theory of Financial Risks, Publication by
the Press Syndicate of the University of Cambridge, (2000);
[12] J. Maskawa, Hamiltonian in Financial Markets, arXiv:cond-mat/0011149 v1, 9 Nov
2000;
[13] Z. Burda et al, Is Econophysics a Solid Science?, arXiv:cond-mat/0301069 v1, 8 Jan
2003;
[14] A. Dragulescu, Application of Physics to Economics and Finance: Money, Income,
Wealth and the Stock Market, arXiv:cond-mat/0307341 v2, 16 July 2003;
[15] A. Dragulescu & M. Yakovenko, Statistical Mechanics of Money, arXiv:condmat/0001432 v4, 4 Mar 2000;
[16] B. Baaquie et al, Quantum Mechanics, Path Integration and Option Pricing:
Reducing the Complexity of Finance, arXiv:cond-mat/0208191v2, 11 Aug 2002;
[17] G. Bonanno et al, Levels of Complexity in Financial markets, arXiv:condmat/0104369 v1, 19 Apr 2001;
[18] A. Dragulescu, & M. Yakovenko, Statistical Mechanics of money, income and wealth:
A Short Survey, arXiv:cond-mat/0211175 v1, 9 Nov 2002;
[19] J. Doyne Farmer, Physics Attempt to Scale the Ivory Tower of Finance, adaporg/9912002 10 Dec 1999;
[20] V. Pareto, Cours d’Economie Politique (Lausannes and Paris),(1897);
[21] L. Batchlier, Annelas Scientifiques de l’Normal Superieure III-17 21-86,(1900); P.
Cootner, The Random Character of Stock Market Prices (Cambridge, MA: MIT Press
) Reprint ,(1964);
[22] F. Black & M. Scholes, Journal of Political Economy, 81, (1973), 637;
[23] R. C. Merton, Journal of Financial Economics, (1976), 125;
[24] M. Nagasawa, Stochastic Processes in Quantum Physics, Birkhauser, (2000);
26
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
[25] E. Purgovecki, Stochastic Quantum Mechanics and Quantum Spacetime:
A
consistent unification of relativity and quantum theory based on stochastic spaces
Reidel, Dordrecht, 1st printing 1984, revised printing, 1986
[26] M. Baxter & E. Rennie, Financial Calculus, Cambridge University Press, (1992).
[27] H. Risken, The Fokker Planck Equation, Springer, (1996);
[28] R.N. Mantegna & H.E. Stanley, An Introduction to Econophysics, Cambridge, (2000);
[29] M.M. Dacrrogna et al, J. Int’l Money & Finance, 12, (1993), 413;
[30] R.N. Mantegna & H.E. Stanley, Nature, 383, (1996), 587;
[31] R.N. Mantegna, Physica A, 179, (1991), 232;
[32] P. Gopikrishnan et al, Phys. Rev. E 60, (1999), 5305;
[33] P. Gopikrishnan et al, Phys. Rev. E 62, (2000) R4493;
[34] P. Gopikrishnan et al, Physica A, 299, (2001), 137;
[35] P. Gopikrishnan et al, Phys. Rev. E 60, (1999) 5305;
[36] V. Plerou et al, Phys. Rev. E 60, (1999) 6519;
[37] W.F. Sharpe, Portfolio Theory & Capital Markets, McGraw Hill, (1970);
[38] E.J. Elton & M.J. Gruber, Modern Portfolio Theory, & Investment Analysis, Wiley,
(1981);
[39] S.M. Ross, Stochastic Processes, John Wiley, (1999);
[40] A. Ott el al., Phys. Rev.Lett. 65, (1990) 2201; J.P.Bouchaud et al., J.Phys.(France)
II 1, (1991), 1465 ; C.-K. Peng et al., Phys. Rev. Lett. 70, (1993), 1343; R.N Mantegna
and H.E Stanley, Nature (London) 376, (1995) 46; T.H.Solomon et al., Phys. Rev.
Lett.71, 3975 (1993); F.Bardou et al., ibid. 72, (1994), 203 ;
[41] J. Klafter and G. Zumofen, Phys. Rev E 49, (1994), 4873,4;
[42] H. Spohn, J. Phys. (France) I 3, (1993), 69;
[43] M. Muskat, The Flow of Homogeneous Fluids Through Porous Media (McGraw-Hill,
New York, (1937);
[44] J. Buckmaster, J. Fluid Mech. 81 (1995), 735;
[45] E.W. Larsen and G.C. Pomraning, SIAM J. Appl. Math. 39, (1980), 201;
[46] W.L Kath, Physica D 12, (1984), 375;
[47] J. Feigenbaum & P. G. O. Freund, Int. J Mod. Phys. B 10, (1996), 3737;
[48] D Sornette, A. Johansen & J.P. Bouchaud, J. Phys. I (France ) 6, (1996), 167;
[49] D. Sornette, Phys. Rep. 297, (1998), 239;
[50] A. Wolf, J.B. Swift, S.L. Swinney & J.A. Vastano, Determining Lyapunov Exponents
From a Time Series, Physica 16D, (1985), 285;
[51] B.B. Mandlebrot, The Fractal Geometry of Nature, Freeman Press, (1977);
[52] G. DeBoek, Ed., Trading on the Edge, Wiley, (1994);
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 11–27
27
[53] J.G. DeGooijer, Testing Nonlinearities in World Stock Market Prices, Economics
Letters, 31, (1989);
[54] E. Peters, A Chaotic Attractor for the S&P 500, Financial Analysts Journal,
March/April, (1991);
[55] E. Peters, Fractal Structure in the Capital Markets, Financial Analysts Journal,
July/August, 31, (1989); P. Cootner, Ed., The Random Character of Stock Market
Prices, Cambridge MIT Press, (1964);
[56] E. Peters, Chaos & Order in the Capital Markets, Wiley, (1996) and references
therein;
[57] L. Andersen L & J. Andreasen, Review of Derivatives Research, 4, (2000), 231;
[58] J.P. Bouchaud et al, Risk 93, (1996), 61;
[59] J.P. Bouchaud & M.Potters, Theory of Financial Risks, Cambridge, (2000);
[60] E. Eberlein et al, Journal of Business 71(3), (1998), 371;
[61] B. Dupire, RISK Magazine, 8, January 1994;
EJTP 3, No. 11 (2006) 29–32
Electronic Journal of Theoretical Physics
Rayleigh process and matrix elements for the
one-dimensional harmonic oscillator
J.H. Caltenco1∗ , J.L. López-Bonilla1† , J. Morales2‡
1
Escuela Superior de Ingenierı́a Mecánica y Eléctrica
Instituto Politécnico Nacional
Edif. Z, Acc. 3-3er Piso Col. Lindavista C.P. 07738 México D.F.
2
Area de Fı́sica AMA, CBI
Universidad Autónoma Metropolitana-Azc.
Apdo. Postal 16-306, CP 02200 México DF
Received 21 December 2005 , Accepted 2 February 2005, Published 25 June 2006
Abstract: We show that, the matrix elements m |e−γ x |n for the one-dimensional harmonic
oscillator have application in Markov process theory, permitting thus to resolve the FokkerPlanck equation for the two-dimensional probability density corresponding to Rayleigh case.
c Electronic Journal of Theoretical Physics. All rights reserved.
Keywords: Matrix elements, One-dimensional harmonic oscillator, Markov process theory,
Fokker-Planck equation
PACS (2006): 03.65.Ta , 03.65.Ge ,02.50.Ga, 46.25.Cc, 02.70.Ns
1.
Introduction
In [1-4] were calculated the matrix elements:
∝
∗
−γx
f (γ) = m| e
ψm
(x) e−γ x ψn (x)dx
|n =
(1)
−∝
for the harmonic oscillator in one dimension, where γ ≥ 0 is an arbitrary parameter.
Then, it was deduced the following result for m ≥ n:
2
m−n 2
γ / m−n
n!
γ
γ
f (γ) =
e 4 Ln
−√
(2)
−
m!
2
2
∗
†
‡
hcalte@maya.esimez.ipn.mx
jlopezb@ipn,mx, lopezbjl@hotmail.com
jmr@correo.azc.uam.mx
30
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 29–32
in terms of the associated Laguerre polynomials Lqn .
It is interesting to observe that the 2nd order differential equation [5-8] defining to
p
Lq permits to show, via (2), that f (γ)satisfies the equation:
1
1 df
d2 f
+
− 2 γ 4 + 4Aγ 2 + 4Q f = 0
2
dγ
γ dγ 4γ
(3)
where A = m + n + 1 and Q = (m − n)2 . That is, (2) is solution of (3), with which it is
possible to obtain [9] the Morse’s radial wave function [10].
In Sec. 2, f (γ)is employed to resolve the nonstationary stochastic Fokker-Planck
equation (FPE) [11] for the Rayleigh distribution.
2.
Two-Dimensional Probability Density Associated to Rayleigh
Process.
The equation (3) has the structure:
D1 (γ)f =
Q
γ2
+A+ 2
4
γ
f
(4)
where it appears the important Bessel’s operator [12]:
DC (γ) =
d2
C d
+
2
dγ
γ dγ
(5)
for the case C = 1. The operator (5) has interesting applications in hydrodynamics,
the theory of subharmonic functions, electrostatics, the Euler-Poisson-Darboux equation,
elasticity, the generalized radiation problem, quantum mechanics and generalized axially
symmetric potential. Here we shall show that, the Bessel’s operator D1 is useful for to
determine the probability density ω associated to Rayleigh process in two dimensions,
because the FPE can adopt a form similar to (4).
In fact, the nonstationary stochastic FPE for the Rayleigh distribution is given by
[11] p. 73:
∂
k
k ∂ 2ω
ω̇ =
(6)
βx−
ω +
∂x
2x
2 ∂ x2
being k and β positive parameters, then the corresponding eigenfunctions X (x) are
solutions of:
2
σ2 λ
σ2 d X
2d X
+ 1+ 2 +
+ x−
σ
X=0
(7)
d x2
x dx
x
β
k
. The operator D1 participates when in (7) we make the following change
with σ 2 = 2β
of functions:
x2
(8)
F (x) = x−1 e 4σ2 X
obtaining thus the relation:
D1 (x)F =
1
λ
x2
− 2− 2
4
4σ
σ
σ β
F
(9)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 29–32
31
On the other hand, in (4) it is possible to realize the formal change of variable:
γ=
i
x ,
σ
i=
√
−1
(10)
then it results the equation:
D1 (x)f =
Q
A
x2
+ 2− 2
4
4σ
x
σ
f
,
(11)
with the same structure as (9); therefore Q = 0, that is m = n, and A = 2n + 1 = 1 + βλ ,
then λ = 2nβ , n= 0,1,2,. . . Besides, F is proportional to f given by (2) with the change
(10):
2
2
x
− x2
F ∝ e 4σ Ln
(12)
2σ 2
then (8), (12) and factors of normalization lead us to the eigenfunctions:
2
x
x − x22
2σ L
e
Xn (x) =
n
n!σ 2
2σ 2
(13)
which are solutions of (7) for λ = 2nβ. The result (13) is our principal aim because
it permits to write immediately the two-dimensional probability density associated to
Rayleigh case, to see [11].
32
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 29–32
References
[1] A. E. Glassgold and D. Holliday, Phys. Rev. A139 (1965) 1717.
[2] J. Morales, J. López-Bonilla and A. Palma, J. Math. Phys. 28 (1987) 1032.
[3] J. Morales and A. Flores-Riveros, J. Math. Phys. 30 (1989) 393.
[4] J. López-Bonilla and G. Ovando, Bull. Irish Math. Soc. N.44 (2000) 61.
[5] C. Lanczos, Linear differential operators, D. Van Nostrand Co., London (1961).
[6] J. D. Talman, Special functions, W. A. Benjamin Inc. New York (1968).
[7] H. Hochstadt, The functions of Mathematical Physics, Dover, New York (1971).
[8] M. Abramowitz and I. A. Stegun, Handbook of mathematical functions, John Wiley
and Sons, New York (1972).
[9] J. H. Caltenco, J. López-Bonilla and R. Peña-Rivero, J. Sci. Res. 50 (2002) 125.
[10] Ch. S. Johnson Jr. and L. G. Pedersen, Problems and solutions in Quantum
Chemistry and Physics, Dover, New York (1986).
[11] R. L. Stratonovich, Topics in the theory of random noise. Vol. I, Gordon and Breach,
New York (1963).
[12] A. Weinstein, Ann. Mat. Pura Appl. 49 (1960) 359.
EJTP 3, No. 11 (2006) 33–70
Electronic Journal of Theoretical Physics
Identical synchronization in chaotic
jerk dynamical systems
Vinod Patidar1∗ and K. K. Sud2
1
Department of Physics
Banasthali Vidyapith Deemed University
Banasthali - 304022, Rajasthan, INDIA
2
Department of Physics
College of Science Campus
M. L. S. University, Udaipur – 313002, INDIA
Received 23 December 2005 , Accepted 24 February 2006, Published 25 June 2006
Abstract:
It has been recently investigated that the jerk dynamical systems are the simplest ever systems, which possess
variety of dynamical behaviours including chaotic motion. Interestingly, the jerk dynamical systems also describe various
phenomena in physics and engineering such as electrical circuits, mechanical oscillators, laser physics, solar wind driven
magnetosphere ionosphere (WINDMI) model, damped harmonic oscillator driven by nonlinear memory term, biological
systems etc. In many practical situations chaos is undesirable phenomenon, which may lead to irregular operations in
physical systems. Thus from a practical point of view, one would like to convert chaotic solutions into periodic limit cycle
or fixed point solutions. On the other hand, there has been growing interest to use chaos profitably by synchronizing chaotic
systems due to its potential applications in secure communication. In this paper, we have made a thorough investigation
of synchronization of identical chaotic jerk dynamical systems by implementing three well-known techniques: (i) PecoraCarroll (PC) technique, (ii) Feedback (FB) technique and (iii) Active Passive decomposition (APD). We have given a
detailed review of these techniques followed by the results of our investigations of identical synchronization of chaos in
jerk dynamical systems. The stability of identical synchronization in all the aforesaid methods has also been discussed
through the transversal stability analysis. Our extensive numerical calculation results reveal that in PC and FB techniques
the x-drive configuration is able to produce the stable identical synchronization in all the chaotic jerk dynamical systems
considered by us (except for a few cases), however y-drive and z-drive configurations do not lead to the stable identical
synchronization. For the APD approach, we have suggested a generalized active passive decomposition, which leads to the
stable identical synchronization without being bothered about the specific form of the jerk dynamical system. Several other
active passive decompositions have also been listed with their corresponding conditional Lyapunov exponents to achieve
the stable identical synchronization in various chaotic jerk dynamical systems.
c Electronic Journal of Theoretical Physics. All rights reserved.
Keywords: Chaos, Jerk dynamical systems, Identical synchronization of chaos, synchronized
chaos
PACS (2006): 05.45.+b; 47.52.+j; 05.45.-a
∗
vinod r patidar@yahoo.co.in
34
1.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
Introduction
Various studies of nonlinear dynamical systems in the last four decades have significantly extended the notion of oscillations in these systems. It has been shown that the
post-transient oscillations in dynamical systems can be associated not only with the regular behavior such as periodic or quasiperiodic oscillations, but also with chaotic behavior
[1-5].
Chaos has long-term unpredictable behavior, which is usually couched mathematically
as sensitivity to initial conditions i.e., where the system’s dynamics takes it, is hard to
predict from the starting point. One way to demonstrate this is to run two identical
chaotic systems side by side, starting both at very close, but not exactly equal initial
conditions. The systems soon diverge from each other, but both retain the same attractor
pattern. An interesting question to ask is: Can we force the two chaotic systems to follow
the same path on the attractor? i.e., Can chaotic systems be made synchronized? The
affirmative answer is possible to this question. It has been shown that some of the ideas of
synchronization can also be extended for the description of particular type of behaviour
in coupled systems possessing chaotic dynamics. For example, Fujisaka and Yamada
[6-8] have demonstrated that two identical systems with chaotic individual dynamics
can change their behaviour from uncorrelated chaotic oscillations to identical chaotic
oscillations as the strength of the coupling between the systems is increased.
Synchronization of chaotic systems is important as the noise-like behavior of chaotic
systems suggests us that such behavior might be useful in some type of private communications. One glance at the Fourier spectrum from a chaotic system suggests the
same. There are typically no dominant peaks, no special frequencies i.e., the spectrum
is broadband. To use a chaotic signal in communications we are immediately led to
the requirement that somehow the receiver must have a duplicate of the transmitter’s
chaotic signal or, better yet, synchronize with the transmitter. In fact, synchronization
is a requirement of many types of communication systems, not only chaotic ones. Unfortunately, if we look at how other signals are synchronized, we will get very little insight
as to how to do it with chaos. New methods are therefore required.
The organization of rest of the paper is as follows: In Section 2, we introduce the
jerk dynamical systems, which are emerged as a sub class of dynamical systems with
rich variety of dynamical behaviour including the chaotic motion. We also summarize
in a tabular form the various classes of jerk dynamical systems, which we have used for
studying the identical synchronization of chaos in these systems. In Section 3, we investigate the identical synchronization of chaos in jerk dynamical system using various
algorithms along with their brief descriptions. Particularly in subsections 3.1, 3.2 and
3.3, we investigate the identical synchronization of various jerk dynamical systems using
Pecora-Carroll (PC) technique, feedback (FB) technique and active-passive decomposition (APD) respectively. We also investigate the stability of identical synchronization
in the aforesaid methods using extensive transversal stability analysis (i.e., extensive
conditional Lyapunov exponents (CLE’s) calculation) as well as Lyapunov function con-
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
35
struction criterion. Finally in Section 4, we summarize the major conclusions drawn from
our study of identical synchronization of chaotic jerk dynamical systems.
2.
Chaotic Jerk Dynamical Systems
In this section, we briefly discuss how the jerk dynamical systems came into existence
in the studies of chaotic dynamics. In recent years, there has been a growing interest to
find three dimensional dynamical systems which are functionally as simple as possible but
nevertheless chaotic [10-22]. Using extensive computer search, Sprott [9] found 19 distinct
chaotic models (one of them is conservative and remaining are dissipative, referred as
Model A to S) with three-dimensional vector fields that consist of five terms including two
nonlinearities or of six terms with one quadratic nonlinearity. Subsequently, Hoover [10]
pointed out that the only conservative system (Model A) found by Sprott is an already
known special case of Nose-Hoover thermostat dynamical system, which exhibits time
reversible Hamiltonian chaos [11]. Except this the other models B to S were apparently
unknown. On the other hand, in 1996 Gottlieb [12] reported that Sprott’s Model A
can be recast into an explicit third order form x = J(x, ẋ, ẍ), which he called ‘jerk
function’. Because it involves a third derivative of x, which in a mechanical system is a
rate of change of acceleration and is called jerk [13]. Since it is known that any explicit
ordinary differential equation can be recast in the form of a system of coupled first order
differential equations but the converse does not hold in general. Even if one may reduce
the dynamical system to a jerk form for each of the phase space variables, the resulting
differential equation may look quite different i.e. there may be different possible jerk
forms of a single dynamical system. With this, Gottlieb [12] asked a provocative question
“What is the simplest jerk function that gives chaos?” By following the study of Gottlieb,
Linz [14] reported that the original Rössler model, Lorenz model and the Sprott’s model
R can be reduced to jerk form and further showed that the jerk form of Rössler and Lorenz
models are rather complicated and are not suitable candidates for the Gottlieb’s simplest
jerk function. However, the jerk form of Sprott’s model R demonstrates the existence of
a much simpler form of jerk function that exhibits chaos. By following the Linz’s study
[14], in 1998 Eichhorn et. al. [15] used the method of Görbner bases and showed that the
fifteen of the Sprott’s chaotic flows [9] can be recast into jerk form. They also showed that
these fifteen models , the Rössler toroidal model [16] and Sprott’s minimal chaotic flow
[17] can be arranged into seven classes (referred as JD1 to JD7) of jerky dynamics as a
hierarchy of quadratic jerk equations with increasingly many terms. Table 1 summarizes
the classification simple polynomial chaotic flows suggested by Eichhorn et al. [15]. The
alphabets in the first column are the labels assigned by Sprott [9], except for ‘TR’ and ‘SJ’
which are the abbereviations for Toroidal Rössler and Simplest Jerk systems respectively.
Such a classification provides a simple mean to compare the functional complexity of
different systems and also demonstrate the equivalence of cases not otherwise apparent.
These seven different classes of chaotic jerk dynamical systems, we have used to study
the identical synchronization in section 3.
36
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
In a subsequent study Eichhorn et. al. [18] examined the simple cases of JD1 and JD2
in more detail and identified the regions of parameter space over which they exhibit chaos.
Patidar and Sud [19] have studied in detail, the dynamical behaviour of a special family
of jerk dynamical system x = −Aẍ − ẋ + G(x), where A is a system parameter, G(x)is
nonlinear function containing one nonlinearity, one system parameter and a constant
term.
Table 1: Basic classes of chaotic dissipative jerky dynamics
Model
Basic classes
and parameter values for which there is chaotic behavior
JD1
I
J
L
N
R
SJ
JD2
k1
k1
k1
k1
k1
k1
M
k1 = −1
Q
S
TR
JD3
F
G
H
=
=
=
=
=
=
−1
−β = −2
−1
−γ = −2
−1
−a = −2.017
k1 = β − 1 = −0.5
k1 = −1
k1 = −β = −0.2
k1
k4
k1
k4
k1
k4
= α − 1 = −0.5
1
= 2α
= 1
= α − 1 = −0.6
1
= 4α
= 0.625
= α − 1 = −0.5
1
= 1
= 2α
JD4
O
k1 =
k4 =
− 21
1
4
JD5
D
JD6
k1 = −1
P
k1 = −1
k4 = 1
JD7
K
k1 = α −
k4 = 1
1
2α
...
x = k1 ẍ + k2 x + xẋ + k3
k2 = −2α = − 0.4
k2 = −αβ = − 4
k2 = −α = − 3.9
k2 = −αγ = − 4
k2 = −α = − 0.9
k2 = − 1
...
x = k1 ẍ + k2 ẋ + x2 + k3
k2 = −β = − 1.7
k2 = β − α = − 2.6
k2 = −α = − 4
k2 = − 1
...
x = k1 ẍ + k2 ẋ + k3 x2 + xẋ + k4
1
− 1 = −2.5
k2 = α − α
k2 = α −
1
2α
k2 = α −
1
α
− 1 = −1.85
− 1 = −2.5
...
x = k1 ẍ + k2 ẋ + k3 x2 + xẍ + k4
k2 = 1 − α = − 1.7
k3
k3
k3
k3
k3
k3
=
=
=
=
=
=
−2α2 = −0.08
αβ (1 − α ) = −4
α (2βγ − α ) = −8.19
α (2β − α γ) = −4
−β = −0.4
0
k3 = −α −
2
β2
4
= −2.4225
k3 = − α4 = −2.4025
k3 = −α2 β = −16
k3 = − 14 (α + β )2 ≈ −0.0858
= −0.25
k3 = − α
2
k3 = −α = −0.4
k3 = − α
= −0.25
2
k3 = −1
...
x = k1 ẍ + k2 x2 + k3 ẋ2 + xẍ
k2 = 1
k3 = −α = −3
...
x = k1 ẍ + k2 ẋ + k3 x2 + k4 ẋ2 + xẍ + k5
k2 = 1 − α = − 1.7
k3 = − 12
k5 = 1/2
...
x = k1 ẍ + k2 ẋ + k3 x2 + k4 ẋ2 + k5 xẋ + xẍ + k6
1
k2 = α − α
k3 = −α = −0.3
− 1 ≈ −2.37
− 12 ≈ −3.53
1
≈ 0.83
k5 = 2 − α = 1.7
k6 = 4α
They have identified the regions of parameter space, where different type of long time
dynamical behaviour dominates, using some analytical methods as well as extensive Lyapunov spectra calculation in complete parameter space by considering five different forms
of nonlinear function G(x) comprising of absolute, quadratic, cubic, quartic and quintic
nonlinearities. They observed that only systems having absolute and quadratic nonlinearities in G(x) exibit chaos. As a result they made an important conclusion for these
jerk dynamical systems that a certain amount of nonlinearity is sufficient for exhibiting
chaotic behaviour but more is not necessarily better. Apart from the above studies on
the dynamics of jerk dynamical system, Patidar et. al. [20] and Patidar and Sud [21]
respectively, have attempted the problems of controlling chaos and synchronization of
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
37
chaos in jerk dynamical systems. In the next section, we investigate the identical synchronization of chaos in jerk dynamical system using various algorithms along with their
brief descriptions.
3.
Identical Synchronization In Chaotic Jerk Dynamical Systems
Early work on synchronization of coupled chaotic systems was done by Fujisaka
and Yamada [6-8]. In that work, some sense of how the dynamics might change was
brought out by a study of the Lyapunov exponents of synchronized, coupled systems.
Although Fujisaka and Yamada [6,-8] were the first to exploit local analysis for the study
of synchronized chaos, their papers went relatively unnoticed. Later, a now-famous paper by Afraimovich et. al. [22] exposed many of the concepts necessary for analyzing
synchronous chaos, although it was not until many years later that wide-spread study of
synchronized, chaotic systems took hold.
As already mentioned, chaotic systems are dynamical systems that defy synchronization due to their essential feature of displaying high sensitivity to initial conditions. As
a result, two identical chaotic systems starting at nearly the same initial points in phase
space develop onto trajectories which become uncorrelated in the course of time. However, it has been shown that it is possible to synchronize these kind of systems, to make
them evolving on the same chaotic trajectory [6, 22, 23, 24, 25]. When we deal with coupled identical systems, synchronization appears as the equality of the state variables while
evolving in time. We refer to this type of synchronization as identical synchronization
(IS). Other names in literature for this notion are: complete synchronization or conventional synchronization [26]. In this section we will discuss main properties of this kind of
synchronization. The appearance of this kind of synchronization have been estabilished
by means of several different coupling mechanisms, such as Pecora and Carroll method
[24, 25, 27], the negative feedback [28], the sporadic driving [29], the active-passive decomposition [30, 31], the diffusive coupling and some hybrid method of Guemez and Matias
[32] etc. In this section, we will concentrate our attention to explain the properties and
stability of synchronized motion in three particular coupling schemes namely Pecora and
Carroll (PC) method, feedback technique and active passive decomposition (APD).
3.1 Pecora Carroll (PC) Technique
Pecora and Carroll [24, 25] discovered a way to achieve identical synchronization. They
take a complete chaotic system and choose a subsystem within it. Then they make a
replica of this subsystem. The original system is called drive (master) system and the
duplicate subsystem is called response (slave). The response is just like the drive except
it is missing one or more variables. The missing variables are sent from the drive to the
response, inputting the variable wherever it is needed in the response subsystem. If a
stable response subsystem has been chosen, then the response’s dynamic variables will
38
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
converge to their counterparts in the drive and will remain synchronized with them as
long as the drive continues to supply the missing variables to the response subsystem i.e.
the key idea for identical synchronization is to choose a stable subsystem.
Consider an autonomous n-dimensional dynamical system
Ẋ = F (X),
(1)
here X = (x1 , x2 , x3 , ..., xn )T and F (X) = (f1 (X), f2 (X), ..., fn (X))T .
Divide the system into two parts in an arbitrary way as:
X = (v, w),
v̇ = g(v, w),
ẇ = h(v, w),
(2)
where v = (x1 , x2 , ..., xm ), g = (f1 (X), f2 (X), ..., fm (X)), w = (xm+1 , xm+2 , ..., xn )
and h = (fm+1 (X), fm+2 (X), ..., fn (X)).
Now create a new subsystem w′ identical to w system and, substitute the set of
variables v for the corresponding v ′ in the function h i.e.,
ẇ′ = h(v ′ , w′ )and v ′ = v i.e. the new subsystem will be drived by the original
system
ẇ′ = h(v, w′ ).
(3)
In such a way we have the following compound system:
v̇ = g(v, w),
ẇ = h(v, w),
ẇ′ = h(v, w′ )
(4)
Now we examine the difference Δw = w − w′ , Under the right condition (i.e. if the
chosen subsystem is stable) both the system will synchronize as time grows i.e. Δw → 0
as t → ∞.
In order to study the dynamics of difference Δw, subtract Eq. (3) from Eq. (2), we
get an equation for the dynamics of Δw as:
ẇ − ẇ′ =
d
(Δw) = Dw h(v, w) Δw,
dt
(5)
where Dw his the matrix of derivatives of hwith respect to w i.e. the Jacobian of the w
subsystem vector field with respect to w only.
The technique described above is also known as complete replacement technique because in this technique we completely replace all the v ′ variables of response subsystem
by their counterparts (v) in the original system.
It is clear that in the phase space of the compound system described by Eq. (4),
there exists a manifold w = w′ such that if the initial conditions lie in this manifold, the
consequent evolution of the system will take place in this manifold as well. In other words,
the manifold w = w′ is an invariant manifold and is called synchronization manifold.
In the above analysis, we have said that both systems will show identical chaotic
oscillations if a stable subsystem is chosen. In order to define which subsystem is stable,
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
39
one needs to study the stability of synchronized oscillations. Since in our case the synchronized oscillations are chaotic, these trajectories are always unstable. However in the
analysis of synchronized chaotic motions one has to distinguish between the instability
for perturbation tangent to synchronization manifold and transverse to it. The regime
of identical chaotic oscillations is stable, when the synchronized trajectories are stable
for the perturbations in the transverse direction to the synchronization manifold. This
analysis is called transversal stability analysis. The Eq. (5) is a basic equation for much
of the discussion on synchronized chaotic systems. Three most frequent criteria for the
stability of synchronized chaotic motions are:
(1) The criterion based on eigenvalues of the Jacobian matrix corresponding to the flow
evaluated on synchronization manifold (i.e. Jacobian matrix defined on difference
system) has been introduced by Fujisaka and Yamada [6, 7], which requires that
for the stable synchronization the largest eigenvalue of the aforesaid Jacobian matrix should be negative. For the case of Pecora Carroll technique discussed above,
this criterion states that the largest eigenvalue of the Jacobian matrix Dw hmust be
negative for onset of synchronization. If the Dw h is constant over the attractor i.e.
response is linear then the calculation of eigenvalues (λ1 , λ2 , ..., λn−m ) is straightforward and hence one can easily conclude whether the synchronization is possible or
not. But the complications arise; when the Jacobian is not constant over the attractor i.e. response system is nonlinear in nature (In case of PC method the response
means the subsystem, but for the other coupling mechanisms, where subsystem is
not required, we will consider the difference system instead of response system i.e.
the basic equation (Eq. (5)), which describe the dynamics of difference of drive and
response systems).
(2) He and Vaidya [27] developed a criterion for chaos synchronization based on the
notion of asymptotic stability of dynamical systems, which refers to the condition for
a given chaotic system with drive-response (master-slave) configuration to reach the
same eventual state at a fixed time irrespective of the choice of initial conditions. One
of the practical way to establish the asymptotic stability of the response subsystem
is to find a suitable Lyapunov function L(Δw)(where Δw represents the difference)
that satisfy the following properties:
a) L(Δw) ≥ 0in a certain region containing the chaotic set of synchronized motion.
b) L(0) = 0, In other words, L(•)is zero everywhere on the synchronization
manifold (i.e.w = w′ ).
c) dL
≤ 0 everywhere in a certain region containing the chaotic set of synchronized
dt
motion and dL
= 0 on the synchronization manifold.
dt
If one succeeds in construction of such Lyapunov function for a particular choice
of response subsystem then one can conclude that the subsystem is stable and the
chaotic set of synchronized oscillations is transversally stable therefore the composite
system evolves towards the regime of identical synchronized chaotic oscillations.
The criterion based on construction of Lyapunov function for the vector field
40
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
of perturbation transverse to the synchronization manifold enables one to prove
that all trajectories in the phase space of the coupled systems are attracted by the
synchronization manifold. Despite the fact that it is a very reliable criterion, which
guarantees the onset of chaotic synchronization, it has two serious disadvantages.
First of all, there is no generic method for construction of Lyapunov function for
an arbitrary system. As a result the search for it is time consuming or practically
impossible even though these systems are known to be synchronized. Thus this
method is not general. The second disadvantage is that the estimate for the minimum
coupling strength necessary to synchronize two systems provided by this method is
usually not accurate. It is found that the systems can be synchronized by much
weaker coupling than predicted by the Lyapunov function method. (The second
disadvantage is not applicable for the PC method as there is no coupling parameter
involved in PC technique. However it is applicable for feedback methods, where we
introduce the coupling between the drive and response systems rather than creating
a new subsystem, as we did in PC method. These methods will be clear in the
subsequent sections).
(3) In contrast with the Lyapunov function criterion, the analysis of transversal Lyapunov exponents is quite straight forward and can be easily employed, even for rather
complicated systems. This can be done by calculating the Lyapunov exponents of
Eq. (5) by using methods similar to ones used for the computation of conventional
Lyapunov exponents for any dynamical systems [33-36]. When all the (n − m) Lyapunov exponents are negative, then the compound system will move towards the
synchronization manifold. Note that the full system (1) has a set of n Lyapunov
exponents (at least one will be positive because we are considering the chaotic system), but the Lyapunov exponents for Eq. (5) do not form a subset of the Lyapunov
exponents of full system because they depend upon the drive variables (v) so called
conditional Lyapunov exponents (CLE’s).
Using Pecora and Carroll approach described above, identical synchronization of chaos has
been demonstrated numerically as well as experimentally on several chaotic systems such
as Lorenz system [24, 25, 27, 37-32], Rossler system [24], the hysteretic circuit [24] Chua’s
circuit [43], driven Chua’s circuit [44], DVP oscillator [45-47], phase-locked loops [48-50]
etc. Now, we demonstrate Pecora and Carroll approach for identical synchronization by
taking chaotic jerk dynamical systems as prototypical examples.
First of all we consider the two chaotic jerk dynamical systems studied by Patidar
and Sud [19], which are members of x = −Aẍ − ẋ + G(x)(for G(x) = B (x 2 + C) and
G(x) = B ( |x| + C)), then we will extend the results to the jerky representations of the
Sprott’s simple chaotic flows by considering their classification according to the hierarchy
of quadratic jerk functions with increasingly many more terms (see Table 1 for detailed
classification).
We consider the jerk equation of the form:
x + A ẍ + ẋ = G(x)
(6)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
41
The dynamical system representation of Eq. (6) in terms of ODE’s by using the transformation ẋ = yandẏ = zwill be:
ẋ = y
ẏ = z
ż = − Az − y + G(x)
(7)
There are three possibilities to implement the PC method in this system.
(i) x-drive configuration: If we consider the x as drive variable then the drive and
response systems for jerk dynamical system (Eq. (7)) will be as
ẋ = y
ẏ = z
ż = − Az − y + G(x)
and
ẏ ′ = z ′
ż ′ = −A z ′ − y ′ + G(x)
and the difference system for Δy = y − y ′ and Δz = z − z ′ in matrix form is
d
(Δy)
Δy
0
1
dt
=
d
Δz
−1 − A
(Δz)
dt
(8)
(9)
Eq. (9) is the basic equation for describing the stability of the perturbation transverse
to the synchronization manifold under x-drive configuration for jerk dynamical system
(Eq. (7)). Since in this equation the coefficient matrix (Jacobian matrix) is a constant
matrix and it is very√easy to calculate the eigenvalues of this matrix. The eigenvalues for
2
this matrix are: −A ± 2 A −4 , for A = 0.6(chaotic case) the eigenvalues are −0.3 ± i 0.95,
which give the following solution for Eq. (9)
Δy = Δz = e−0.3t (K1 cos (0.95t) + K2 sin (0.95t))
(10)
here K1 and K2 are constants of integration. It is clear that as t → ∞, Δy = Δz = 0,
hence drive and response systems synchronize. If we numerically calculate the Lyapunov
exponents of the subsystem (subjected to x-drive condition) i.e. conditional Lyapunov
exponents (CLE’s), these come out to -2.9986E-1, -3.0013E-1 (for both the cases (i) G(x)
= B(x 2 + C) with B = 0.58 and C = −1 (ii) G(x) = B( |x| + C) with B = 1.0 and
C = −2.0). As all the CLE’s are negative and hence synchronization is possible.
In the above paragraph, we have analyzed the possibility of identical synchronization
in chaotic jerk dynamical system (Eq. (7)) for x-drive configuration using eigenvalues of
the Jacobian matrix and CLE’s. Now we analyze the possibility of synchronization using
the Lyapunov function construction method.
If we consider the Lyapunov function
1
1+A
2
2
2
2
L(Δy, Δz) =
(11)
(Δy) + (Δz) + (Δy + Δz) + A (Δy) ,
2
A
here Δy = y − y ′ and Δz = z − z ′ . Lyapunov function L (Δy, Δz)is positive definite
and L(0, 0) = 0 i.e., it is zero everywhere on the synchronization manifold (y = y ′ and
z = z ′ ). Now the time rate of change of Lyapunov function is given by
dL
dt
(Δy, Δz) =
1+A
A
(Δy) dtd (Δy) + (Δz) dtd (Δz)
+ (Δy + Δz) dtd (Δy) +
d
(Δz)
dt
+ A (Δy) dtd (Δy)
42
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
= − 2A (Δz)2 + (Δy)2
(12)
≤ 0and also dL
= 0 on the synchronization manifold, hence the
for A = 0.6 > 0, dL
dt
dt
synchronization is possible i.e. Δy = Δz = 0 as t → ∞.
From the analysis given above, we found that all the three criteria guarantee the
identical synchronization of chaotic jerk dynamical systems described by Eq. (7) using
the x-drive configuration.
Fig. 1 Identical synchronization of chaos in jerk dynamical system having quadratic nonlinearity
using Pecora-Carroll technique (i.e., Eq. (8) with G(x) = B(x2 + C)) for A = 0.6, B = 0.58 and
C = 1.0: (a) time history of y-variables of drive and response systems (b) the difference between
y-variables of drive and response as a function of time (c) time history of z-variables of drive
and response systems (d) the difference between z-variables of drive and response as a function
of time.
We have also numerically solved the drive-response system of jerk dynamical system
for x-drive configuration i.e., Eq. (8) by using fourth-order Runge-kutta integrator with
the step size 0.01. The results of the numerical calculations are shown in Figures 1 and 2
for G(x) = B(x2 + C)with A=0.6, B=0.58 and C = −1. Particularly in Figures 1(a) and
1(b), we have shown the y−variables of both drive and response systems and difference
between y−variables of both drive and response systems respectively. It is clear that ast
increases the y-variables of drive and response systems synchronize i.e., difference becomes
zero. Similar behaviour is discussed in Figures 1(c) and 1(d) for z-variables. There are
more ways to visualize identical synchronization between the drive and response systems.
Some of them we have shown in Figure 2. Particularly in Figure 2(a), we have plotted the
y-variable of drive versus y-variable of response while in Figure 2(b) z-variable of drive
versus z-variable of response (after some transient time die out). Both plots are straight
lines inclined with an angle of 45o and passing through the origin (0,0) which also show
the synchronized behaviour of drive and response systems.
In Figure 2(c) we have shown the trajectories of the drive and response systems in the
xy-plane. We see that the drive and response start with different initial conditions but
after some time both converge to the same trajectory. A better view has been shown in
Figure 2(d) in which we have plotted the trajectory in the ΔyΔz − plane (i.e. difference
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
43
Fig. 2 Identical synchronization of chaos in jerk dynamical system having quadratic nonlinearity
using Pecora-Carroll technique (i.e., Eq. (8) with G(x) = B(x2 + C)) for A = 0.6, B = 0.58 and
C = −1.0: (a) y-variable of drive versus y-variable of response system (b) z-variable of drive
versus z-variable of response system (c) the trajectory of drive and response systems in yz-plane
(d) the trajectory of the difference of y-variables and z-variables of drive and response systems
in difference plane.
Fig. 3 Identical synchronization of chaos in jerk dynamical system having absolute nonlinearity
using Pecora-Carroll technique (i.e., Eq. (8) with G(x) = B(|x| +C)) for A = 0.6, B = 1.0
and C = −2.0: (a) time history of y-variables of drive and response systems (b) the difference
between y-variables of drive and response as a function of time (c) time history of z-variables
of drive and response systems (d) the difference between z-variables of drive and response as a
function of time.
plane), it is a spiral ending at a fixed point (0,0). Results of the similar numerical
calculations for Eq. (8) with G(x) = B(|x | + C) for A = 0.6, B = 1.0 and C = -2.0 are
shown in Figures 3 and 4.
(ii) y-drive configuration: If we consider y as a drive variable then the drive and
response systems for jerk dynamical system (Eq. (7)) will be as
44
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
Fig. 4 Identical synchronization of chaos in jerk dynamical system having absolute nonlinearity
using Pecora-Carroll technique (i.e., Eq. (8) with G(x) = B(|x| +C)) for A = 0.6, B = 1.0 and
C = −2.0: (a) y-variable of drive versus y-variable of response system (b) z-variable of drive
versus z-variable of response system (c) the trajectory of drive and response systems in yz-plane
(d) the trajectory of the difference of y-variables and z-variables of drive and response systems
in difference plane.
ẋ = y
ẏ = z
ż = − Az − y + G(x)
and
ẋ′ = y
ż ′ = −A z ′ − y + G(x′ )
(13)
and the difference system for Δx = x − x′ and Δz = z − z ′ will be as:
d
(Δx)
dt
d
(Δz)
dt
= y −y = 0
= −A(Δz) + G(x) − G(x′ )
(14)
Eq. (14) is the basic equation for describing the stability of the perturbation transverse to
the synchronization manifold under y-drive configuration for jerk dynamical system (Eq.
(7)). Since in this equation the coefficient matrix (Jacobian matrix) is not a constant
matrix as in case of x-drive configuration. So it is not easy to calculate the eigenvalues of
this matrix because these will depend on the time evolution of x and z variables of both
drive and response systems. Instead of finding the eigenvalues, we numerically calculate
the conditional Lyapunov exponents and which are 2.9899E-4 (≈ 0) & -6.0029E-1 (for
G(x) = B(x 2 + C);A = 0.6, B = 0.58 and C = −1.0) and 5.4307E-4 (≈ 0) & -6.0054E-1
(for G(x) = B(|x|+C);A = 0.6, B = 1.0 and C = −2.0). Since one of them (in both
cases) is approximately zero or positive definite and hence identical synchronization is
not possible.
We have also numerically solved the drive-response system of jerk dynamical system
for y-drive configuration i.e., Eq. (13) by using fourth-order Runge-kutta integrator with
the step size 0.01. The results of the numerical calculations are shown in Figure 5 (for
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
Fig.5 Identical synchronization of chaos in
Fig.6 Identical synchronization of chaos in jerk
jerk dynamical system having quadratic nonlinearity using Pecora-Carroll technique (Eq.
(13) with G(x) = B(x2 + C)) for A = 0.6,
B = 0.58 and C = −1.0: (a) the difference
between x-variables of drive and response as a
function of time (b) the difference between zvariables of drive and response as a function of
time (c) ) x-variable of drive versus x-variable
of response system (d) z-variable of drive versus z-variable of response system.
dynamical system having absolute nonlinearity
using Pecora-Carroll technique (Eq. (13) with
G(x) = B(|x| +C)) for A = 0.6, B = 1.0
and C = −2.0: (a) the difference between xvariables of drive and response as a function of
time (b) the difference between z-variables of
drive and response as a function of time (c) )
x-variable of drive versus x-variable of response
system (d) z-variable of drive versus z-variable
of response system.
45
G(x) = B(x 2 + C); A = 0.6, B = 0.58 and C = −1.0) and in Figure 6 (for G(x) =
B(|x | + C);A = 0.6, B = 1.0 and C = −2.0). Particularly in Frames (a) of both figures,
we have shown the difference between x-variable of the drive and response systems while
in Frames (b) the difference between the z-variables of the drive and response systems.
We observe that the difference between the x-variables of drive and response systems is
constant and equal to the initial difference. It is so because one of the CLE’s is equal to
zero for y-drive configuration, which suggests that neither the initially nearby trajectories
diverge nor converge. However the difference between the z-variables of the drive and
response systems is large and oscillatory and hence the drive and response systems do
not synchronize for y-drive configuration. In Frames (c) of both the figures, we have
plotted the x-variables of drive versus the x-variable of response system. It is a straight
line passing through the point (0, 16) not through origin, which suggests that x-variables
of drive and response systems do not synchronize (x = x′ ) but have a linear relationship
46
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
(x′ = x + 16). However the plot of z versus z ′ (see Frames (d) of both the Figures)
is a complicated geometrical structure, which shows the non-synchronized behaviour of
z−variables of drive and response systems and also suggests that zand z ′ have a nonlinear
relationship as evident from a complicated attractor type structure in Frames (d).
(iii) z-drive configuration: If we consider z as the drive variable then the drive and
response systems for jerk dynamical system (Eq. (7)) will be as:
ẋ = y
ẏ = z
ż = − Az − y + G(x)
and
ẋ′ = y ′
ẏ ′ = z
(15)
and the difference system for Δx = x − x′ and Δy = y − y ′ in matrix form is
d
(Δx)
0
1
Δx
dt
=
(16)
d
0
0
Δy
(Δy)
dt
Eq. (16) is the basic equation for describing the stability of the perturbation transverse
to the synchronization manifold under z-drive configuration for jerk dynamical system
(Eq. (7)). Since in this equation the coefficient matrix (Jacobian matrix) is a constant
matrix and it is very easy to calculate the eigenvalues of this matrix. The eigenvalues
for this matrix are: 0, 0 for any value ofA and hence the trajectories for x(t)&x′ (t)and
y(t) & y ′ (t)respectively, will remain apart by a constant distances |x(0) − x′ (0)|and |z(0) −
z ′ (0)| , in this case also synchronization is not possible. If we numerically calculate the
conditional Lyapunov exponents, these turn out to be (0.0, 0.0), which are consistent with
the above results. From the analysis of identical synchronization using PC technique for
the jerk dynamical system (Eq. (7)), we can summarize the results in the Table 2.
Table 2: Conditional Lyapunov exponents (CLE’s) for different combination of drive
and response systems for jerk dynamical system having quadratic and absolute nonlinearities.
Drive Variable
x
y
z
x
y
z
...
x + Aẍ + ẋ = B(x2 + C ); A = 0.6; B = 0.58; C = −1
Response Subsystem
Conditional Lyapunov Exponents
Whether IS is possible or not?
(y, z)
-2.9986E-1, -3.0013E-1
Yes
(x, z)
2.9899E-4, -6.0029E-1
No
(x, y)
0.0000E-0, 0.0000E-0
No
...
x + Aẍ + ẋ = B(|x| + C); A = 0.6; B = 1; C = −2
(y, z)
-2.9986E-1, -3.0013E-1
Yes
(x, z)
5.4307E-4, -6.0054E-1
No
(x, y)
0.0000E-0, 0.0000E-0
No
In the above analysis, we have discussed in detail the possibilities of identical synchronization of chaos in jerk dynamical systems having quadratic and absolute nonlinearities
using the Pecora-Carroll approach. Now we extend our calculations for the identical
synchronization of chaos using PC approach to the jerk dynamical representations of the
Sprott’s simple chaotic flows [9] by considering their classification [15] according to the
hierarchy of quadratic jerk functions with increasing many more terms. In Table 3, we
have summarized all these results. We observe from the Table 3 that the x-drive configuration for all the jerk dynamical systems guarantees the identical synchronization except
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
47
for the models L, R and D, where no subsystem is stable. However the y-drive and z-drive
configurations are unstable for all the jerk dynamical systems.
48
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
3.2 Feedback (FB) Technique
In Section 3.1, we observed that Pecora and Carroll method of chaos synchronization
works fairly good for the jerk dynamical systems. However it requires dividing the original
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
49
system into two stable subsystems. In this section, we address the same problem of
synchronization but in different way by considering the following question: Can one
make a chaotic trajectory of one system to synchronize with a chaotic trajectory of other
identical chaotic system, starting with different initial conditions without dividing the
original system into two stable subsystems? It is possible by means of feedback method
or one-way coupling method [45, 47, 51-53]. By one-way coupling we mean that the
behaviour of second chaotic system (response) is dependent on the behaviour of the
first identical system (drive) but the first one (drive) is not influenced by the behaviour
of second (response) system. The one way coupling method is also called as feedback
method as in this we choose a drive variable from the drive system and feedback control
is applied to the response system. The feedback control is proportional to the difference of
the chosen variable from drive and its counterpart from response system. Under suitable
conditions, as time elapses, the amount of feedback decreases and soon both the drive and
response systems achieve complete synchronization by following the same trajectory and
afterward the feedback amount becomes zero and the identical synchronization persists.
The similar feedback method has been used by Singer et. al. [54], Pyragas [55], Chen
and Dong [56], Pyragas [57], Kapitaniak and Chua [58], Kapitaniak et. al. [59] for
the controlling of chaos, where the drive system exhibits the periodic motion of desired
periodicity while the response system is chaotic and the aim of the feedback control is to
bring the response system to the periodicity of drive system. In the following, we discuss
the formulation and stability of the feedback method for identical synchronization of
chaos.
Consider an n-dimensional dynamical system, which is chaotic, as
Ẋ = F (X),
(17)
where X = (x1 , x2 , x3 , ..., xn )T and F (X) = (f1 (X), f2 (X), ..., fn (X))T . Now choose
a dynamical variable as drive variable from it e.g. xi (1 ≤ i ≤ n). Consider another
chaotic system identical to Eq. (17) but starting from different initial conditions (i.e.
with different variables), as
Ẋ ′ = F (X ′ ),
(18)
where X ′ = (x1 , x2 , ..., xn )T and F (X ′ ) = (f1 (X ′ ), f2 (X ′ ), ..., fn (X ′ ))T . Now the feedback control, which is proportional to the difference of the drive variable xi and its
counterpart xi in the response, is applied to response system. Hence the response system
looks as
′
′
′
′
ẋ1 = f1 (x1 , x2 , ...., xn )
.
′
′
′
′
′
ẋi = fi (x1 , x2 , ...., xn ) − c(xi − xi )
,
(19)
.
.
′
′
′
′
ẋn = fn (x1 , x2 , ...., xn )
where c is a constant and termed as feedback constant or coupling strength. The pair
of drive and response dynamical systems (i.e. Eqs. (17) and (19)) synchronize if the
50
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
dynamical system describing the evolution of the difference,
⎡ ⎤ ⎡ ′
⎤ ⎡
⎤
′
′
′
ẋ1 − ẋ1
f1 (x1 , x2 , ..., xn ) − f1 (x1 , x2 , ..., xn )
ė1
⎢. ⎥ ⎢.
⎥ ⎢.
⎥
⎢ ⎥ ⎢ ′
⎥ ⎢
⎥
′
′
′
′
⎢ ėi ⎥ ⎢ ẋi − ẋi ⎥ ⎢ fi (x1 , x2 , ..., xn ) − fi (x1 , x2 , ..., xn ) − c(xi − xi ) ⎥
=
=
⎢. ⎥ ⎢
⎥ ⎢
⎥,
⎢ ⎥ ⎢.
⎥ ⎢.
⎥
⎣. ⎦ ⎣.
⎦ ⎣.
⎦
′
′
′
′
ėn
ẋn − ẋn
fn (x1 , x2 , ..., xn ) − fn (x1 , x2 , ..., xn )
(20)
possesses a stable fixed point at the origin E = 0,where E = (e1 , e2 , ..., en )T and
′
ei = xi − xi for i = 1 to n. The dynamics of the error system can be understood by
studying the following linearized system for small E:
Ė = DF · E,
(21)
where DF is the Jacobian (matrix of derivatives ) of response system with respect to
Ei.e.
⎤
⎡ ∂f1
∂f1
1
. . ∂e
. ∂f
∂e1
∂ei
n
⎢ . . .
. . . ⎥
⎥
⎢ ∂f
∂f
∂fi ⎥
i
i
⎢
.
−
c
.
.
⎢
∂e
∂e
∂e
n ⎥,
i
DF = ⎢ 1
(22)
⎥
.
.
.
.
.
.
⎥
⎢
⎣ . . .
. . . ⎦
∂fn
∂fn
n
. ∂ei
. . ∂f
∂e1
∂en
All the derivatives in the Jacobian matrix are evaluated at E = 0i.e., on the synchronization manifold. It is clear that the coupling strength c also appears in equation of error
system and hence it also affects the stability of synchronization. Now the question arises:
for which selection of drive variable and coupling strength the identical synchronization
will be stable. To study the stability of identical synchronization in this case, one can
use the three basic criteria introduced in the Section 3.1: (i) If the largest eigenvalue
of the Jacobian matrix DF appeared in Eq. (21) is negative then drive and response
systems will possess identical synchronization [6-8]. (ii) If a suitable Lyapunov function
[27] L(E)exists, which satisfies: (a)L(E) ≥ 0, (b) L(0) = 0, (c) dL(E)/dt ≤ 0 and (d)
dL(0)/ = 0. (iii) If all the Lyapunov exponent of the system described by Eq. (21) (condt
ditional Lyapunov exponents) are negative, then the identical synchronization between
drive and response systems will be stable [24, 25].
Now we use the feedback technique described above to demonstrate the identical
synchronization of chaos in jerk dynamical systems. First of all we consider the two jerk
dynamical systems, which have been studied in detail by Patidar and Sud [19], then we
will extend the results to the jerky representations of the Sprott’s simple chaotic flows
by considering their classification (as suggested by Eichhorn et al. [15]) according to the
hierarchy of quadratic jerk functions with increasing many more terms [see Table 1 for
detailed classification]
First, we consider the following family of jerk dynamical system:
x = −A ẍ − ẋ + G(x),
(23)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
51
where A is a system parameter and G(x) is a nonlinear function. The dynamical system
representation of Eq. (23) in terms of first order ODE’s by using the transformations
ẋ = yand ẏ = zwill be as:
ẋ = y
ẏ = z
.
(24)
ż = −Az − y + G(x)
Now choose a dynamical variable x as drive variable and applied the feedback control
as explained in the previous subsection. The response system after inclusion of feedback
term will look like as:
ẋ′ = y ′ − c(x′ − x)
ẏ ′ = z ′
,
(25)
ż ′ = −Az ′ − y ′ + G(x′ )
where c is the coupling strength between the drive and response system. The dynamical
system for the evolution of difference between drive and response systems described by
Eqs. (24) and (25) will be as:
Ė = DF · E,
(26)
where E = (x′ − x, y ′ − y, z ′ − z)T and
DF =
"
−c
0
∂G/∂x
1
0
0
1
−1 −A
#
.
(27)
The derivative in the Jacobian matrix DF is evaluated atE = 0. It is clear that for
a nonlinear functionG(x), the Jacobian matrix given above is not a constant matrix
and it is not easy to calculate its eigenvalues. For analyzing the stability of identical
synchronization in this case, we have performed the numerical calculation of the Lyapunov
exponents of Eq. (26) (i.e. conditional Lyapunov exponents).
First of all we have numerically integrated the coupled systems described by Eqs. (24)
and (25) with A = 0.6, G(x) = 0.58 (x 2 − 1) and for different values of coupling strength
(c). In Figure 7, We have shown the solutions for two specific values of coupling strength
c = 0and c = 0.8. Particularly in Frames (a), (b) and (c) respectively, we have shown the
solution of coupled systems in xx′ -plane, yy ′ -plane and zz ′ -plane for c = 0 i.e when both
the systems are evolving independently from different initial conditions. It is clear that
these figures correspond to an unsynchronized motion. While in Frames (d), (e) and (f)
respectively, we have shown the solution of coupled systems in xx′ -plane, yy ′ -plane and
zz ′ -plane for c = 0.8. In all these frames ((d), (e) and (f)), we observe a line inclined
at an angle of 45o and passing through the origin, which suggests the equality of all
three dynamical variable of drive with its counterparts in response systems i.e. identical
synchronization. For both cases, we have also calculated the Lyapunov exponents of
the response system. For c = 0, the drive and response systems are chaotic and evolve
independently and hence both have the same set of Lyapunov exponents i.e.(0.6832E01, 0.0, -0.6683). For c = 0.8, the response system is dependent on the behaviour of
drive system however the drive system is not influenced by the behaviour of response
system. Hence the drive system possesses the same Lyapunov exponents as for c = 0,
52
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
Fig. 5 Identical synchronization of chaos in jerk dynamical system using feedback technique
(Eqs. (24) and (25)) for A = 0.6and G(x)=0.58(x 2 − 1): Frames (a), (b) and (c) respectively,
show the unsynchronized motion of x, y and z variables of drive and response systems, when
both systems are uncoupled i.e. c = 0.0 while Frames (d), (e) and (f) respectively, show the
synchronized motion of x, y and z variables of drive and response systems, when both systems
are coupled through coupling strength c = 0.8.
while the values of Lyapunov exponents of response system i.e. conditional Lyapunov
exponents (CLE’s) are (-0.4653E-1, -0.2705E-0, -0.1085E+1). From the values of CLE’s
it is clear that all the CLE’s are negative or the largest CLE is negative and hence identical
synchronization is stable for c = 0.8. For getting the complete information for the value
of coupling strength, at which the identical synchronization is stable or unstable, we have
calculated the conditional Lyapunov exponents for a certain range of coupling strength
(c) in step of 0.01.
In Figure 8, we have shown the plot of largest CLE as a function of coupling strength.
We observe from this figure that when c is greater than 0.65 then the largest CLE is
negative and hence identical synchronization is stable while for c less than 0.65, the
largest CLE is positive and identical synchronization is unstable.
We have also numerically integrated the coupled systems described by Eqs. (24) and
(25) with A = 0.6, G(x) = ( |x| − 2) and for different values of coupling strength (c). In
Figure 9, we have shown the solutions for two specific values of coupling strength c = 0
(unsynchronized) and c = 0.7(synchronized). For both cases, we have also calculated the
Lyapunov exponents of the response system. For c = 0, when both the drive and response
systems are chaotic and evolve independently hence have the same set of Lyapunov exponents i.e. (+0.3148E-01, 0.0, -0.6314E-0) while for c = 0.7 the response system is
dependent on the behaviour of drive system however the drive system is not influenced
by the behaviour of response system hence the drive system possess the same Lyapunov
exponents as for c = 0, while the values of Lyapunov exponents of response system i.e.
conditional Lyapunov exponents (CLE’s) are (-0.2269E-1, -0.3072E-0, -0.9716E-0). All
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
53
Fig. 6 Largest conditional Lyapunov exponent (CLE of Eq. (26)) as a function of coupling
strength (c) for A = 0.6and G(x)=0.58(x 2 − 1).
the CLE’s are negative or the largest CLE is negative and hence identical synchronization is stable for c = 0.7. For getting the complete information for the value of coupling
strength at which identical synchronization is stable or unstable, we have calculated the
conditional Lyapunov exponents for a certain range of coupling strength (c) in step of
0.01.
In Figure 10, we have shown the plot of largest CLE as a function of coupling strength.
We observe from this figure that when c is greater than 0.55 then the largest CLE is
negative and hence identical synchronization is stable while for c less than 0.55 largest
CLE is positive and identical synchronization is unstable. Finally, we extend the feedback
technique for identical synchronization of chaos in jerk dynamical representations of the
Sprott’s simple chaotic flows [9] by considering their classification [15] according to the
hierarchy of quadratic jerk functions with increasing many more terms.
In Figures 11 to 14, we have shown the largest CLE as a function of coupling strength
for the jerk dynamical representations of Sprott’s simple chaotic flows, when x-variable
is used as a drive for calculating the feedback control term. In the respective frames we
have also depicted the threshold value of coupling strength beyond which the largest CLE
is negative i.e. identical synchronization is possible.
3.3 Active Passive Decomposition (APD)
In Sections 3.1 and 3.2, we have discussed two different techniques for synchronizing two
identical chaotic systems as well as their application to various jerk dynamical systems.
In Section 3.1, we observed that the drive and response technique introduced by Pecora
54
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
Fig. 7 Identical synchronization of chaos in jerk dynamical system using feedback technique
(Eqs. (24) and (25)) for A = 0.6and G(x)=(|x| -2): Frames (a), (b) and (c) respectively,
show the unsynchronized motion of x, y and z variables of drive and response systems, when
both systems are uncoupled i.e. c = 0.0 while Frames (d), (e) and (f) respectively, show the
synchronized motion of x, y and z variables of drive and response systems, when both systems
are coupled through coupling strength c = 0.7.
Fig. 8 Largest conditional Lyapunov exponent (CLE of Eq. (26)) as a function of coupling
strength (c) for A = 0.6and G(x)=(|x| -2).
and Carroll [24, 25] works fairly for synchronizing various identical jerk dynamical systems. The x-drive configuration for all jerk dynamical systems guarantees the identical
synchronization except for the models L, R and D, where no subsystem is found to be
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
55
Fig. 9 Largest conditional Lyapunov exponent (CLE of Eq. (26)) as a function of coupling
strength (c) for JD1.
Fig. 10 Largest conditional Lyapunov exponent (CLE of Eq. (26)) as a function of coupling
strength (c) for JD2.
stable. However the y-drive and z-drive configurations for all the jerk dynamical systems
lead to unstable synchronization.
One important point we note that in case of Pecora and Carroll technique of identical
synchronization only a finite number of possible decompositions exist, which are bounded
56
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
Fig. 11 Largest conditional Lyapunov exponent (CLE of Eq. (26)) as a function of coupling
strength (c) for JD3.
by the number of different subsystems. For a n-dimensional dynamical system, number of
possible decomposition in terms of drive and response subsystem is n(n-1)/2. In general
only a few of the possible response subsystems possess negative conditional Lyapunov
exponents and may be used to implement synchronizing systems. Similarly, from Section
3.2, we observed that the feedback technique for synchronization of two identical chaotic
systems also works for various jerk dynamical systems. However in this case also the
identical synchronization of chaos in jerk systems is possible only when we use the xvariable as a drive i.e. the only x-drive configuration is stable. As we know that for
the numerical simulations, it is very easy to divide the original dynamical system into
drive and response subsystems (for PC technique) and to choose a drive variable (for
feedback technique). However in the real physical system there are certain limitations.
In addition, not all the possible combinations of drive and response systems are stable.
So it may create problems while implementing the above discussed techniques for the
identical synchronization everywhere.
To avoid these difficulties Kocarev, Parlitz and their group [30, 31, 60] proposed a
general drive response scheme named as Active Passive Decomposition (APD). The basic
idea of the active passive synchronization approach consists in a decomposition of a given
chaotic system into an active and a passive part where different copies of the passive part
synchronize when driven by the same active component. In the following, we explain the
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
57
Fig. 12 Largest conditional Lyapunov exponent (CLE of Eq. (26)) as a function of coupling
strength (c) for JD4, JD5, JD6, JD7.
basic concept and terminology of the active passive decomposition.
Consider an autonomous n-dimensional dynamical system, which is chaotic as
Ż = F (Z),
(28)
where Z = (z1 , z2 , z3 , ..., zn )T and F (Z) = (f1 (Z), f2 (Z), ..., fn (Z))T .
Now rewrite this autonomous system as a non-autonomous system that possesses
certain synchronization properties as follows:
Ẋ = G(X, S),
(29)
where X = (x1 , x2 , ..., xn )T is the new state vector corresponding to Z and S is some
vector valued function of time given by
S = H(X),
(30)
Ṡ = H(X, S),
(31)
The pair of functions G and H constitutes a decomposition of the original vector field
F . The crucial point of this decomposition is that suitable choices of the function H for
which any system
(32)
Ẋ ′ = G(X ′ , S),
58
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
that is given by same non-autonomous vector field G, the same driving S, but with
′
′
′
different variables X ′ = (x1 , x2 , ..., xn )T , synchronizes with the original system (29), i.e.
X − X ′ → 0 for t → ∞. More precisely, synchronization of the pair of identical systems
(29) and (32) occurs if the dynamical system describing the evolution of the difference
E = X ′ − X,
Ė = Ẋ ′ − Ẋ = G(X ′ , S) − G(X, S),
(33)
= G(X + E, S) − G(X, S),
(34)
Ė = DF (X, S) · E,
(35)
possesses a stable fixed point at the origin E = 0. Here E = (e1 , e2 , ..., en )T and
′
ei = xi − xi for i = 1 to n. The dynamics of the error system can be understood by
studying the following linearized system for small E:
In this decomposition of the vector field F into G and H, the system described by vector
field G (Eq. (29)) is a passive system whereas the component described by H is an
active component and hence this technique is called active passive decomposition (APD).
We note that in APD approach the freedom to choose H that defines the drive signal,
leads to a large flexibility in applications. Thus APD is different from the Pecora-Carroll
synchronization approach, where only a finite number of possible coupling exists that is
given by the number of stable subsystems of the dynamical system. Now the question
arises that how to decompose the original vector field F into the active H and passive
G components, so that the copies of the passive part possess identical synchronization
i.e. X = X ′ . To study the stability of identical synchronization in APD approach, one
can again make use of the three basic criteria introduced in the Section 3.1: (i) If the
largest eigenvalue of the Jacobian matrix DF (X, S)appeared in Eq. (35) is negative
then the copies of the passive system driven by same active component will possess
identical synchronization [6-8]. (ii) If a suitable Lyapunov function [27] L(E)exists, which
satisfies: (a)L(E) ≥ 0, (b) L(0) = 0, (c) dL(E)/dt ≤ 0 and (d) dL(0)/dt = 0. (iii) If
all the Lyapunov exponent of the system described by Eq. (35)(conditional Lyapunov
exponents) are negative, then the copies of the passive system driven by same active
component will possess identical synchronization [24, 25].
Instead of decomposing a given chaotic system one may also synthesize it starting from
a stable linear system Ẋ = A · X given by some constant matrix A where an appropriate
nonlinear function S = H(X)is added such that the complete system
Ẋ = A · X + S,
(36)
is chaotic [61]. In this case S = H(X) is treated as active component and Ẋ = A · Xis as
passive. Now we consider the copy of the passive part with different variables but driven
by same active component i.e.
Ẋ ′ = A · X ′ + S,
(37)
In this case the error dynamics is given by
Ė = Ẋ ′ − Ẋ = A · (X ′ − X) = A · E,
(38)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
59
Since Ais a constant matrix hence if the eigenvalues of matrix A are all negative then the
dynamical system describing the evolution of error (Eq. (38)) will possess a stable fixed
point at the origin therefore two chaotic systems described by Eqs. (36) and (37) possess
identical synchronization independent of initial conditions and active signalS.
The APD approach described above has been applied to demonstrate identical synchronization of chaos in Lorenz, Rossler and Chua circuit [30, 31]. In this study they
have also suggested a possible application of APD synchronization approach to communication and demonstrated it through experimental realization in Chua circuit as well as
numerical simulations in Lorenz and Rossler systems. They also showed that how this
approach can be used to construct high dimensional synchronizing systems in a systematic way using low dimensional systems as building blocks. Later Parlitz et. al. [60]
showed that how synchronization using APD can be used for modeling and time series
analysis. Later on Jinlan et. al. [62] used APD scheme to realize the spatiotemporal
chaotic synchronization. Now, we use APD approach described above to demonstrate the
identical synchronization of chaos in jerk dynamical systems.
We consider a general form of jerk dynamical system as:
x = J(x, ẋ, ẍ),
(39)
The dynamical system representation of Eq. (39) in terms of first order ODE’s by using
the transformations ẋ = yand ẏ = zwill be as:
ẋ = y
ẏ = z
,
ż = J(x, y, z)
(40)
where J(x, y, z)is a jerk function. Now we start with the simplest way to implement the
APD scheme, which is described in Eqs. (36)-(38). We rewrite Eq. (40) as:
ẋ = y
ẏ = z
,
ż = ax + by + cz + J(x, y, z) − ax − by − cz
(41)
T
which"is in form of
# Ẋ = A · X + S, whereX = (x, y, z) , S = J(x, y, z) − ax − by − cz,
0 1 0
A = 0 0 1 and a, band care arbitrary real constants which ensure the desired
a b c
identical synchronization. In the above we add and subtract the same term ax + by + cz
in the evolution of z-variable of Eq. (40). Now by assumingS = J(x, y, z)−ax−by−cz as
an active element for the APD scheme, we consider the copy of the functionally identical
jerk system as Eq. (40) but with different variables and same active element (drive) S
as:
ẋ′ = y ′
ẏ ′ = z ′
,
(42)
ż ′ = ax′ by ′ + cz ′ + J(x, y, z) − ax − by − cz
The error between the variables of Eqs. (41) and (42) is governed by the error equation
as:
(43)
Ė ′ = A · E,
60
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
where E = (x′ − x, y ′ − y, z ′ − z)T . Since Ais a constant matrix hence if the eigenvalues
of matrix A are all negative then the dynamical system describing the evolution of error
(Eq. (43)) will possess a stable fixed point at the origin therefore two chaotic systems
described by Eqs. (41) and (42) possess identical synchronization independent of initial
conditions. The characteristic equation of A is:
λ 3 − cλ 2 − bλ − a = 0.
(44)
The eigenvalues (roots of Eq. (44)) λ 1 , λ 2 and λ 3 satisfy the followings:
λ1 + λ2 + λ3 = c
λ 1 λ 2 + λ 2 λ 3 + λ 3 λ 1 = −b .
λ 1λ 2λ 3 = a
(45)
To ensure the identical synchronization, all the eigenvalues should be negative. Since
a, b and c are arbitrary real constant so it is always possible to choose these constants
in such a way that all the roots of Eq. (44) be negative. In addition, in the above
analysis J(x, y, z)is any jerk function i.e. the above analysis is independent of the jerk
function and hence can be applied to all the jerk dynamical systems. It has been shown
that many three dimensional dynamical systems can be transformed into jerk dynamical
system [15], hence the above analysis can be applied to any three dimensional system
that can be transformed into jerky dynamics.
Now we consider a special family of jerk dynamical systemx = −Aẍ − ẋ + G(x),
where G(x)is a nonlinear function, which has been studied in detail by Patidar and Sud
[19]. If we implement the above formalism (Eqs.(40) to (45)) to these systems then the
active and passive systems will be as follows:
and
ẋ = y
ẏ = z
ż = ax + by + cz − Az − y + G(x) − ax − by − cz
(46)
ẋ′ = y ′
ẏ ′ = z ′
.
ż ′ = ax′ + by ′ + cz ′ − Az − y + G(x) − ax − by − cz
(47)
In Figure
$ 15, we have shown the results of numerical calculation of Euclidean error
(i.e.|e| = (x′ − x)2 + (y ′ − y)2 + (z ′ − z)2 ) between the systems described by Eqs. (46)
and (47) as a function of time (t) for A = 0.6, G(x) = 0.58(x2 − 1)and three specific
sets of eigenvalues of matrix A: (i)λ 1 = −1.2, λ 2 = −1.5and λ 3 = −1.7(ii)λ 1 = −3.6,
λ 2 = −4.5and λ 3 = −5.1 (iii)λ 1 = −7.2, λ 2 = −9.0and λ 3 = −10.2, which correspond
to (i) a = -3.06, b = -6.39 and c = - 4.4 (ii) a = -82.62, b = -57.51 and c = - 13.2 (iii)
a = -660.96, b = -230.04 and c = - 26.4 respectively.
In Figure 16, we have shown the results of numerical computation of Euclidean error
forG(x) = (|x| − 2), however the other parameters are same as for Figure 15. We observe
from Figures 15 and 16 that the Eucledean error decays to zero after a certain time and
identical synchronization of active and passive system occurs. We also note that the
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
61
$
Fig. 13 Euclidean error (i.e.|e| = (x′ − x)2 + (y ′ − y)2 + (z ′ − z)2 ) between the systems described by Eqs. (46) and (47) as a function of time (t) for A = 0.6, G(x) = 0.58(x 2 − 1) and
three specific sets of parameters (i) a = -3.06, b = -6.39 and c = - 4.4 (ii) a = -82.62, b =
-57.51 and c = - 13.2 (iii) a = -660.96, b = -230.04 and c = - 26.
$
Fig. 14 Euclidean error (i.e.|e| = (x′ − x)2 + (y ′ − y)2 + (z ′ − z)2 ) between the systems described by Eqs. (46) and (47) as a function of time (t) for A = 0.6, G(x) = (|x| - 2) and three
specific sets of parameters (i) a = -3.06, b = -6.39 and c = - 4.4 (ii) a = -82.62, b = -57.51
and c = - 13.2 (iii) a = -660.96, b = -230.04 and c = - 26.
results in these two figures are independent of the form of nonlinear function G(x). In
addition, we also observe that for the third parameter set (a, b, c) = (-660.96, -230.04,
-26.4) for which the eigenvalues are larger in magnitude the decay is faster than the
second and first parameter sets.
62
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
In the above described active passive decomposition the Jacobian matrix appeared
in the error equation (Eq. (43)) is a constant matrix and hence the stability of identical
synchronization can be analyzed very easily by simply calculating the eigenvalues of the
Jacobian matrix. The above described way of active passive decomposition provides us
the flexibility to choose the value of constants a, b and c freely to attain the identical synchronization. However many more ways are possible to decompose the given dynamical
system into active and passive parts. To illustrate other active passive decompositions,
we again consider the same family of jerk dynamical systems as considered above. For
example, one of the possible active passive decomposition is given by
ẋ = −x + S
ẏ = z
ż = −Az − y + G(x)
(48)
S =x+y
(49)
ẋ′ = −x′ + S
ẏ ′ = z ′
.
ż ′ = −Az ′ − y ′ + G(x′ )
(50)
with
and
The dynamical system describing the evolution of the difference between systems described by Eqs. (48) and (50) is given as,
#
#" ′
# "
" ′
ẋ − ẋ
x −x
−1
0
0
y′ − y .
ẏ ′ − ẏ =
0
0
1
(51)
′
′
(G(x )−G(x))/(x′ − x)
−1
−A
z′ − z
ż − ż
It is clear that in this case the Jacobian matrix appearing in the error equation is not
constant and hence not easy to calculate its eigenvalues easily. For analyzing the stability
of identical synchronization, we have used the numerical computation of the Lyapunov
exponents of Eq. (51) (i.e. conditional Lyapunov exponents). If we consider G(x) =
0.58(x2 − 1)and A = 0.6then conditional Lyapunov exponents are λ1 = −0.3006E0, λ2 =
−0.3039E0, λ3 = −0.9991E0and if we consider G(x) = (|x| − 2) and then conditional
Lyapunov exponents are λ1 = −0.3014E0, λ2 = −0.3012E0, λ3 = −0.9999E0. It is clear
that for both cases all the three conditional Lyapunov exponents are negative and hence
the identical synchronization is stable. Other possible active passive decompositions and
their corresponding conditional Lyapunov exponents for the above discussed family of
jerk dynamical system with G(x) = 0.58(x2 − 1)and G(x) = (|x| − 2) are given in
Tables 4 and 5 respectively.
From Tables 4 and 5, it is clear that several active passive decompositions are possible
which lead to stable identical synchronization unlike to Pecora-Carroll and feedback techniques where only a few drive-response configurations lead to identical synchronization.
Finally, we extend the active passive decomposition approach for identical synchronization of chaos to the jerk dynamical representations of the Sprott’s simple chaotic flows [9]
by considering their classification [15] according to the hierarchy of quadratic jerk functions with increasing many more terms. In Table 6, we have given a few possible active
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
63
Table 4: Examples of active passive decompositions of the jerk dynamical system having
quadratic nonlinearity and their corresponding conditional Lyapunov exponents (CLE’s)
ẋ = y − x + s
ẏ = z
ż = −Az − y + B(x2 + C)
ẋ = −x − z + s
ẏ = z
ż = −Az − y + B(x2 + C)
ẋ = y
ẏ = z − x + s
ż = −Az − y + B(x2 + C)
ẋ = y
ẏ = z − y + s
ż = −Az − y + B(x2 + C)
ẋ = y
ẏ = −y + s
ż = −Az − y + B(x2 + C)
ẋ = y
ẏ = −x − y + s
ż = −Az − y + B(x2 + C)
ẋ = x − x2 + s
ẏ = z
ż = −Az − y + B(x2 + C)
ẋ = y
ẏ = y − xz + s
ż = −Az − y + B(x2 + C)
...
x = −Aẍ − ẋ + B(x2 + C); A = 0.6; B = 0.58; C = −1
ẋ′ = y ′ − x′ + s
s=x
ẏ ′ = z ′
ż ′ = −Az ′ − y ′ + B(x′2 + C)
ẋ′ = −x′ − z ′ + s
s=x+y+z
ẏ ′ = z ′
ż ′ = −Az ′ − y ′ + B(x′2 + C)
ẋ′ = y ′
s=x
ẏ ′ = z ′ − x′ + s
ż ′ = −Az ′ − y ′ + B(x′2 + C)
ẋ′ = y ′
s=y
ẏ ′ = z ′ − y ′ + s
ż ′ = −Az ′ − y ′ + B(x′2 + C)
ẋ′ = y ′
s=y+z
ẏ ′ = −y ′ + s
ż ′ = −Az ′ − y ′ + B(x′2 + C)
ẋ′ = y ′
s=x+y+z
ẏ ′ = −x′ − y ′ + s
ż ′ = −Az ′ − y ′ + B(x′2 + C)
ẋ′ = x′ − x′2 + s
s = x2 + y − x
ẏ ′ = z ′
ż ′ = −Az ′ − y ′ + B(x′2 + C)
ẋ′ = y ′
s = xz + z − y
ẏ ′ = y ′ − x′ z ′ + s
ż ′ = −Az ′ − y ′ + B(x′2 + C)
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
= −0.1005E0
= −0.3188E0
= −0.1181E + 1
= −0.1798E0
= −0.2354E0
= −0.1184E + 1
= −0.3695E − 1
= −0.1555E0
= −0.4081E0
= −0.9567E − 1
= −0.5749E0
= −0.9308E0
= −0.2313E − 2
= −0.6016E0
= −0.1000E + 1
= −0.4979E0
= −0.5013E0
= −0.6013E0
= −0.2620E0
= −0.3024E0
= −0.1884E + 1
= −0.8211E − 1
= −0.1118E0
= −0.5253E0
Table 5: Examples of active passive decompositions of the jerk dynamical system having absolute
nonlinearity and their corresponding conditional Lyapunov exponents (CLE’s)
ẋ = y − x + s
ẏ = z
ż = −Az − y + B(|x| + C)
ẋ = −z + s
ẏ = z
ż = −Az − y + B(|x| + C)
ẋ = −x − z + s
ẏ = z
ż = −Az − y + B(|x| + C)
ẋ = y
ẏ = z − x + s
ż = −Az − y + B(|x| + C)
ẋ = y
ẏ = z − y + s
ż = −Az − y + B(|x| + C)
ẋ = y
ẏ = −x + s
ż = −Az − y + B(|x| + C)
ẋ = y
ẏ = −y + s
ż = −Az − y + B(|x| + C)
ẋ = y
ẏ = −x − y + s
ż = −Az − y + B(|x| + C)
ẋ = x − x2 + s
ẏ = z
ż = −Az − y + B(|x| + C)
...
x = −Aẍ − ẋ + B(|x| + C); A = 0.6; B = 1.0; C = −2
ẋ′ = y ′ − x′ + s
s=x
ẏ ′ = z ′
ż ′ = −Az ′ − y ′ + B(|x′ | + C)
ẋ′ = −z ′ + s
s=y+z
ẏ ′ = z ′
ż ′ = −Az ′ − y ′ + B(|x′ | + C)
ẋ′ = −x′ − z ′ + s
s=x+y+z
ẏ ′ = z ′
ż ′ = −Az ′ − y ′ + B(|x′ | + C)
ẋ′ = y ′
s=x
ẏ ′ = z ′ − x′ + s
ż ′ = −Az ′ − y ′ + B(|x′ | + C)
ẋ′ = y ′
s=y
ẏ ′ = z ′ − y ′ + s
ż ′ = −Az ′ − y ′ + B(|x′ | + C)
ẋ′ = y ′
s=x+z
ẏ ′ = −x′ + s
ż ′ = −Az ′ − y ′ + B(|x′ | + C)
ẋ′ = y ′
s=y+z
ẏ ′ = −y ′ + s
ż ′ = −Az ′ − y ′ + B(|x′ | + C)
ẋ′ = y ′
s=x+y+z
ẏ ′ = −x′ − y ′ + s
ż ′ = −Az ′ − y ′ + B(|x′ | + C)
ẋ′ = x′ − x′2 + s
s = x2 + y − x
ẏ ′ = z ′
ż ′ = −Az ′ − y ′ + B(|x′ | + C)
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
= −0.7373E − 1
= −0.4001E0
= −0.1127E + 1
= −0.2887E − 2
= −0.7248E − 1
= −0.5270E0
= −0.1601E0
= −0.3662E0
= −0.1074E + 1
= −0.2011E − 1
= −0.1577E0
= −0.4227E0
= −0.1468E0
= −0.5344E0
= −0.9160E0
= −0.1138E − 2
= −0.8297E − 3
= −0.6007E0
= −0.1599E − 2
= −0.5985E0
= −0.9991E0
= −0.4997E0
= −0.4992E0
= −0.6017E0
= −0.2601E0
= −0.3005E0
= −0.3616E + 1
64
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
passive decompositions for the jerk dynamical representations of some of the Sprott’s
simple chaotic flows and their corresponding CLE’s (we have given the results for only
one model from each class i.e., one each from JD1 to JD6 ).
Table 6: Examples of active passive decompositions for the jerk dynamical representations of some of the Sprott’s simple chaotic flows and their corresponding conditional
Lyapunov exponents (CLE’s).
JD1
Model J
ẋ = y − x + s
ẏ = z
ż = k1 z + k2 x + xy + k3
ẋ = −x + s
ẏ = z
ż = k1 z + k2 x + xy + k3
ẋ = y
ẏ = z − y + s
ż = k1 z + k2 x + xy + k3
ẋ = y
ẏ = −x − y + s
ż = k1 z + k2 x + xy + k3
JD2
Model S
ẋ = −x + s
ẏ = z
ż = k1 z + k2 y + x2 + k3
ẋ = y
ẏ = z − y + s
ż = k1 z + k2 y + x2 + k3
ẋ = y
ẏ = −x − y + s
ż = k1 z + k2 y + x2 + k3
JD3
Model F
ẋ = y − x + s
ẏ = z
ż = k1 z + k2 y + k3 x2
+xy + k4
ẋ = −x + s
ẏ = z
ż = k1 z + k2 y + k3 x2
+xy + k4
...
x = k1 ẍ + k2 x + xẋ + k3 , k1 = −2.0 ; k2 = −4.0 ; k3 = −4.0
ẋ′ = y ′ − x′ + s
s=x
ẏ ′ = z ′
ż ′ = k1 z ′ + k2 x′ + x′ y ′ + k3
ẋ′ = −x′ + s
s=x+y
ẏ ′ = z ′
ż ′ = k1 z ′ + k2 x′ + x′ y ′ + k3
ẋ′ = y ′
s=y
ẏ ′ = z ′ − y ′ + s
ż ′ = k1 z ′ + k2 x′ + x′ y ′ + k3
ẋ′ = y ′
s=x+y+z
ẏ ′ = −x′ − y ′ + s
ż ′ = k1 z ′ + k2 x′ + x′ y ′ + k3
...
x = k1 ẍ + k2 ẋ + x2 + k3 , k1 = −1.0 ; k2 = −4.0 ; k3 = −16.0
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
λ1
λ2
λ3
= −0.1603E − 1
= −0.4438E0
= −0.2544E + 1
= −0.1399E0
= −0.9985E0
= −0.1862E + 1
= −0.1496E0
= −0.5205E0
= −0.2332E + 1
= −0.4996E0
= −0.5002E0
= −0.2000E + 1
ẋ′ = −x′ + s
λ1 = −0.4971E0
s=x+y
ẏ ′ = z ′
λ2 = −0.5007E0
λ3 = −0.1003E + 1
ż ′ = k1 z ′ + k2 y ′ + x′2 + k3
ẋ′ = y ′
λ1 = −0.6668E − 1
s=y
λ2 = −0.5307E0
ẏ ′ = z ′ − y ′ + s
λ3 = −0.1401E + 1
ż ′ = k1 z ′ + k2 y ′ + x′2 + k3
ẋ′ = y ′
λ1 = −0.4958E0
s=x+y+z
λ2 = −0.5004E0
ẏ ′ = −x′ − y ′ + s
λ3 = −0.1004E + 1
ż ′ = k1 z ′ + k2 y ′ + x′2 + k3
...
x = k1 ẍ + k2 ẋ + k3 x2 + xẋ + k4 , k1 = −0.5 ; k2 = −2.5 ; k3 = −0.25; k4 = 1
ẋ′ = y ′ − x′ + s
ẏ ′ = z ′
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+x′ y ′ + k4
ẋ′ = −x′ + s
ẏ ′ = z ′
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+x′ y ′ + k4
s=x
λ1 = −0.8347E − 1
λ2 = −0.1432E0
λ3 = −0.1275E + 1
s=x+y
λ1 = −0.9779E − 1
λ2 = −0.4059E0
λ3 = −0.1000E + 1
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
ẋ = y
ẏ = z − y + s
ż = k1 z + k2 y + k3 x2
+xy + k4
ẋ = y
ẏ = −y + s
ż = k1 z + k2 y + k3 x2
+xy + k4
ẋ = y
ẏ = −x − y + s
ż = k1 z + k2 y + k3 x2
+xy + k4
JD4
Model O
ẋ = y − x + s
ẏ = z
ż = k1 z + k2 y + k3 x2
+xz + k4
ẋ = −x + s
ẏ = z
ż = k1 z + k2 y + k3 x2
+xz + k4
ẋ = −x − z + s
ẏ = z
ż = k1 z + k2 y + k3 x2
+xz + k4
ẋ = y
ẏ = z − x + s
ż = k1 z + k2 y + k3 x2
+xz + k4
ẋ = y
ẏ = z − y + s
ż = k1 z + k2 y + k3 x2
+xz + k4
ẋ = y
ẏ = −x − y + s
ż = k1 z + k2 y + k3 x2
+xz + k4
65
ẋ′ = y ′
λ1 = −0.6420E − 2
ẏ ′ = z ′ − y ′ + s
s=y
λ2 = −0.4293E0
′
′
′
′2
ż = k1 z + k2 y + k3 x
λ3 = −0.1065E + 1
′
′
+x y + k4
ẋ′ = y ′
λ1 = −0.1957E − 2
ẏ ′ = −y ′ + s
s=y+z
λ2 = −0.5019E0
′
′
′
′2
ż = k1 z + k2 y + k3 x
λ
3 = −0.1000E + 1
+x′ y ′ + k4
′
′
ẋ = y
λ1 = −0.4964E0
ẏ ′ = −x′ − y ′ + s
s
=
x
+
y
+
z
λ2 = −0.5003E0
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
λ3 = −0.5039E0
′
′
+x y + k4
...
x = k1 ẍ + k2 ẋ + k3 x2 + xẍ + k4 , k1 = −0.5 ; k2 = −1.7 ; k3 = −1.0; k4 = 0.25
ẋ′ = y ′ − x′ + s
ẏ ′ = z ′
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+x′ z ′ + k4
′
ẋ = −x′ + s
ẏ ′ = z ′
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+x′ z ′ + k4
′
ẋ = −x′ − z ′ + s
ẏ ′ = z ′
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+x′ z ′ + k4
ẋ′ = y ′
ẏ ′ = z ′ − x′ + s
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+x′ z ′ + k4
ẋ′ = y ′
ẏ ′ = z ′ − y ′ + s
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+x′ z ′ + k4
′
ẋ = y ′
ẏ ′ = −x′ − y ′ + s
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+x′ z ′ + k4
s=x
λ1 = −0.2513E − 1
λ2 = −0.3436E0
λ3 = −0.9209E0
s=x+y
λ1 = −0.1325E0
λ2 = −0.1375E0
λ3 = −0.9996E0
s=x+y+z
λ1 = −0.1181E0
λ2 = −0.1189E0
λ3 = −0.1032E + 1
s=x
λ1 = −0.5036E − 1
λ2 = −0.7334E − 1
λ3 = −0.1455E0
s=y
λ1 = −0.1205E0
λ2 = −0.4668E0
λ3 = −0.6832E0
s=x+y+z
λ1 = −0.2681E0
λ2 = −0.4996E0
λ3 = −0.5004E0
66
JD5
Model D
ẋ = y − x + s
ẏ = z
ż = k1 z + k2 x2 + k3 y 2 + xz
ẋ = −x − z + s
ẏ = z
ż = k1 z + k2 x2 + k3 y 2 + xz
ẋ = y
ẏ = z − y + s
ż = k1 z + k2 x2 + k3 y 2 + xz
ẋ = y
ẏ = −x − y + s
ż = k1 z + k2 x2 + k3 y 2 + xz
JD6
Model P
ẋ = y − x + s
ẏ = z
ż = k1 z + k2 y + k3 x2
+k4 y 2 + xz + k5
ẋ = −x + s
ẏ = z
ż = k1 z + k2 y + k3 x2
+k4 y 2 + xz + k5
ẋ = y
ẏ = z − y + s
ż = k1 z + k2 y + k3 x2
+k4 y 2 + xz + k5
ẋ = y
ẏ = −x − y + s
ż = k1 z + k2 y + k3 x2
+k4 y 2 + xz + k5
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
...
x = k1 ẍ + k2 x2 + k3 ẋ2 + xẍ, k1 = −1.0 ; k2 = 1.0 ; k3 = −1.0
ẋ′ = y ′ − x′ + s
λ1 = −0.8313E − 1
s=x
ẏ ′ = z ′
λ2 = −0.2672E0
λ3 = −0.1598E + 1
ż ′ = k1 z ′ + k2 x′2 + k3 y ′2 + x′ z ′
ẋ′ = −x′ − z ′ + s
λ1 = −0.5850E0
s=x+y+z
ẏ ′ = z ′
λ2 = −0.4168E0
λ3 = −0.2352E + 1
ż ′ = k1 z ′ + k2 x′2 + k3 y ′2 + x′ z ′
ẋ′ = y ′
λ1 = −0.3648E − 1
s=y
λ2 = −0.4033E0
ẏ ′ = z ′ − y ′ + s
λ3 = −0.1744E + 1
ż ′ = k1 z ′ + k2 x′2 + k3 y ′2 + x′ z ′
ẋ′ = y ′
λ1 = −0.4979E0
s=x+y+z
λ2 = −0.5011E0
ẏ ′ = −x′ − y ′ + s
λ3 = −0.1184E + 1
ż ′ = k1 z ′ + k2 x′2 + k3 y ′2 + x′ z ′
...
x = k1 ẍ + k2 ẋ + k3 x2 + k4 ẋ2 + xẍ + k5 ; k1 = −1.0; k2 = −1.7; k3 = −0.5; k4 = 1.0; k5 = 0.5
ẋ′ = y ′ − x′ + s
ẏ ′ = z ′
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+k4 y ′2 + x′ z ′ + k5
ẋ′ = −x′ + s
ẏ ′ = z ′
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+k4 y ′2 + x′ z ′ + k5
ẋ′ = y ′
ẏ ′ = z ′ − y ′ + s
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+k4 y ′2 + x′ z ′ + k5
′
ẋ = y ′
ẏ ′ = −x′ − y ′ + s
ż ′ = k1 z ′ + k2 y ′ + k3 x′2
+k4 y ′2 + x′ z ′ + k5
s=x
λ1 = −0.2839E − 1
λ2 = −0.2464E0
λ3 = −0.1119E + 1
s=x+y
λ1 = −0.1826E0
λ2 = −0.2120E0
λ3 = −0.9995E0
s=y
λ1 = −0.3656E − 1
λ2 = −0.4006E0
λ3 = −0.8625E0
s=x+y+z
λ1 = −0.3878E0
λ2 = −0.4997E0
λ3 = −0.5007E0
Conclusions
In this paper, we have investigated identical synchronization in chaotic jerk dynamical systems by implementing three algorithms: Pecora-Carroll (PC) technique, Feedback
(FB) technique and active passive decomposition (APD. We have also investigated the
stability of identical synchronization in the aforesaid methods using extensive transversal
stability analysis (i.e., extensive conditional Lyapunov exponents (CLE’s) calculation)
as well as Lyapunov function construction criterion. We summarize the results of our
investigations as follows:
• In case of Pecora-Carroll (PC) technique- Our extensive numerical calculation reveals
that the x-drive configuration for all the jerky representation of the Sprott’s simple
chaotic flows by considering their classification into seven different classes of jerky
dynamics (according to the hierarchy of quadratic jerk functions with increasing
many more terms), guarantees the identical synchronization except for the models
‘L’, ‘R’ and ‘D’, where no subsystem is found to be stable. However y-drive and
z-drive configurations are unstable for all the jerk dynamical systems.
• In Case of feedback (FB) technique- we have done the extensive calculation for
conditional Lyapunov exponents by implementing FB technique with x-drive configuration in the jerky representations of the Sprott’s chaotic flows by considering
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
67
their classification into seven different classes of jerky dynamics (according to the
hierarchy of quadratic jerk functions with increasing many more terms). We have
done the extensive calculations of CLE’s for the complete range of coupling strength
i.e. 0 ≤ c ≤ 1 for all the jerk systems. We observed that the feedback technique
works well for synchronizing all the chaotic jerk dynamical systems considered by
us. However y-drive and z-drive configurations do not lead to stable identical synchronization.
• In case of active passive decomposition (APD) –we have suggested how it is possible to decompose any jerk dynamical system into active and passive elements to
achieve the stable synchronization without being bothered about the specific form of
jerk function. Besides this generalized decomposition of jerk dynamical system into
active and passive elements, we have also suggested several other active and passive decompositions for the jerky representation of the Sprott’s simple chaotic flows
to achieve the stable identical synchronization and calculated their corresponding
conditional Lyapunov exponents.
As the jerk dynamical systems describe various phenomena in physics and engineering
such as electrical circuits, mechanical oscillators, laser physics, solar wind driven magnetosphere ionosphere (WINDMI) model, damped harmonic oscillator driven by nonlinear
memory term, biological systems etc. It has been also shown that many three dimensional
dynamical systems can be transformed into jerk dynamical system [15], hence results of
our investigations for the identical synchronization of chaos are of great use in diverse
areas of science, where we would like to behave two or more systems in synchronized
manner although maintaining their chaotic nature.
68
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
References
[1] S. Wiggins, Introduction to applied nonlinear dynamical systems and chaos; SpringerVerlag (1990).
[2] E. Ott, Chaos in dynamical systems; Cambridge University Press, Cambridge (1993).
[3] J. Argyris, G. Faust, M. Hasse, An exploration of chaos; North-Holland, Amsterdam
(1994).
[4] S. H. Strogatz, Nonlinear dynamics and chaos: with applications to Physics, Biology,
Chemistry and Engineering; Addison-Wesley, Reading, RI. (1994)
[5] R. C. Hillborn, Nonlinear dynamics and chaos; Oxford University Press, Oxford
(2001).
[6] H. Fujisaka, T. Yamada, Progress in Theoretical Physics 69, 32 (1983).
[7] H. Fujisaka, T. Yamada, Progress in Theoretical Physics 70, 1240 (1983).
[8] T. Yamada, H. Fujisaka, Progress in Theoretical Physics 72, 885 (1984).
[9] J. C. Sprott, Phys. Rev. E 50, R647 (1994).
[10] W. G. Hoover, Phys. Rev. E 51, 759 (1995).
[11] H. A. Posch, W. G. Hoover, F. J. Vesely, Phys. Rev. A 33, 4253 (1986).
[12] H. P. W. Gottlieb, American Journal of Physics 64, 525 (1996).
[13] S. H. Schot, American Journal of Physics 46, 1090 (1978).
[14] S. J. Linz, American Journal of Physics 65, 523 (1997).
[15] R. Eichhorn, S. J. Linz., P. Hänggi, Phys. Rev. E 58, 7151 (1998).
[16] O. E. Rössler, Ann.(N.Y.) Acad. Science 316, 376-392 (1979).
[17] J. C. Sprott, Physics Letters A 228, 271 (1997).
[18] R. Eichhorn, S. J. Linz., P. Hänggi, Chaos, Solitons and Fractals 13, 1 (2002).
[19] Vinod Patidar and K. K. Sud, Pramana-J.Phys. 64, 75 (2005).
[20] Vinod Patidar, N. K. Pareek and K. K. Sud, Physics Letters A 304, 121 (2002).
[21] Vinod Patidar and K. K. Sud, Ind. J. Phys. 76B, 553 (2002).
[22] V. S. Afraimovich, N. N. Verichev, M. I. Rabinovich, Inv. VUZ Rasiofiz RPQAEC
29, 795-803 (1986).
[23] A. S. Pikovski, Z. Phys. B. 55, 149 (1984).
[24] L. M. Pecora, T. L. Carroll, Physical Review Letters 64, 821-824 (1990).
[25] L. M. Pecora, T. L. Carroll, Physical Review A 44, 2374-2383 (1991).
[26] K. Pyragas, Phys. Rev. E 54, R4508-R4511 (1996).
[27] R. He, P. G. Vaidya, Physical Review A 46, 7387-7392 (1992).
[28] T. Kapitaniak, Phys. Rev. E 50, 1642-1644 (1994).
[29] R. E. Amritkar, N. Gupte, Phys. Rev. E 47, 3889-3895 (1993).
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
69
[30] L. Kocarev, U. Parlitz, Phys. Rev. Lett. 74, 5028-5031 (1995).
[31] U. Parlitz, Phys. Rev. Lett. 76, 1232-1235 (1996).
[32] J. Guemez, M. A., Matias, Phys. Rev. E 52, R2145-R2148 (1995).
[33] L. Benettin, L. Galgani, A. Giorgilli, J.-M. Strelcyn, Mechanica 15, 9-20 (1980).
[34] L. Benettin, L. Galgani, A. Giorgilli, J.-M. Strelcyn, Mechanica 15, 21-30 (1980).
[35] A. Wolf, J. B. Swift, H. L. Swinney, J. A. Vastano, Physica D 16, 285-317 (1985).
[36] G. Rangrajan., S. Habib, R. D. Ryne, Phys. Rev. Lett. 80, 3747-3750 (1998).
[37] T. L. Carroll, L. M. Pecora, IEEE Trans. Circ. Systs. 38, 453 (1991).
[38] T. L. Carroll, L. M. Pecora, Int. J. Bif. Chaos 2, 659-667 (1992).
[39] T. L. Carroll, L. M. Pecora, Physica D 67, 126-140 (1993).
[40] T. L. Carroll, L. M. Pecora, IEEE Trans. Circ. Systs. 40, 646 (1993).
[41] K. M..Cuomo, A. V. Oppenheim, Phys. Rev. Lett. 71, 65-68 (1993).
[42] K. M. Cuomo, A. V. Oppenheim, S. H. Strogatz, IEEE Trans. Circ. Systs. 40, 626-633
(1993).
[43] L. O. Chua, L. Kocarev, K. Eckert, M. Itoh, Int. J. Bif. Chaos 2, 705 (1992).
[44] L. Kocarev, K. S.Halle, K. Eckert, L. O. Chua, U. Parlitz, Int. J. Bif. Chaos 2, 973
(1992).
[45] K. Murali, M. Lakshmanan, Phys. Rev. E 48, R1624-R1626 (1993).
[46] M. Lakshmanan, K. Murali Current Science 67, 989 (1994).
[47] M. Lakshmanan, K. Murali, Chaos in nonlinear oscillators:
synchronization, World Scientific, Singapore (1996).
controlling and
[48] T. Endo, L. O. Chua, Int. J. Bif. Chaos 1, 701 (1991).
[49] M. De Sousa Viera, A. J. Lichtenberg, M. A. Lieberman, Int. J. Bif. Chaos 1, 691
(1991).
[50] M. De Sousa Viera, A. J. Lichtenberg, M. A. Lieberman, Phys. Rev. A 46, R7359R7362 (1992).
[51] J. K. John, R. E. Amritkar, Phys. Rev. E. 49, 4843-4847 (1994).
[52] J. K. John, R. E. Amritkar, Int. J. Bif. & Chaos 4, 1687-1695 (1994).
[53] R. E. Amritkar, A. Maybhate Parameter estimation and synchronization using
adaptive control; in Nonlinear dynamics and computational dynamics (Ed. Sheorey
V. B.), Narosa Publishing House, New Delhi (1999)
[54] J. Singer, Y.-Z. Wang, H. H. Bau, Phys. Rev. Lett. 66, 1123-1125 (1991).
[55] K. Pyragas, Phys. Lett. A 170, 421-428 (1992).
[56] G. Chen, X. Dong, Int. J. Bif. Chaos 2, 207 (1992).
[57] K. Pyragas, Phys. Lett. A 181, 203-210 (1993).
[58] T. Kapitaniak, L. O. Chua, Int. J. Bif. Chaos 4, 477 (1994).
70
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 33–70
[59] T. Kapitaniak, L. O. Chua, G. Q. Zhong, Int. J. Bif. Chaos 4, 483 (1994).
[60] U. Parlitz, L. Kocarev, T. Stojanovski, H. Preckel, Phys. Rev. E 53, 4351-4361
(1996).
[61] C. W. Wu, L. O Chua, Int. J. Bifurcation & Chaos 3, 1619 (1993).
[62] W. Jinlan, C. Guangzhi, Q. Tuanfa, N. Wansun, W. Xuming, Phys. Rev. E 58,
3017-3027 (1998).
EJTP 3, No. 11 (2006) 71–83
Electronic Journal of Theoretical Physics
Second Order Perturbation of Heisenberg
Hamiltonian for Non-Oriented Ultra-Thin
Ferromagnetic Films
P. Samarasekara
∗
Department of Physics,
University of Ruhuna,
Matara, Sri Lanka
Received 25 January 2006 , Accepted 15 March 2006, Published 25 June 2006
Abstract:
The second order perturbation of magnetic energy for ferromagnetic thin films of two and three layers has
been studied using classical Heisenberg Hamiltonian. According to our model, the film with two layers is equivalent to an
oriented film, when anisotropy constants do not vary inside the film. But the energy of films with three layers indicates
periodic variation. Introducing second order perturbation induces some sudden overshooting of energy curves, compared
with smooth energy curves obtained for oriented ferromagnetic ultra thin films in one of our previous report. After taking the
fourth order anisotropy into account, the overshooting part dominates by reducing the smooth part of energy graphs. Several
minimums can be observed in last 3-D graph implying that the film with N=3 can be oriented in some preferred directions by
applying a certain value of stress. The shape of graphs of energy variation of all sc(001), fcc(001) and bcc(001) ferromagnetic
ultra thin films with second (or fourth) order anisotropy is exactly same. Easy and hard directions of these all types with
the effect of second order anisotropy only are 34.40 and 124.40 , respectively. The angle between easy and hard directions
J
ω
Ks
ω
is exactly 900 as expected. Although these simulations were given for
only, this same approach can be carried out for any values of
(2)
J Dm
, ω
ω
,
and
(2)
(4)
Dm
= 10, Kωs = 10 and Dωm = 5 values
ω
(4)
Dm
or any type of ferromagnetic material.
ω
= 10,
Considering the other terms such as dipole interaction and demagnetization factor really complicates the simulation.
c Electronic Journal of Theoretical Physics. All rights reserved.
Keywords: Heisenberg Hamiltonian, Ferromagnetic Films,
PACS (2006): 75.10.Hk , 75.30.Gw , 75.70.-i
1.
Introduction:
The research of exchange anisotropy has received a wide attention in last decade, due
to the difficulties of physical understanding of exchange anisotropy and to its application
in magnetic media technology and magnetic sensors [4] . The magnetic properties of ferromagnetic thin films and multi layers have been extensively investigated because of their
∗
pubudus@phy.ruh.ac.lk
72
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
potential impact on magnetic recording devices. The magnetic properties of thin films of
ferromagnetic materials have been investigated using the Bloch spin-wave theory earlier
[5] . The magnetization of some thin films shows an in plane orientation due to the dipole
interaction. Due to the broken symmetry uniaxial anisotropy energies at the surfaces of
the film, the perpendicular magnetization is preferential. But due to the strain induced
distortion in the inner layers, bulk anisotropy energies will appear absent or very small
in the ideal crystal. Very thin films indicate a tetragonal distortion resulting in stressinduced uniaxial anisotropy energy in the inner layers with perpendicular orientation of
easy axis. The magnetic in-plane anisotropy of a square two-dimensional (2D) Heisenberg
ferromagnet in the presence of magnetic dipole interaction has been determined earlier
[6] . The long range character of the dipole interaction itself is sufficient to stabilize the
magnetization in 2-D magnet. Also the easy and hard axes of the magnetization with
respect to lattice frame are determined by the anisotropies. Magnetic properties of the
Ising ferromagnetic thin films with alternating superlattice layers were investigated [7] .
In addition to these, Monte Carlo simulations of hysteresis loops of ferromagnetic thin
films have been theoretically traced [8] . Since the surfaces slightly distort the symmetry
of the system under consideration, physical quantities in the vicinity of surfaces generally
deviate from those in the bulk.
The second order perturbation solution of Heisenberg Hamiltonian for ultra-thin ferromagnetic films has been found without considering the effect of stress induced anisotropy
and demagnetization factor in some early reports [1] . The stress of a film arises mainly
due to the difference between thermal expansion coefficients of the film and the substrate.
When the film is cooled down or heated after annealing or deposition, the stress takes
place in the film. The effect of stress on the coercivity and anisotropy of sputtered ferromagnetic thin films has been previously studied by us [2,3] . The contribution of stress to
the magnetic energy is given by Ks sin2θ. Here Ks depends on the magnetization and the
magnitude of stress, and θ is the angle between the stress and the magnetization. The
theoretical values of demagnetization factor within the film plane and normal to the film
plane are 0 and 1 in SI units, respectively.
2.
The Model
The Heisenberg Hamiltonian of ferromagnetic films can be formulated as following.
H=−
(2)
(4)
n 3(S
n )
m .S
m .rmn )(rmn .S
ω S
J
z 2
z 4
Sm .Sn +
( 3
−
)
−
D
(S
)
−
Dλm (Sm
)
m
λ
m
5
2 m,n
2 m=n rmn
rmn
m
m
−
m,n
− (Nd S
m −
n /μ0 )].S
[H
Ks Sin2θm
m
m and S
n are two spins. Above equation can be simplified into following form [1].
Here S
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
E(θ)= - 12
N
m,n=1
−
[(JZ|m−n| − ω4 Φ|m−n| ) cos(θm − θn ) −
N
73
3ω
Φ|m−n|
4
cos(θm + θn )]
(2)
(4)
(Dm
cos2 θm +Dm
cos4 θm + Hin sin θm + Hout cos θm )
m=1
+
N
N
Nd
cos(θm − θn ) − Ks
sin 2θm
μ
0
m=1
m,n=1
(2)
(1)
(4)
Here J, Z|m−n| , ω, Φ|m−n| , θ, Dm , Dm , Hin , Hout , Nd , Ks , m, n and N are spin exchange
interaction, number of nearest spin neighbors, strength of long range dipole interaction,
constants for partial summation of dipole interaction, azimuthal angle of spin, second
and fourth order anisotropy constants, in plane and out of plane applied magnetic fields,
demagnetization factor, stress induced anisotropy constant, spin plane indices and total
number of layers in film, respectively. When the stress applies normal to the film plane,
the angle between mth spin and the stress is θm .
The azimuthual angles of spins can be given as θm = θ + εm and θn = θ + εn . After
substituting these new angles in above equation number 1, the cosine and sine terms can
be expanded up to the second order of εm and εn as following.
E(θ)=E0 +E(ε)+E(ε2 )+———
If the third and higher order perturbations are neglected, then
E(θ) = E0 + E(ε) + E(ε2 )
Here
(2)
N
N
1
ω
3ω
cos 2θ
Φ|m−n|
(JZ|m−n| − Φ|m−n| ) +
E0 = −
2 m,n=1
4
8
m,n=1
2
− cos θ
N
(2)
Dm
m=1
4
− cos θ
N
m=1
(4)
Dm
− N (Hin sin θ + Hout cos θ −
Nd
+ Ks sin 2θ)
μ0
(3)
N
N
N
3ω
(4)
(2)
2
sin 2θ
Dm
εm
E(ε) = −
Dm εm + 2 cos θ sin 2θ
Φ|m−n| (εm + εn ) + sin 2θ
8
m=1
m=1
m,n=1
−Hin cos θ
E(ε2 ) =
N
εm + Hout sin θ
m=1
N
m=1
εm − 2Ks cos 2θ
N
εm
m=1
N
N
ω
3ω
1
cos 2θ
(JZ|m−n| − Φ|m−n| )(εm − εn )2 −
Φ|m−n| (εm + εn )2
4 m,n=1
4
16
m,n=1
2
2
−(sin θ − cos θ)
N
m=1
(2) 2
Dm
εm
2
2
2
+ 2 cos θ(cos θ − 3 sin θ)
N
(4) 2
Dm
εm
m=1
N
N
N
N
Hin
Hout
Nd
2
2
2
sin θ
cos θ
+
(εm − εn ) + 2Ks sin 2θ
ε2m
εm +
εm −
2
2
2μ
0
m,n=1
m=1
m=1
m=1
74
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
After using the constraint
N
εm = 0, E(ε) = α
.ε
m=1
Here α
(ε) = B(θ)
sin 2θ are the terms of matrices with
N
3ω
(4)
Φ|λ−m| + Dλ (2) + 2Dλ cos2 θ
Bλ (θ) = −
4 m=1
(4)
Also E(ε2 ) = 21 ε.C.ε
Here the elements of matrix C can be given as following,
Cmn = −(JZ|m−n| −
+δmn {
N
λ=1
ω
3ω
2Nd
Φ|m−n| ) −
cos 2θΦ|m−n| +
4
4
μ0
[JZ|m−λ| − Φ|m−λ| (
ω 3ω
(2)
+
cos 2θ)] − 2(sin2 θ − cos2 θ)Dm
4
4
(4)
+4 cos2 θ(cos2 θ − 3 sin2 θ)Dm
+ Hin sin θ + Hout cos θ −
4Nd
+ 4Ks sin 2θ}
μ0
(5)
Therefore, the total magnetic energy given in equation 2 can be deduced to [1]
1
1
· C+ · α
· ε + ε · C · ε = E0 − α
E(θ) = E0 + α
2
2
(6)
Here C+ is the pseudo-inverse given by
C · C+ = 1 −
E
N
(7)
Here E is the matrix with all elements Emn =1.
3.
Results and Discussion:
First the energy will be found for a film with two layers (N=2). Since it is reasonable
to assume that the anisotropy energies remain constant within an ultra-thin film with
(2)
(2)
(4)
(4)
two layers, D1 =D2 and D1 =D2 .
From equation 5, C11 =C22 and C12 =C21 .
+
+
+
+
Therefore from equation 7, C12
= C21
= 2(C211−C22 ) = −C11
= −C22
+
Using above results, α
· C+ · α
= (α1 − α2 )2 C11
But from equation 4, for a film with two layers α1 = α2
Therefore α
· C+ · α
= 0, and the total energy in equation 6 is E0 . This means that
the energy of a film with two layers is reduced to the energy of the oriented film.
Now if the anisotropy constants change within the film, then C12 =C21 and C22 =C11 .
+
+
+
+
21
= C221 +C11 ) .
= −C22
and C21
= 2(CC1122C+C
= −C12
Therefore, C11
2
22 −C )
21
2(C21 −C11 C22
+
+
= (α1 − α2 )(C21
α2 − C12
α1 )
Hence, α
· C+ · α
Now the total energy is different from that of an oriented film. The matrix elements
of C+ , which were found using equation 7 for a film with three layers (N=3), contained
about 20 terms of Cmn elements.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
75
The C+
31 has been given as an example below.
+
A
C31 = D
Here A=-[(C22 C11 -C12 C21 )-0.67(C11 -C21 )(C12 -C22 )]C31 -0.67C32 (C11 -C21 )2
-0.33(C22 C11 -C12 C21 )(C11 -C21 )
D=C31 [(C11 C23 -C13 C21 )(C12 -C22 )-(C13 -C23 )(C22 C11 -C12 C21 )]
-(C11 C23 -C13 C21 )(C11 -C21 )C32 +(C22 C11 -C12 C21 )(C11 -C21 )C33
Therefore, the final result became really complicated. To avoid this problem, the
matrix elements were found for special case, C · C + =1.
·α
Under this condition, energy is given by equation 6 only if E
= 0.
Therefore, α1 + α2 + α3 = 0. This implies that the average value of first perturbation
α1 +α2 +α3
) is zero under this condition. Now the C+ is the standard inverse of a matrix,
(
3
+
nm
given by matrix elementCmn
= cof actorC
. Also the matrix elements Cmn can be given
det C
as following according to equation 5.
2Nd
ω
Φ1 (1 − 3 cos 2θ) +
4
μ0
ω
2Nd
C31 = −JZ2 + Φ2 (1 − 3 cos 2θ) +
4
μ0
ω
2Nd
(2)
C33 = J(Z1 + Z2 ) − (Φ1 + Φ2 )(1 + 3 cos 2θ) −
+ (2 cos 2θ)Dm
4
μ0
(4)
4 cos2 θ(cos2 θ − 3 sin2 θ)Dm
+ Hin sin θ + Hout cos θ + 4Ks sin 2θ
2Nd
ω
(2)
+ (2 cos 2θ)Dm
2JZ1 − Φ1 (1 + 3 cos 2θ) −
2
μ0
2
2
2
(4)
4 cos θ(cos θ − 3 sin θ)Dm + Hin sin θ + Hout cos θ + 4Ks sin 2θ
C12 = C21 = C23 = C32 = −JZ1 +
C13 =
C11 =
+
C22 =
+
(8)
Because it is reasonable to assume that second or fourth order anisotropy constants
(2)
(2)
(2)
(4)
(4)
(4)
do not change inside an ultra thin film, D1 =D2 =D3 and D1 =D2 =D3 . Therefore
for the convenience the matrix elements C+
mn will be given in terms of C11 , C22 , C32 , and
C31 only.
+
C11
=
+
C12
=
+
C13
=
+
C22
=
2
C11 C22 − C32
+
= C33
2
2
C11 (C11 C22 − C31 ) + 2C32 (C31 − C11 )
C32 C31 − C32 C11
+
+
+
= C21
= C23
= C32
2
2
C11 (C11 C22 − C31
) + 2C32
(C31 − C11 )
2
C32
− C22 C31
+
= C31
2
2
(C31 − C11 )
C11 (C11 C22 − C31 ) + 2C32
2
2
− C31
C11
2
2
C11 (C11 C22 − C31
) + 2C32
(C31 − C11 )
(9)
Both matrices C and C+ are highly symmetric about both matrix diagonals. From
equation 6,
+
+
+
+
(α12 + α32 ) + C32
(2α1 α2 + 2α2 α3 ) + C31
(2α1 α3 ) + α22 C22
]
E(θ) = E0 − 0.5[C11
(10)
76
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
Because Cmn terms contain sometimes about nine different terms for N=3, the product
of two Cmn terms and hence C+
mn terms will contain about 80 terms. For an example, the
term C11 contains nine different terms, and therefore C+
22 will contain about 81 terms. To
simplify the problem, only the magnetic exchange interaction, second order anisotropy
and stress induced anisotropy terms have been considered for this simulation. For sc(001)
lattice, Z0 =4, Z1 =1, Z2 =0 [1], and from above equations number 4 and 8,
(2)
α = α1 = α2 = α3 =Dm sin2θ, C12 = C21 = C23 = C32 = −J, C13 = C31 = 0
(2)
(2)
C11 = C33 = J + (2 cos 2θ)Dm + 4Ks sin 2θ andC22 = 2J + (2 cos 2θ)Dm + 4Ks sin 2θ
But from equation 10,
E(θ) = E0 − 0.5α2
C11 (2C22 − 4C32 + C11 ) + C31 (4C32 − 2C22 − C31 )
2
2
C11 (C11 C22 − C31
) + 2C32
(C31 − C11 )
(2)
E(θ) = E0 −
1.5[Dm ]2 sin2 2θ
(2)
(2 cos 2θ)Dm + 4Ks sin 2θ
(2)
From equation number 3, E0 = -8J-3[Dm cos2 θ+Ks sin2θ]
(2)
Therefore, E(θ)= -8J-3[Dm cos2 θ+Ks sin2θ]−
(2)
1.5[Dm ]2 sin2 2θ
(2)
(2 cos 2θ)Dm +4Ks sin 2θ
(2)
Dm
= 10, ω = 10, Kωs = 10, then
sin2 2θ
= −80 − 30[cos2 θ+sin2θ]− cos7.5
2θ+2 sin 2θ
The graph between E(θ)
and θ is given in figure 1. If Ks is also a variable, then
ω
E(θ)
sin2 2θ
2
= −80 − 3[10cos θ + Kωs sin2θ]− 20 cos150
ω
2θ+4 Ks sin 2θ
J
ω
E(θ)
ω
If
ω
The 3-D plot of E(θ)
versus Kωs and θ is given in figure 2.
ω
= 0.
When the energy is minimum, ∂E
∂θ
After taking the derivative of above equation, at minimum energy the Kωs satisfies
following cubic equation.
2( Kωs )3 cos2θ+10( Kωs )2 [cos2 2θ-sin2θ]+25( Kωs )cot2θ[2cos2θ–sin2 2θ]
+125[-2cos2θ+2cos2 2θ-sin2 2θ]=0
Using above equation, the preferential direction of orientation for different values of
stress can be calculated.
For bcc(001) lattice Z 0 =0, Z1 =4 and Z2 =0 [1], and hence C12 = C21 = C23 = C32 =
−4J, C13 = C31 = 0,
(2)
+ 4Ks sin 2θ,
C11 = C33 = 4J + (2 cos 2θ)Dm
(2)
C22 = 8J + (2 cos 2θ)Dm
+ 4Ks sin 2θ.
E0 of this lattice is exactly same as that of sc(001). Finally, we can show that energy
E(θ) of bcc(001) lattice is exactly same as that of sc(001) lattice. Using Z0 =4, Z1 =4 and
Z2 =0 for fcc(001) lattice [1]
C12 = C21 = C23 = C32 = −4J, C13 = C31 = 0,
(2)
C11 = C33 = 4J + (2 cos 2θ)Dm
+ 4Ks sin 2θ,
(2)
+ 4Ks sin 2θ.
C22 = 8J + (2 cos 2θ)Dm
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
77
(2)
But E0 = -14J-3[Dm cos2 θ+Ks sin2θ]
(2)
Therefore, E(θ)= -14J-3[Dm cos2 θ+Ks sin2θ]−
(2)
1.5[Dm ]2 sin2 2θ
(2)
(2 cos 2θ)Dm +4Ks sin 2θ
(2)
Dm
Using ωJ = 10, ω = 10, Kωs = 10 for this simulation,
E(θ)
sin2 2θ
= −140 − 30[cos2 θ+sin2θ]− cos7.5
ω
2θ+2 sin 2θ
The graph between E(θ)
and θ in this case is similar to the graph given in figure 1.
ω
According to graph, the easy and hard directions of this film are 34.40 and 124.40 ,
respectively. These angles are higher than those of oriented ferromagnetic thin films
investigated in one of our early report. Therefore, introducing second order perturbation
has increased the angles corresponding to easy and hard directions. Also these angles of
easy and hard directions are valid for all three type of ferromagnetic materials described
in this report. The separation between any two consecutive minimums or maximums is
1800 , similar to regular sine or cosine curve.
When Ks is also a variable,
E(θ)
sin2 2θ
= −140 − 3[10cos2 θ + Kωs sin2θ]− 20 cos150
ω
2θ+4 Ks sin 2θ
ω
versus Kωs and θ in this case is similar to the graph given in
The 3-D plot of E(θ)
ω
figure 2. This graph indicates a slow variation of energy with the variation of the stress
or the angle. The graph corresponding to the minimum energy of bcc(001) and fcc(001)
are exactly same as that of sc(001). Taking dipole interaction into account will give an
energy equation with at least fifty terms, and therefore the effect of dipole interaction
will not be discussed in this report.
After taking fourth order anisotropy term into account for sc(001)
C12 = C21 = C23 = C32 = −J,
C13 = C31 = 0,
(4)
(2)
+ 4Ks sin 2θ
+ 4 cos2 θ(cos2 θ − 3 sin2 θ)Dm
C11 = C33 = J + (2 cos 2θ)Dm
(2)
(4)
C22 = 2J + (2 cos 2θ)Dm
+ 4 cos2 θ(cos2 θ − 3 sin2 θ)Dm
+ 4Ks sin 2θ
(2)
(4)
E(θ)= -8J-3[Dm cos2 θ+Dm cos4 θ+Ks sin2θ]
(2)
−
If
(4)
Dm
= 5,
ω
E(θ)
=
ω
(4)
1.5[Dm + 2Dm cos2 θ]2 sin2 2θ
(2)
(4)
(2 cos 2θ)Dm + 4 cos2 θ(cos2 θ − 3 sin2 θ)Dm + 4Ks sin 2θ
then
7.5[1+cos2 θ]2 sin2 2θ
−80 − 15[2cos2 θ+cos4 θ+2sin2θ] − cos 2θ+cos
2 θ(cos2 θ−3 sin2 θ)+2 sin 2θ
and θ is given in figure 3. According to figure 1 and 3,
The graph between E(θ)
ω
introducing the fourth anisotropy has destroyed the smoothness of the curve. Sudden periodical overshooting can be observed in the graph. The separation between any adjacent
minimum and maximum is very small. The first and second minimums of overshooting can be observed at 740 and 160.50 . The curved parts with smooth maximums and
minimums in figure 1 have been flattened in figure 3.
If Ks is a variable,
78
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
E(θ)
ω
=
−80 − 3[10cos2 θ+5cos4 θ +
−
Ks
sin2θ]
ω
150[1 + cos2 θ]2 sin2 2θ
20 cos 2θ + 20 cos2 θ(cos2 θ − 3 sin2 θ) + 4 Kωs sin 2θ
The 3-D plot of E(θ)
versus Kωs and θ is given in figure 4. According to this graph,
ω
the energy varies rapidly with angle and stress after taking the effect of fourth order
anisotropy into account. Several minimums of energies can be observed at different angles
and stresses. The stresses corresponding to the different directions of preferred orientation
can be determined from this graph.
4. Conclusion:
Although the ultra-thin ferromagnetic film with two layers behaves as an oriented
film when anisotropy constants remain same for both layers, the energy of film with
three layers indicates periodic variation. Some sudden overshooting of energies can be
observed after taking second order perturbation into consideration, compared with energy
curves of oriented ferromagnetic ultra thin films described in one of our previous report.
The way of energy variation of all sc(001), fcc(001) and bcc(001) ferromagnetic ultra
thin films with second (or fourth) order anisotropy are exactly same. Easy and hard
directions of these sc(001), fcc(001) and bcc(001) ferromagnetic ultra thin films with
the effect of second order anisotropy only are 34.40 and 124.40 , respectively. Here these
angles have been measured with respect to the normal drawn to the film. The angle
between easy and hard directions is exactly 900 as expected. These angles corresponding
to easy and hard directions are higher than those of oriented ultra ferromagnetic thin
films studied in our early report. Introducing the fourth order anisotropy reduces the
smoothness of energy graphs, and indicates several minimums in 3-D graph. Although
(2)
(4)
these details were given for ωJ = 10, Dωm = 10, Kωs = 10 and Dωm = 5 values only, all these
(2)
(4)
simulations can be performed for any values of ωJ , Dωm , Kωs and Dωm or any other type of
ferromagnetic material than sc(001), fcc(001) and bcc(001). Since tedious simulations
have to be carried out after considering all the terms, above few terms have been taken
into account. According to graph 4, the film can be preferentially oriented in some certain
directions by applying some certain stresses.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
79
References
[1] K.D. Usadel and A. Hucht, Phys. Rev. B 66, 024419-1 (2002).
[2] P. Samarasekara and F.J. Cadieu: Jpn. J. Appl. Phys. 40, 3176 (2001).
[3] P. Samarasekara and F.J. Cadieu, Chinese J. Phys. 39(6), 635 (2001).
[4] David Lederman, Ricardo Ramirez and Miguel Kiwi, Phys. Rev. B(70), 18442 (2004).
[5] Martin J. Klein and Robert S. Smith, Phys. Rev. 81, 378 (1951).
[6] M. Dantziger, B. Glinsmann, S. Scheffler, B. Zimmermann and P.J. Jensen, Phys.
Rev. B(66), 094416 (2002).
[7] M. Bentaleb, N. El Aouad and M. Saber: Chinese J. Phys. 40(3), 307 (2002).
[8] D. Zhao, Feng Liu, D.L. Huber and M.G. Lagally: J. Appl. Phys. 91(5), 3150 (2002).
80
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
Fig. 1 Graph between
E(θ)
ω
and θs for
Ks
ω =10
with the effect of second order anisotropy
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
Fig. 2 3-D plot of
E(θ)
ω
versus
Ks
ω
and θs with the effect of second order anisotropy
81
82
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
Fig. 3 Graph between
anisotropy
E(θ)
ω
and θs for
Ks
ω =10
with the effect of second and fourth order
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 71–83
Fig. 4 3-D plot of
E(θ)
ω
versus
Ks
ω
and θs with the effect of second and fourth order anisotropy
83
EJTP 3, No. 11 (2006) 85–109
Electronic Journal of Theoretical Physics
Frameable Processes with Stochastic Dynamics
Enrico Capobianco
∗
Questlab, Research and Statistics
Mestre-Ve, Italy
Received 14 February 2006 , Accepted 5 April 2006, Published 25 June 2006
Abstract: A crucial goal in many experimental fields and applications is achieving sparse
signal approximations for the unknown signals or functions under investigation. This fact allows
to deal with few significant structures for reconstructing signals from noisy measurements or
recovering functions from indirect observations. We describe and implement approximation and
smoothing procedures for volatility processes that can be represented by frames, particularly
wavelet frames, and pursue these goals by using dictionaries of functions with adaptive degree
of approximation power. Volatility is unobservable and underlying the realizations of stochastic
processes that are non-i.i.d., covariance non-stationary, self-similar and non-Gaussian; thus,
its features result successfully detected and its dynamics well approximated only in limited
time ranges and for clusters of bounded variability. Both jumps and switching regimes are
usually observed though, suggesting that either oversmoothing or de-volatilization may easily
occur when using standard and non-adaptive volatility models. Our methodological proposal
combines wavelet-based frame decompositions with blind source separation techniques, and uses
greedy de-noisers and feature learners.
c Electronic Journal of Theoretical Physics. All rights reserved.
Keywords: Frames and Wavelets, Blind Source Separation, Sparse Approximation, Greedy
Algorithm, Volatility Smoothing
PACS (2006): 02.50.Fz, 05.40.-a,02.50.-r, 02.50.Ey
1.
Introduction and Volatility Background
Volatility processes have been for almost two decades a cross-disciplinary research
subject and have suggested many challenging problems to financial mathematicians and
statisticians. The main interest of this work too is in volatility, particularly the approximation of its dynamics and its statistical modelling. It is useful to start from the following
general classification.
∗
The author carried out most of this work at CWI, Amsterdam (NL).
86
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
Volatility can be specified as conditionally dependent on past squared returns and own
lags, thus representing the class of Generalized Autoregressive Conditionally Heteroscedastic (GARCH) processes [30, 7], or as a stochastically indipendent process, characterized
by markovian structure and a noise source independent from the disturbance term in the
conditional mean equation, thus referring to Stochastic Volatility (SV) processes [33].
The main stylized facts about volatility models appear from many empirical studies
[47], and among them the most important are (1) heavy-tailed (leptokurtic) marginal
return distributions; (2) volatility clustering, and thus tail dependence; (3) second-order
dependence, visible in absolute and squared transformed returns; (4) long memory and
covariance non-stationarity.
The GARCH and SV classes are often selected according to different goals, since they
refer to different information sets. In the first model class there is observation-driven
2
2
+ . . . + αp yt−p
(i.e., ARCH(p)
information, with yt | Yt−1 ∼ N (0, σt2 ), for σt2 = α0 + α1 yt−1
case); thus, the information set Ft for the index retun process is formed by past squared
observations up to time t-p, i.e., σ{Ys2 : s ≤ t − p}.
In the other class of models this is not possible, since they are driven by parameters and include both observable and unobservable variables; the index return process
is distributed according to yt | ht ∼ N (0, exp(ht )), and the volatility is specified as
ht = γ0 + γ1 ht−1 + ηt , ηt ∼ N ID(0, ση2 ). With ηt Gaussian, ht is autoregressive of orγ0
der one, and covariance stationarity follows if | γ1 |< 1. Then μh = E(ht ) = 1−γ
,
1
σ2
E(y 4 )
σh2 = var(ht ) = 1−γη 1 and the kurtosis is (σ 2t)2 = 3exp(σh2 ) ≥ 3, resulting in fatter tails
y
than the Gaussian ones.
Non-parametric and semi-parametric approaches have been introduced with the aim
of relaxing assumptions (see for instance the seminal work of [31]) on the error probability
density of regression models, or on the unobservable volatility function itself. Usually one
may end up with iteratively smoothing the unspecified function, as in log-transformed
multiplicative models [34] where backfitting estimation procedures [35] are adopted.
In these cases, the flexibility allowed so to account for the unrestricted structure in
the volatility function may be not sufficient when dealing with stochastic volatility, for
then the consistency of estimation procedures may fail. This fact occurs when a certain
orthogonality is built in the model, which relies on the statistical independence between
the information sets to which the volatility and the conditional mean dynamics refer.
The procedures that we suggest and implement are recursive in nature, being this an
important condition for building effective stochastic dynamics models, and are designed
for operating under both orthogonal and non-orthogonal conditions. The reference framework is that of non-orthonormal and redundant frame-based systems, together with that
of orthonormal systems built on atomic dictionaries o approximating functions.
It has been shown in many studies (see also [21] for an excellent review) that from such
systems, non-linear estimators can be built and result effective and computationally fast.
For inhomogeneous function classes they result superior to linear estimators employed by
backfitting methods with some kind of smoother. We illustrate how sparsified frame-based
volatility models may be investigated by a novel combination of greedy approximation and
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
87
space dimensionality reduction techniques that lead to near-optimal volatility smoothing
and feature detection.
The paper is organized as follows: Section 2 reviews frames, while Section 3 and 4
illustrate, respectively, multiresolution and greedy approximation techniques. Section 5
suggests an estimation technique based on blind source separation analysis and reports
experiments on modeling and smoothing volatility from a stock index return series. The
conclusions are in Section 6.
2.
Frames
A general way of approximating families of functions belonging to general spaces
refers to frames [29, 20], which represent redundant sets of approximating vectors. Throughout the paper, volatility processes will be considered to be non-stationary or piecewise
stationary, as in local stationary processes [19, 48]; thus, their realizations belong to spaces
that allow for spatial inhomogeneity and a certain complexity of dependence structure.
Together with the variance positivity constraint, a local boundedness condition for
the volatility paths is also assumed; this allows for preventing the volatility function from
explosive behavior. We thus consider frameable volatility functions those ones that can
be represented by frames or similarly derived systems.
Frame components are not linearly independent, but despite this aspect which seems
to penalize them from a computational standpoint, there are advantages in using frames
since they lead to numerically stable and robust-to-noise reconstruction algorithms while
also allowing for increased feature detection power, due to their flexibility.
A formal specification of frames requires the presence of a system {γk }M
k=1 and bounds
A and B such that:
2
A || x || ≤
M
k=1
|< x, γk >|2 ≤ B || x ||2
(1)
∀x ∈ RN .
A frame operator associated to them is F such that:
[F x]M
k=1 =< x, γk >
(2)
Noting that the redundancy of the frame comes from M ≥ N , and it is measured
by the ratio M/N , the frame operator, when multiplied by its transpose F ∗ , shows two
properties:
• F ∗ F is invertible and A−1 and B −1 are its bounds;
• a dual frame {γ˜k }M
k=1 is defined such that:
γ˜k = (F ∗ F )−1 γk
(3)
where now the associated dual operator is:
F̃ = F (F ∗ F )−1
(4)
88
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
A tight frame is defined when A = B, while an orthonormal basis requires a value of
1 for the bounds; this allows for an improved reconstrution power, as in this case F ∗ F is
diagonal and col(F ) are orthogonal, thus suggesting that a basis can be formed. Another
aspect of interest is given by the reconstruction step, where an inverse or pseudoinverse
of the F operator is required.
The standard formula is delivered by:
F − = (F ∗ F )−1 F ∗
(5)
but it can be shown that F − = F˜∗ , which means that one choice for the pseudoinverse is
provided by the transpose of the operator associated to the dual frames, and this leads
to a reconstruction formula which depends on this last operator as follows:
x = F˜∗ F x =
M
< x, γk > γ˜k
(6)
k=1
This linear reconstruction leads with noisy signals to a change in the previous expansion so to include the effects of noise according to:
M
[ < x, γk > + ǫk ] γ˜k
(7)
k=1
given the noise ǫ.
Note that a generalized reproducing kernel [42] is found whenever we have a vector
xn ⊂ x (i.e., a subset of the frame coefficients [F x]M
k=1 =< x, γk >) for which we compute
the denoising projection:
xk < γ˜k , γk >
(8)
P xn =
k
Since for a sequence of frame coefficients it holds that x = P x, our previous relation
is valid for x too, due to the kernel < γ˜k , γk >. The mean square error (MSE) is found
as:
M
1
1
E || x − x̂ ||2 = σ 2
|| γ˜k ||2
N
N k=1
(9)
with σ 2 the variance of the noise term. Given the definition of the dual frames, and
replacing in the MSE formula, one may achieve some optimality results, as when an
orthogonal matrix can be used a sequence of uniform frames is obtained which asymptotically approaches a tight frame. And it is precisely for this class of frames that the
MSE results optimal, i.e., minimum.
The MSE computed with frames is of course related to sparsity and coherence measures; we might expect that more sparsity and coherence bring MSE closer to its minimum, due to the fact that the contribution coming from informative and/or correlated
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
89
structures dominates that related to the noise, and a better performance in terms of
minimizing errors is achieved by greedy approximation algorithms. These aspects will be
introduced afterwards, while now special cases are illustrated.
3.
Multiresolution Feature Learning
3.1 Wavelet Frames
Wavelet frames are constructed by sampling from the Continuous Wavelet Transform
(W T c ) [20] over time and scale coordinates. For f ∈ L2 (R) (with < ., . > the L2 inner
product) and given an analyzing (admissible) wavelet with its doubly-indexed generated
family, the W T c is:
− 12
c
W T (f )jk =< f, ψjk >=| j |
f (t)ψ(
t−k
)dt
j
(10)
The function f can be recovered from the following reconstruction formula:
f=
c−1
ψ
W T c (f )jk ψjk
djdk
j2
(11)
and this comes from the ”resolution of identity formula”:
2
∀f, g ∈ L (R),
c
dkdj
%
W T c (f )jk W
T (g)jk 2 = cψ < f, g >
j
(12)
Given a constant cψ < ∞ and integration ranging from −∞ to ∞, the W T c maps f into
an Hilbert space, and its image, say L2W T c (R), is a closed subspace and a Reproducing
Kernel Hilbert Space too, since from the resolution formula it is enough to replace g with
ψ so to get:
c
%
T (g)jk > = < ψj ′ k′ , ψjk >
K(j, k, j ′ , k ′ ) =< W
(13)
More generally, given a scaling function φ, such that its dilates and translates constitute orthonormal bases for all the Vj subspaces obtained as scaled versions of the
subspace V0 to which φ belongs, and given a mother wavelet ψ together with the terms
indicated with ψjk and generated by j−dilations and k−translations, such that ψjk (x) =
j
2 2 ψ(2j x − k), one obtains differences among approximations computed at successively
coarser resolution levels.
Thus, a so-called Multiresolution Analysis (MRA) [41, 20] is obtained, i.e. a sequence
¯ j∈Z Vj = L2 (R),
of closed subspaces satisfying . . . , V2 ⊂ V1 ⊂ V0 ⊂ V−1 ⊂ V−2 ⊂ . . ., with ∪
j
∩j∈Z Vj = {0} and the additional condition f ∈ Vj ⇐⇒ f (2 .) ∈ V0 . The last condition is
a necessary requirement for identifying the MRA, meaning that all the spaces are scaled
versions of a central space, V0 .
90
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
An MRA approximates L2 [0, 1] through Vj generated by orthonormal scaling functions
φjk , where k = 0, . . . , 2j − 1. These functions allow also for the sequence of 2j wavelets
ψjk , k = 0 . . . , 2j − 1 to represent an orthonormal basis of L2 [0, 1].
Signal decompositions with the MRA property have also near-optimal properties in
a quite wide range of inhomogeneous function spaces, Sobolev, Holder, for instance, and
in general all Besov and Triebel spaces [46].
Generally speaking, with a Discrete Wavelet Transform (DWT) a map f → w from
the signal domain to the wavelet coefficient domain is obtained; one applies, through a
bank of quadrature mirror filters, the transformation w = W f , so to get the coefficients for
high scales (high frequency information) and for low scales (low frequency information).
A sequence of smoothed signals and of details giving information at finer resolution
levels is found from the wavelet signal decomposition and may be used to represent a
signal expansion:
f (x) =
cj0 k φj0 k (x) +
j>j0
k
djk ψjk (x)
(14)
k
where φj0 k is associated with the corresponding coarse resolution coefficients cj0 k and
&
&
djk are the detail coefficients, i.e., cjk = f (x)φjk (x)dx and djk = f (x)ψjk (x)dx. In
short, the first term of the right hand side of (1) is the projection of f onto the coarse
approximating space Vj0 while the second term represents the cumulated details.
3.2 De-noising and De-correlation
In the wavelet-based representations of signals sparsity inspires strategies that eliminate redundant information, not distinguishable from noise; this can be done in the
wavelet coefficients domain, given the relation between true and empirical coefficients:
d˜jk = djk + ǫt
(15)
The wavelet shrinkage principle [24, 25, 26] applies a thresholding strategy which
yields de-noising of the observed data; it operates by shrinking wavelets coefficients to
zero so that a limited number of them will be considered for reconstructing the signal.
Given that a better reconstruction might be crucial for financial time series in order to
capture the underlying volatility structure and the hidden dependence, de-noising can be
useful for spatially heterogeneous signals.
Financial time series are realizations of non-stationary and non-Gaussian stochastic
processes; a reason why wavelet systems could be effectively used when dealing with these
signals, concerns the ability of wavelets to de-correlate along time and across scales.
The de-correlation effect of the wavelet coefficients is one of the main properties that
wavelet transforms bring into the analysis [37, 2]; thus, from a structure with long range
dependence (LRD), either short range dependence (SRD) or de-correlation are found,
when looking at wavelet coefficients at each scale 2j or resolution level j.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
91
Given a probability space (Ω, F, P), consider a stochastic process and one of its realizations observed with strong dependence structure; following [1], and with a slight
variation in the notation compared to before due to the frequency (scale) characterization, let E[dx (j, k)] = 0 and the variance be:
2
E[dx (j, k) ] = Γx (v)2j | Ψ0 (2j v) |2 dv
(16)
and it represents a measure of the power spectrum Γx (.) at frequencies vj = 2−j v0 , with
Ψ0 the Fourier Transform of ψ0 .
Given Γx (v) ∼ cf | v |−α and v → 0, then:
E[dx (j, k)2 ] ∼ 2jα cf C(α, ψ0 )
for j → ∞, and
C(α, ψ0 ) =
| v |−α | Ψ0 (v) |2 dv
(17)
(18)
for α ∈ (0, 1).
One can then look at the variance law as follows:
var(dx (j, k)) ≈ 2jα
(19)
for j → ∞.
The decay of the covariance function is much faster in the wavelet expansion coefficients domain than in the domain originated by long memory processes. The covariance
function of the wavelet coefficients is controlled by M , the number of vanishing moments;
when they are present in sufficiently high number they lead to high compression power.
The sequence of detail signals or wavelet expansion coefficients is a stationary process if the number of vanishing moments M satisfies the constraint that the variance of
dx (j, k) shows scaling behaviour in a range of cut-off values j1 ≤ j ≤ j2 which has to be
determined. The sequence no longer shows LRD but only SRD when M ≥ α2 , and the
higher M the shorter the correlation left, due to:
E[dx (j, k)dx (j, k ′ )] ≈ | k − k ′ |α−1−2M , f or | k − k ′ | → ∞
(20)
These assumptions don’t rely on a Gaussian signal, and could be further idealized by
assuming E[dx (j, k)dx (j, k ′ )] = 0 for (j, k) = (j, k ′ ).
Thus, with an LRD process, the effect of the wavelet transform is clear, bringing
back decorrelation or small SRD due to the control of non-stationarity and dependence
through the M parameter. Across scales, instead, a certain degree of independence is
obtained, so that the detail series might individually contribute to different information
content in terms of frequency.
92
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
With regard to non-stationarity, an almost natural condition of financial time series,
especially when measured at very high frequencies; wavelets stationarize the data when
they are observed in their transformed wavelet coefficients, and they enable this change
in a resolution-wise fashion. As suggested by [16] X(t) is a stationary process if and only
if it is stationary at all levels of resolutions:
Definition 1: k-stationarity
Stationarity at the kth level of resolution if ∀n ≥ 1, t1 , . . . , tn ∈ T and l ∈ Z holds when:
d
[X(t1 + 2−k l), . . . , X(tn + 2−k )] = [X(t1 , . . . , X(tn )]
(21)
3.3 Multiscale Decomposition
We face two problems, when approximating the volatility function and estimating the
model parameters involved: the role of smoothness and the presence of noise. Following [5], we might rely on quadratic information from the data, leading to non-negative
estimators of the following kind:
σ̃ 2 (t) =
αi ri2 (t)
(22)
i
for i αi = 0 and i αi2 = 1. This can just be an initial estimate for a more calibrated
and robust procedure, since it can be improved by estimators that better account for
smoothness and sparsity.
Consider the L2 wavelet decomposition for the volatility function, expressed this time
through inner products:
2
σW
(t) =
k
< σ 2 (t), φj0,k > φj0,k (t) +
j>j0
< σ 2 (t), ψj,k > ψj,k (t)
(23)
k
where a smooth part is combined with a cumulated sequence of details obtained at different scales.
We can then apply the same decomposition to σ̃ 2 , being σ2 unobservable, and obtain
a perturbed version of the previous approximation:
2
σ̂ 2 (t) = σ̃W
(t) + ǫ(t)
(24)
where ǫ = W ξ represents a wavelet transformed disturbance.
In order to sparsify the estimator, we can apply noise shrinkage techniques or use
function expansions in different bases or overcomplete dictionaries.
If instead of considering the L2 elements we consider other function classes more
suitable for inhomogeneous behavior we can build non-linear estimators for the volatility
function through wavelet-type families [32].
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
93
3.4 Overcomplete Representations
Function dictionaries are collections of parameterized waveforms or atoms [15] ; they are
available for approximating many classes of functions, formed directly from a particular
family or from merging two or more dictionary classes.
Particularly in the latter case an overcomplete dictionary is composed, with linear
combinations of elements that may represent remaining dictionary structures, thus originating a non-unique signal decomposition.
Wavelet packets (WP) [17] are one example of overcomplete representations; they
represent an extension of the wavelet transform and allows for better adaptation due to
an oscillation index f which delivers a richer combination of functions.
& +∞
Given the admissibility condition −∞ W0 (t)dt = 1, ∀(j, k) ∈ Z 2 , following [39] we
have:
− 12
2
∞
t
hi−2k Wf (t − i)
W2f ( − k) =
2
i=−∞
(25)
where f relates to the frequency and h to the low-pass impulse response of a quadrature
mirror filter. Also the following holds:
∞
1
t
2− 2 W2f +1 ( − k) =
gn−2k Wf (t − n)
2
n=−∞
(26)
where g is this time an high-pass impulse response.
For compactly supported wave-like functions Wf (t), finite impulse response filters of
a certain length L can be used, and by P -partitioning in (j, f )-dependent intervals Ij,f
one finds an orthonormal basis of L2 (R), i.e., a wavelet packet:
j
{2− 2 Wf (2−j t − k), k ∈ Z, (j, f ) | Ij,f ∈ P }
(27)
One thus obtains a better domain compared to simple wavelets for selecting a basis to
represent the signal and can always select an orthogonal wavelet transform by changing
the partition P and defining w0 = φ(t) and Wf = ψ, from the so-called WP transform
(WPT).
Correspondingly, Cosine Packets (CP) and the related transform (CPT) suggest optimal bases in terms of compression power and sparsity [27], and optimal bases for dealing
with non-stationary processes with time-varying covariance operators [44].
The building blocks in CP are localized cosine functions, i.e., localized in time and
forming smooth basis functions, and they form almost eigenvectors for certain classes
of non-stationary processes, and thus almost diagonal operators for approximating the
covariance function.
The CPT has an advantage over the classic Discrete Cosine Transform (DCT); the
latter defines an orthogonal transformation and thus maps a signal from the time to the
94
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
frequency domain, but it is not localized in time and thus is not able to adapt well to
non-stationary signals.
Depending on the taper functions we select, the cosine packets decay to zero within the
interval where they are defined and in general determine functions adapted to overcome
the limitations of DCT.
The DCT-II transform is defined as:
n−1
2
(2i + 1)kπ
sk
)
(28)
gk =
fi+1 cos(
n i=0
2n
for k = 0, 1, . . . , n − 1, and scale factor sk resulting 1 if k = 0 or n, and
An inverse DCT is instead given by:
n−1
2
(2i + 1)kπ
)
fi+1 =
gk sk cos(
n k=0
2n
√1
2
if k = 0 or n.
(29)
for i = 0, 1, . . . , n − 1.
4.
Greedy Algorithms
Optimal algorithms often require adaptive signal approximation techniques based on
sparse representations. Sparsity refers to the possibility of considering only few elements
of a dictionary of approximating functions selected among a redundant set. In this way,
by restricting the search to a sub-set of the original atomic dictionary, one may combine
fast convergence through computing less inner products.
The MP algorithm [43] is a good example, and it has been successfully implemented in
many studies for its simple structure and effectiveness. A signal is decomposed as a sum
of atomic waveforms, taken from families such as Gabor functions, Gaussians, wavelets,
wavelet and cosine packets, among others. We focus on the WP and CP Tables, whose
signal representations are given by:
W P (t) = jf k wj,f,k Wj,f,k (t) + resn (t)
and
CP (t) = jf k cj,f,k Cj,f,k (t) + resn (t)
In summary, the MP algorithm approximates a function with a sum of n elements,
called atoms or atomic waveforms, which are indicated with Hγi and indexed by a dictionary Γ of functions whose form should ideally adapt to the characteristics of the signal
at hand.
The MP decomposition exists in orthogonal or redundant version and refers to a
greedy algorithm which at successive steps decomposes the residual term left from a
projection of the signal onto the elements of a selected dictionary, in the direction of that
one allowing for the best fit.
At each time step the following decomposition is computed, yielding the coefficients
hi which represent the projections, and the residual component, which will be then reexamined and in case iteratively re-decomposed according to:
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
f (t) =
n
hi Hγi (t) + resn (t)
95
(30)
i=1
and
(1)
(2)
(3)
following the procedure:
inizialize with res0 (t) = f (t), at i=1;
&
compute at each atom Hγ the projection μγ,i = resi−1 (t)Hγ (t)dt;
find in the dictionary the index with the maximum projection,
γi = argminγ∈Γ || resi−1 (t) − μγ,i Hγ (t) ||,
which equals from the energy conservation equation argmaxγ∈Γ | μγ,i |;
(4) with the nth MP coefficient hn (or μγn ,n ) and atom Hγn the computation of the
updated nth residual is given by:
resn (t) = resn−1 (t) − hn Hγn (t);
(5) repeat the procedure until i ≤ n.
With H representing an Hilbert Space, the function f ∈ H can thus be decomposed
as f =< f, gγ0 > gγ0 + Rf , with f approximated in the gγ0 direction, orthogonal to Rf ,
such that f 2 =|< f, gγ0 >|2 +Rf 2 .
Thus, to minimize the Rf term requires a choice of gγ0 in the dictionary such that
the inner product term is maximized (up to a certain optimality factor. The choice of
these atoms from the D dictionary occurs by choosing an index γ0 based on a certain
choice function conditioned on a set of indexes Γ0 ∈ Γ.
The main aspect of interest for the computational learning power of the MP algorithm has appeared in our study like in many others, and refers to how it is capable of
dealing efficiently with the so-called coherent structures compared to the dictionary noise
components [22].
5.
Non-parametric Estimation
5.1 Data
Figure 1 reports one-day 1m returns compared to the 5m aggregated values, together
with their absolute and squared transforms. We deal with stock returns observed at very
high frequencies from the Nikkei 225 composite index; the data were collected minute by
minute and refer to a certain market activity year, 1990.
We have approximately 35,000 observed values, covering intra-day market activity
and excluding holidays and weekends; we have transformed the prices in financial returns
by taking, as usual, the logarithms of ratios of consecutive time point prices.
We have also generated a temporally aggregated (every five minutes) series, which
basically smooths the original series, at the price of losing high frequency content. More
than 7,000 5-minute observations remain available to conduct a compared analysis.
We note a self-similar behavior and the typical function shape conditioned to the
different intensity of intra-day activity hours according to usual market technical phases.
In defining self-similarity, addressed as the property of self-affinity by [45], from [6] we
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
0
50
100
150
200
250
0.04
0.03
0.02
0.01
0
A.
50
100
150
200
250
Time
50
100
150
200
250
C.
0.06
Time
0
10
20
30
Time
40
50
D.
0.0
0.0
-0.2
0.01
0.05
-0.1
0.02
0.10
0.03
0.0
0.15
0.04
0.05
0.20
0.2
0.1
0
B.
0.25
Time
0.0
-0.2
0.0
-0.1
0.05
0.0
0.10
0.1
0.15
0.2
0.20
96
0
10
20
30
Time
40
50
E.
0
10
20
30
Time
40
50
F.
Fig. 1 A) Raw 1m returns. B) Absolute 1m returns. C) Squared 1m returns. D) Raw 5m
returns. E) Absolute 5m returns. F) Squared 5m returns.
have the following definition:
Definition 2: self-similarity
Given α ∈ (0, 1), γ > 0, f : Rd → R and x̄ ∈ Rd , a local re-normalization operator
γ
family, Rα,x̄
can be constructed such that:
γ
Rα,x̄
f (x) =
1
[f (x̄ + γx) − f (x̄)]
γα
∀x ∈ Rd
(31)
Thus, for instance, a Gaussian process X defined on a probability space (Ω, F, P ), is a
d
γ
X = X], ∀γ ∈ R+ and ∀x̄ ∈ Rd .
self-similar process of degree α if [Rα,x̄
As a remark, working with just one long realization of the underlying return process
means accepting the limitations that necessarily follow with regard to asymptotic inference, but at the same time represents a de facto typical situation in non-experimental
contexts, where the suggested techniques might be used.
Dealing with non-stationarity is imposed by real circumstances; the observed series is
subject to regime changes and external factors whose impact on the dynamics of returns
and structure of volatility is undoubtedly relevant. The observed index return time series
and the underlying volatility process represent a challenging context for approximation
and estimation techniques.
5.2 Iterative Smoother
Instead of using very noisy squared return sequences, we may run the MP greedy algorithm directly over the returns and then look at the computed absolute and squared
residuals for investigating the volatility features.
The corresponding autocorrelation function delivers a diagnostic measure of the feature detection power of the MP algorithm run over WP and CP dictionaries. Thus, in
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
97
our experiments, a main issue is to control the behavior of the transformed MP residual
terms after n approximation steps.
We have examined (see Figure 2) wavelet and cosine packets, whose time-frequency
partitioning role is complementary. While cosine packets build a partition in the time
domain and then run over each interval further frequency segmentation, wavelet packets
work the other way around, by first segmenting along frequencies and then along time
time. Thus, the former face frequency inhomogeneous smoothness while the latter deal
with time inhomogeneity [28].
The within-block coefficients of the WP and CP structures describe the contribution
at each resolution level of both time and frequency components in representing the signal
features under a varying oscillation index varying from 0 to 2J − 1 (right-wise).
The WP Table presents sets of coefficients stored in sequency order according to an
increasing oscillation index; at level 0 the original signal is represented and at level 1
the two subsets w1,0 and w1,1 have scale 2, corresponding to c1 and d1 obtained with the
DWT. The CP table presents instead blocks ordered by time and the coefficients within
the blocks are ordered by frequency.
Level 0
Level 0
Level 1
Level 1
Level 2
Level 2
Level 3
Level 3
Level 4
Level 4
Level 5
Level 5
Level 6
Level 6
0
1000
2000
3000
4000
5000
6000
7000
A.
B.
Fig. 2 CP table (A) and WP table (B) with signal segmentation level-by-level.
The way these plots should be interpreted suggests that the low frequency information
in the signal is expected to be concentrated on the left side and the high frequency
information on the right side of the table. For the CP table, the high frequency part
of the signal is now expected on the left side, while the low frequency behavior appears
from the right side.
We investigate the performance of the MP algorithm when is applied on an ad hoc
restricted and selected dictionary, based on a certain range of most independently informative resolution levels. Details are introduced next.
98
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
5.3 Blind Source Separation
The goal of searching for statistically independent coordinates characterizing certain objects and signals, or otherwise for least dependent coordinates, due to a strong dependence
in the nature of the stochastic processes observed through the structure of the index series,
leads to Independent Component Analysis (ICA) [13, 18].
Combination these goals with the search for sparse signal representations suggests hybrid forms of Sparse Component Analysis (SCA) [40, 23, 49]. With SCA one attempts to
combine the advantages delivered by sparsity and independence of signal representations,
which transfer to better compression power and estimation in statistical minimax sense.
By assuming that the sensor outputs are indicated by xi , i = 1, . . . , n and represent
a combination of independent, non-Gaussian and unknown sources si , i = 1, . . . , m, a
non-linear system Y = f (X) could be approximated by a linear one AS, where X = AS.
Instead of computing f (X) one may now work for estimating the sources S together with
the m × m mixing matrix A, where usually m << n, with n the number of sensor signals,
but with m = n holding in many cases too.
Independent components can be efficiently computed by ad-hoc algorithms such as
joint approximate diagonalization of eigenmatrices for real signals (JadeR) [14]. For
Gaussian signals, the Independent Components are exactly the known Principal Components; with non-Gaussian signals ICA delivers superior performance, due to the fact that
it relies on high order statistical independence information.
The JadeR algorithm is the ICA algorithm that we have applied in our experiments
to deliver estimates for the separating or de-mixing matrix B = A−1 , and obtain the
Y = BX, such that (under a perfect separation) Y = BAS = S. As a matter of fact, the
solution holds approximately and up to permutation P and scaling D, i.e., Y = DP S.
The de-correlation and rotation steps which have to be implemented will deliver a set of
approximate m independent components.
5.4 Volatility Source Separation
We set the following system to represent a volatility process:
y t = A t xt + ǫ t
(32)
where the observed returns are indicated by yt , the mixing matrix At is to be estimated,
together with the sources or latent variables xt ; the noise ǫt is superimposed to the system
2
the volatility
dynamics, with an i.i.d. (0, σǫ,t ) distribution. We indicate with vt = σǫ,t
process.
Ideally the volatility sources have a sparse representation, represented through the
following system:
xt = Cjf t Φjf t + ηt
(33)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
99
The compression and decorrelation properties of wavelet transforms can be supported
by a more effective search for least dependent components through combined MRA and
ICA steps. These steps can be set to work together in an hybrid method, as described
below.
Since the sources are unobservable, estimating them and the mixing matrix is quite
complicated; we can either build an optimization system with a regularized objective
function through some smoothness priors, so to estimate the parameters involved, or we
can proceed more recursively in the mean square sense.
One route is going through the iterations of the MP algorithm, by processing the
observed returns with the WP and CP libraries, and by looking at:
Yt ≈ Pt Φt + ξt = At Ct Φt + ξt
(34)
where the noise is including an approximation error from the system equation and residual
measurement effects ǫt .
We start by considering the detail signals obtained through WP and CP transforms,
which refer to different degrees of resolution. Then, we combine an ICA step with the
MP algorithm operating on WP and CP tables; through such a joint search for sparsity
and statistical independence we are basically adopting an hybrid SCA solution.
We aim to optimize sparsity through the choice of ad hoc function dictionaries and
optimze the performance of non-linear thresholding estimators. Furthermore, we search
least dependent coordinates such that an almost diagonal covariance operator is achieved,
helping the interpretation of latent volatility features.
5.5 Mixing Matrix Estimates
In Table 1 below we report the two estimated mixing matrices A, where the observed
sensor signals are those computed at each resolution levels by the WP and the CP transforms. These signals are passed through the ICA algorithm for the extraction of ”m”
possible sources which we set equal to the number of sensors.
We look at the results of this table so to extract from each detail level an approximate
value indicating its contribution to the global signal features, independently from the
other levels. The highest values computed suggest what are the dominant independent
components on a scale-dependent basis, without identifying their specific nature or the
underlying economic factors, being them system dynamics or pure shocks.
From the WP estimated mixing matrix A we note a strong within-level factor always
dominating apart from levels 5 and 6, where a mutual cross-influence appears to dominate. From the CP extimated mixing matrix A things change substantially, since each
level depends mainly from out-of-level factors, i.e., components belonging to other resolution levels, and only negligibly influenced by within-level factors. ICA selects the finest
resolution levels of the WP Table, while for the CP Table it delivers a mix of components
which are not concentrated at the finest resolutions.
100
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
Resol. lev.
0
1
2
3
4
5
6
level 0
0.2218
0.0028
0.0085
0.0047
0.0023
0.0069
0.0085
level 1
0.0002
0.1951
−0.0013
0.0001
−0.0189
−0.0035
−0.0037
level 2
0.0068
0.0003
−0.167
0.0015
0.0007
0.0019
−0.001
level 3
0.0031
−0.0057
−0.0008
−0.1438
−0.0019
−0.0045
0.0059
level 4
0.0012
−0.0125
0.0017
0.0028
−0.1318
0.0117
0.0
level 5
0.0032
−0.0023
0.0014
−0.0045
0.0008
−0.0011
−0.1147
level 6
0.0023
−0.0009
−0.0018
0.0047
−0.0082
−0.121
0.0017
level 3
0.1868
−0.0008
−0.0038
−0.0114
−0.0057
−0.0031
0.006
level 4
−0.0022
0.1832
0.0011
−0.0002
−0.0191
−0.0083
0.0053
level 6
0.0014
0.0046
0.1748
−0.0021
0.0036
−0.0052
0.002
level 2
−0.0089
−0.0023
−0.0031
−0.1712
0.0031
0.0057
−0.0006
level 5
0.006
0.0142
−0.0059
0.002
0.1482
−0.0053
0.0035
level 0
0.0029
0.0062
0.008
0.0031
0.0021
0.1261
0.0033
level 1
0.0012
0.0033
0.0013
0.0013
0.0041
0.0039
−0.1204
WP-A
CP-A
Table 1 Weights of the estimated ICA mixing matrix A distributed across resolu-
tion levels for residual 5m series obtained in WP/CP tables.
The diagonalization achieved by ICA with WP delivers a resolution-wise ordered
sequence of informative coefficients, with the highest scale delivering the most informative
component, while for CP the order is not the same. Lower scale sequences are more
Gaussian and stationary, due to aggregation effects, and thus it is likely that a certain loss
of efficiency follows from the fact of missing finer detail structure. Conversely, higher scale
sequences, being locally more irregular series, might present better efficiency together with
faster convergence rates.
The coefficient sequences obtained with WP and CP transforms have resolution-wise
approximation power and thus perform more accurately, compared to the original series,
with regard to the sample path from which the N data are observed. Processes for which
a convergence rate N −1 is reached by pointwise continuous real valued functions are called
irregular path processes [8] and the rate is called super-optimal. For stationary Gaussian
processes with differentiable paths these rates are not achievable; similarly, lower scale
sequences, compared to higher scale ones, reflect this aspect.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
101
5.6 Realised-Integrated Volatility
Due to the fact that daily squared returns don’t help too much in forecasting the latent
volatility structure because of noise and of different dynamics, a measure has been suggested so to obtain a more accurate estimate of the volatility function with high frequency
observations.
The realised volatility function is obtained by σ̂ 2 (t) = Ti=1 ri2 (t), and thus fulfills the
scope of approximating the integral of the unobservable variable by averaging a certain
number of 1m or 5m intraday values ri2 (t) [3, 4].
&
It holds [38] that σ̂t2 → σt2 = σs2 ds, i.e., the integrated volatility is approximated
by the realised volatility obtained according to the quadratic variation (QV) principle
derived from the following definition:
Definition 3: pth variation
given Xt and the partition T = {t0 , t1 , . . . , tn } of [0, t], the pth variation of Xt over T is:
(p)
Vt (T ) =
n
k=1
| Xtk − Xtk−1 |p
(35)
(2)
If || T ||= max1≤k≤n | tk − tk−1 | goes to 0, then lim||T ||→0 Vt (T ) =< X >t , where
the limit is the QV of Xt . This implies, in turn, that a convergence in probability applies,
&1 2
2
i.e., nj=1 rn,t,j
→n→∞ 0 σt+τ
dτ , where the cumulative squared high frequency returns
are employed rather than the daily values, so to improve the volatility prediction power.
The QV principle and the realised-integrated volatility relation lead quite naturally
[36] towards a non-parametric regression model as a de facto reference setting. With a
normalized sampling index t = ni , i = 0, . . . , n, and with yi/n = n(x(i+1)/n − xi/n )2 indi& (i+1)/n 2
σ (x)s ds, from this latter expression one can recover
cating noisy estimates of n i/n
2
σ (xt ) and thus the whole procedure leads to a model like yt ≈ σ 2 (xt ) + ǫt .
Since both the regressors and the disturbances are not indipendent and identically
distributed random variables, the volatility estimator can be found through families of
non-parametric estimators where the search for more spatially homogeneous observational
points require a re-mapping of the original values to some more regular domain, i.e., a
ν-indexed grid such that σ̂n2 (x̄) = νi Ki,n (xν , x̄). Our next step is to identify efficient
and feasible solutions for smoothing volatility.
5.7 Experiments
For pointwise volatility estimation a possibility is to use a parametric model such as a
GARCH or SV, computed directly on the residuals obtained with the MP algorithm,
after a certain number of steps [10, 11, 12]. The focus is on detecting the time varying
volatility features. We thus show parametric estimates from a model designed for the
series of MP residues obtained after 100 and 500 iterations, i.e. by using 100 or 500
approximating structures.
102
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
We have adopted an MA(1)-GARCH(1,1) model, thus balancing serial correlation in
the returns, and have used a Student’s t as marginal distribution. The model should be
written with a conditional mean equation (including an intercept and the MA(1) term in
k) like:
yt = k + ξt σt , with σt =
2
ht = a1 ht−1 + b1 yt−1
$
ht
(36)
(37)
where yt | Ψt−1 ∼ i.i.d.(0, ht ), ξt ∼ N (0, 1), and given the set of past information Ψt−1 .
By the prediction error decomposition, the log-likelihood function for a sample y1 , . . . , yi
is given by:
li (Θ) = logLT (Θ) =
T
i=1
T
T
xi
1
loghi +
log g( 1/2 )
logp(yi | Ψi−1 ) = −
2 i=1
hi
i=1
(38)
where g(.) is the Student’s t distribution, an heavy tailed conditional distribution, given
by g(x) = c
1
v+1
x2
(1+ v−2
) 2
, where c =
Γ( v+1
)
2
1
(π(v−2)) 2 Γ( v2 )
.
−1
Note that x = yi hi 2 are the standardized residuals, and that the degrees of freedom
are estimated together with the other parameters, say Θ, in the model. We then check
diagnostic and statistical properties, reported in Table 2, where t-stat is calculated from
the estimated standard deviations; Deg. Fr. refers to the return Student’s t distribution;
LB stands for Ljung-Box statistics for estimated squared standardized residuals; MaxLik
is the estimated value of the likelihood function.
Parameters
(G)hf100w.res
(G)hf500w.res
(G)hf100c.res
(G)hf500c.res
0.30
0.19
0.33
0.21
(33.87)
(18.53)
(30.08)
(17.7)
ARCH
0.012
0.014
0.076
0.078
t-stat
(9.95)
(8.51)
(9.6)
(5.96)
GARCH
0.973
0.98
0.699
0.20
(341.75)
(387.8)
(25.89)
(1.895)
Deg.Fr.
2.88
5.29
2.84
5.16
LB
20.07
10.86
40.25
17.49
5327.87
7229.1
5176.29
6628.33
MA(1)
t-stat
t-stat
MaxLik
Table 2 GARCH (G) estimates for the residual series with 100 and 500 runs of MP with WP
and CP tables indicated by hf100/500w.res and hf100/500c.res.
We observe that in general 500 iteration residues seem to suggest better models, as far
as concerns likelihood estimated values and LB statistics. For CP and WP the estimates
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
103
of the parameters change, particularly for the GARCH one, resulting for CP smaller in
absolute terms and surprisingly not significant for the 500 iteration residues.
Some plots are now reported with regard to the quantiles with respect to the Studentt distributions of the residual sequences obtained by MP after 100 and 500 iterations,
together with the ACF of the squared residuals (see Figure 3). With 500 iterations the
distributional properties suggested by the QQ-plots improve, for a better alignment with
the ideal benchmark distributions.
Ideally, coherent structures should be removed and the algorithm should be stopped
when dictionary noise is encountered. We observe second order statistical properties since
features relate now to the volatility process characterizing the signal. When no structure
is found this fact has to be interpreted as the evidence that only pure volatility aspects
are left in the residual series, now clean of dependence, seasonal and non-stationary
components.
Our results and diagnostic plots indicate that volatility is smoother when estimated
from 500 MP iterations, compared to only 100. The possibilities are that it is either
approximating the true volatility or that the MP algorithm is overfitting. For the CP case
the same observation could be done, but due to the behavior of the GARCH parameter
estimates, we might conclude that more than possible oversmoothing, MP has produced
de-volatilization in the returns process.
Figure 4 is about the performance of the MP algorithm when examined through the
residues obtained at varying approximation power employed. For the case under study,
we consider the L2 and L1 errors, from respectively squared and absolute transformed
residual terms. We compare them with the number of MP approximating, possibly
coherent, structures employed, up to 500, which corresponds to the L0 norm of the
expansion coefficients, i.e., a measure of sparsity.
We note from the CP plots (C-D) that MP has a fast convergence; with both L2 and
L1 norms, the first minimal turning point is at 100 structures, while the second one is at
200 structures. For the L2 norm the successive decay is smooth, while for the L1 norm is
slightly steeper in approaching the new minimum at approximately 500.
From the WP plots (A-B) the limit of 100 structures is confirmed, but then convergence is lost. This confirms that by iterating 500 times the MP algorithm surely oversmooths the volatility function with WP approximating atoms, and instead of learning
its structure it yields de-volatilization in the CP case.
We thus can conclude that the number of 100 iterations represents a fairly good
benchmark iteration number of the MP algorithm for exploring the dynamics and learning
the features of stock index return volatility, while at the same time preventing from
possible numerical instability and overfitting.
6.
Conclusions
Financial time series are very complex structures and we show that in order to investigate their structure and detect volatility features it might be useful to adopt sparse
104
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
representations and independent component decompositions, together with greedy approximation techniques. An example is offered by ICA applied to wavelet and cosine
packets, and forming an hybrid form of SCA when combined with greedy feature learners. It appears from the experiments that the there is a clear improvement in the feature
detection power of the latent volatility structure underlying a stock index return series.
The resolution selection operated by ICA, concentrated on high scale signals for WP
function dictionaries, eliminates redundant information by keeping highly localized time
resolution power without simultaneously losing too much frequency resolution, due to the
fact that low scale information can be reproduced by averaging high scales.
When ICA is applied to a CP library, it doesn’t really build a sparse representation,
since the CP coordinates are already naturally endowed with that property; in terms of
decomposing the signal, the advantage of using a CP transform is thus in the inherent
diagonalization power with respect to the covariance operator.
The hybrid SCA method that we have designed for our application yields least dependent resolution levels which are used for calibrating the MP algorithm. The latter
achieves a better detection power for the dependence structure in the return series due
to the fact that the almost independent (or least dependent) coordinates along which the
iterations are run, allow the MP algorithm to be more orthogonalized, and thus more
efficient in retrieving the most coherent atomic structures.
Acknowledgements.
The author, while at CWI, was a recipient of the 2001/02 ERCIM Research Fellowship.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
105
References
[1] P. Abry, P. Flandrin, M.S. Taqqu, and D. Veitch, Wavelets for the analysis, estimation
and synthesis of scaling data, in ”Self-similar network traffic and performance
evaluation” (C. Park, and W. Willinger, Eds.), Wiley, New York, (2000) 39-88.
[2] P. Abry, D. Veitch, and P. Flandrin, Long range dependence: revisiting aggregation
with wavelets, Journal of Time Series Analysis, 19(3), (1998), 253-266.
[3] T. Andersen, T. Bollerslev, F.X. Diebold, and H. Ebens, The distribution of stock
return volatility, Jourbal of Financial Economics, 61, (2001), 43-76.
[4] T. Andersen, T. Bollerslev, F.X. Diebold, and P. Labys, The distribution of exchange
rate volatility, Journal of American Statistical Association, 96, (2001), 42-55.
[5] A. Antoniadis, and, and C. Lavergne, Variance Function Estimation in regression by
Wavelet Methods, in ”Wavelets and Statistics” (A. Antoniadis, and G. Oppenheim,
Eds.), Springer-Verlag, New York, (1995), 31-42.
[6] A. Benassi, Locally Self Similar Gaussian Processes, in ”Wavelets and Statistics” (A.
Antoniadis, and G. Oppenheim, Eds.), Springer-Verlag, New York, (1995), 43-54.
[7] T. Bollerslev, Generalized Autoregressive Conditional Heteroskedasticity, Journal of
Econometrics, 31, (1986), 307-327.
[8] D. Bosq, ”Nonparametric Statistics for Stochastic Processes. Estimation and
Prediction”. Springer-Verlag, New York 1998.
[9] A. Bruce, and H.V. Gao, ”S+Wavelets”, StaSci Division, MathSoft Inc, Seattle 1994.
[10] E. Capobianco, Statistical Analysis of Financial Volatility by Wavelet Shrinkage,
Methodology and Computing in Applied Probability, I(4), (1999), 423-443.
[11] E. Capobianco, Indipendent Multiresolution Component Analysis and Matching
Pursuit, Computational Statistics and Data Analysis, 42(3), (2003), 385-402.
[12] E. Capobianco, Frameable non-stationary processes and volatility applications, IMA
preprint 2011, Institute for Mathematics and its Applications, Minneapolis, MN, 2004.
[13] J. Cardoso, Source separation using higher order moments, Proceedings International
Conference on Acoustic, Speech and Signal Processing, (1989), 2109-2112.
[14] J. Cardoso, and A. Souloumiac, Blind beamforming for non-Gaussian signals, IEE
Proceedings F., 140(6), (1993), 771-774.
[15] S. Chen, D. Donoho, and M.A. Saunders, Atomic Decomposition by Basis Pursuit,
SIAM Review, 43(1), (2001), 129-159.
[16] B. Cheng, and H. Tong, K-stationarity and wavelets, Journal of Statistical Planning
and Inference, 68, (1998), 129-144.
[17] R.R. Coifman, Y. Meyer, and M.V. Wickerhauser, Wavelet Analysis and Signal
Processing, in ”Wavelets and their Applications” (M.B. Ruskai et. al., Eds), Jones
and Bartlett, Boston, (1992), 153-178.
[18] P. Comon, Independent Component Analysis - a new concept?, Signal Processing,
36(3), (1994), 287-314.
[19] R. Dahlhaus, Fitting time series models to nonstationary processes, The Annals of
Statistics, 25, (1997), 1-37.
106
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
[20] I. Daubechies, ”Ten Lectures on wavelets”, SIAM, Philadelphia 1992.
[21] R.A. DeVore, Nonlinear Approximation, Acta Numerica, (1998), 51-150.
[22] G. Davis, S. Mallat, and M. Avellaneda, Greedy adaptive approximations,
Constructive Approximation, 13(1), (1997), 57-98.
[23] D. Donoho, Sparse Components of Images and Optimal Atomic Decompositions, J.
Constructive Approximation, 17, (2001), 353-382.
[24] D. Donoho, and I.M. Johnstone, Ideal Spatial Adaptation via Wavelet Shrinkage,
Biometrika, 81, (1994), 425-455.
[25] D. Donoho, and I.M. Johnstone, Adapting to unknown smoothness via wavelet
shrinkage, Journal of American Statistical Association, 90, (1995), 1200-1224.
[26] D. Donoho, and I.M. Johnstone, Minimax Estimation via Wavelet Shrinkage, The
Annals of Statistics, 26, (1998), 879-921.
[27] D. Donoho, S. Mallat, and R. von Sachs, Estimating Covariances of Locally
Stationary Processes: Rates of Convergence of Best Basis Methods, Tech. Rep. 1998517, Dep. Statistics, Stanford University, Stanford, (1998).
[28] D. Donoho, M. Vetterli, R.A. DeVore, and I. Daubechies, Data Compression and
Harmonic Analysis, IEEE Transactions in Information Theory, 44(6), (1998), 24352476.
[29] R.J. Duffin and A.C. Schaeffer, A class of non-harmonic Fourier series, Transactions
of the American Mathenatical Society, 72, (1952), 341-366.
[30] R.F. Engle, Autoregressive Conditional Heteroscedastic models with estimates of the
variance of th UK inflation, Econometrica, 50, (1982), 987-1007.
[31] R.F. Engle, and G. Gonzales-Rivera, Semiparametric ARCH Models, Journal of
Business and Economic Statistics, 9, (1991), 345-359.
[32] H.Y. Gao, Wavelet shrinkage estimates for heteroscedastic regression models, Tech.
Rep., MathSoft Inc., Seattle, (1997).
[33] E. Ghysels, A.C. Harvey, and E. Renault, Stochastic Volatility, in ”Handbook of
Statistics” (G.S. Maddala and C.R. Rao, Eds.), vol. 14. Amsterdam: Elsevier Science,
(1996), 118-191.
[34] C.M. Hafner, Estimating high frequency foreign exchange rate volatility with
nonparametric ARCH models, Journal of Statistical Planning and Inference, 68, (1998),
247-269.
[35] T.J. Hastie, and R.J. Tibshirani, ”Generalized Additive Models”. Chapman & Hall,
London, 1990.
[36] M. Hoffmann, Lp estimation of te diffusion coefficient, Bernoulli, 5(3), (1999), 447481.
[37] I.M. Johnstone, and B.W. Silverman, Wavelet threshold estimators for data with
correlated noise, Journal Royal Statistical Socoety Ser. B., 59, (1997), 319-351.
[38] I. Karatzas, and E. Shreve, Brownian Motion and Stochastic Calculus, SpringerVerlag, New York, 1988.
[39] H. Krim, and J.C. Pesquet, On the statistics of Best Bases criteria, in ”Wavelets
and Statistics”, (A. Antoniadis, and G. Oppenheim, Eds.), Springer-Verlag, New York,
(1995), 193-207.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
107
[40] M.S. Lewicki, and T.J. Sejnowski, Learning Overcomplete Representations, Neural
Computation, 12(2), (2000), 337-365.
[41] S. Mallat, Multiresolution approximations and wavelet orthonormal bases of L2 (R),
Transactions of American Mathematical Society, 315, (1989), 69-87.
[42] S. Mallat, ”A wavelet tour of signal processing”, Academic Press, 1999.
[43] S. Mallat, and Z. Zhang, Matching Pursuit with time frequency dictionaries, IEEE
Transactions in Signal Processing, 41, (1993), 3397-3415.
[44] Mallat, S., Papanicolaou, G. and Z. Zhang, Adaptive Covariance Estimation of
Locally Stationary Processes, The Annals of Statistics, 26(1), (1998), 1-47.
[45] B.B. Mandelbrot, L. Calvet, and A. Fischer, The Multifractal model of asset returns,
Disc. Pap. 1164, Cowles Foundation, Yale University, New Haven, (1997).
[46] I. Meyer, ”Wavelets: algorithms and applications”, SIAM, Philadelphia, 1993.
[47] T. Mikosch, and C. Starica, Limit theory for the sample autpcorrelations and
extremes of a GARCH(1,1) process, The Annals of Statistics, 28(5), (2000), 1427-1451.
[48] M.H. Neumann and R. von Sachs, Wavelet Thresholding: beyond the Gaussian
I.I.D. situation, in ”Wavelets and Statistics”, (A. Antoniadis and G. Oppenheim Eds.),
Springer-Verlag, New York, (1995), 301-329.
[49] M. Zibulewsky, and B.A. Pearlmutter, Blind Source Separation by Sparse
Decomposition in a Signal Dictionary, Neural Computation, 13(4), (2001), 863-882.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
10
15
108
7040
7040
10
70387039
xres
-10
-5
-5
0
0
xres
5
5
7039
1
-20
-10
0
10
20
-10
10
-5
0
A.
Quantiles of t(2.88872294767274) distribution
5
10
Quantiles of t(5.2936730230749) distribution
7040
0
xres
-4
-5
-2
xres
0
2
5
4
6
7040
B.
-6
-10
2
-8
1
-30
-20
-10
0
10
20
Quantiles of t(2.84688052704445) distribution
-10
C.
-5
0
5
10
Quantiles of t(5.16731001818244) distribution
Series : resq500w
D.
ACF
0.0
0.0
0.2
0.2
0.4
0.4
ACF
0.6
0.6
0.8
0.8
1.0
1.0
Series : resq100w
30
2
1
0
2000
4000
6000
0
2000
Lag
4000
6000
Lag
E.
F.
Series : resq500c
ACF
0.0
0.0
0.2
0.2
0.4
0.4
ACF
0.6
0.6
0.8
0.8
1.0
1.0
Series : resq100c
0
2000
4000
6000
0
Lag
2000
4000
6000
Lag
G.
H.
Fig. 3 QQ-plots of standardized residuals vs quantiles of a Student’s t, for 100 (A,B)
and 500 (C,D) MP iterations run on respectively WP/CP Tables; correspondent
ACF of squared residuals for WP (E,F) and CP (G,H).
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 85–109
109
0.08
0.05
0.02
0
100
200
300
400
500
A.
0.3
0.2
0.1
0.0
0
100
200
300
400
500
B.
0.3
0.2
0.1
0.0
0
100
200
300
400
500
C.
0.5
0.3
0.1
0
100
200
300
400
500
D.
Fig. 4 L2 error vs number of approximating structures, for WP (A) and CP (C)
and L1 error vs L0 norm for WP (B) and CP (D).
EJTP 3, No. 11 (2006) 111–122
Electronic Journal of Theoretical Physics
Ab-initio Calculations for Forbidden M1/E2 Decay
Rates in Ti XIX ion
A. Farrag
∗
Physics Department,Faculty of Science,
Cairo University, Cairo, Egypt
Received 27 February 2006 , Accepted 5 April 2006, Published 25 June 2006
Abstract: The rates of the electric quadrupole E2 and magnetic dipole M1 forbidden
transitions in the ground configuration and some excited configurations of the Ti XIX ion have
been calculated. The multiconfiguration Hartree - Fock (MCHF) method has been used. The
relativistic corrections are included in the Breit - Pauli approximation. A detailed comparison
of the present theoretical results with previous calculations and the available data in literature
is presented.
c Electronic Journal of Theoretical Physics. All rights reserved.
Keywords: Ab-initio Calculations, Multiconfiguration Hartree - Fock, Ti XIX Ion, Breit -Pauli
Approximation
PACS (2006): 31.15.Ar, 31.15.Ne, 21.10.Ky
1.
Introduction
Recently there has been considerable interest in the forbidden electric quadrupole
and magnetic dipole transitions in ionic systems. These forbidden lines are spectral
lines which arise from transitions that are forbidden by selection rules for the electric
dipole E1 transitions. The M1 and E2 transitions have been found to be very useful in
the diagnostics of astrophysical and laboratory plasmas [1,2] and are necessary for the
interpretation of the observed line intensities and may contribute to the width and shape
of the spectral line associated with nearby allowed transitions [3,4]. These transitions
often represent optically thin lines i.e isolated lines with low transition probabilities, due
to low self-absorption effects in the plasma. Moreover, the M1and E2 lines frequently
occur at longer wavelengths, if compared with the electric-dipole allowed transitions, since
they may connect the levels within the same electron configuration. Thus, by having lines
∗
azza farrag@hotmail .com
112
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 111–122
in the visible or near uv range, high resolution techniques can be employed in order to
obtain detailed information including the shape of the lines.
Traditionally, the interest in these lines has come from the astrophysicists and has
been restricted to light elements. More recently, however, such lines have been observed
in fusion plasmas. In particular, Tokamak plasmas have been found to be a rich source
of forbidden lines in heavy elements, heavier than those observed in the stars; these lines
have been used for diagnostic purposes in Tokamak machines.
A current interest in the forbidden transitions of the metallic impurity elements in
Tokamak discharges results from their use for certain diagnostic purposes. The intensities
of these transitions allow measurements of the concentrations of impurity ions which
originate in the high temperature interior of the discharge.
Many calculations of the dipole oscillator strengths have been made for the Be- like
ions [5, 6], and dipole transition rates are now known quite accurately for many transitions
of he Be-isoelectronic sequence. Some studies of the forbidden transition rates have
appeared in literature [7, 8, 13, 15]. Given the importance of forbidden transition for the
Be-like –Ti ion, it is important to present accurate data for these transition rates.
Forbidden transitions for highly ionized Be like ions are expected to play an increasingly important role in fusion plasma diagnostics in future, but until recently, very little
was known about these transitions. This paper intends to fill this gap.
The theory is presented in section 2. Section 3 displays the results of the present calculations together with a comparison of these results with the available data in literature;
finally a conclusion is given in the last section 4.
2.
Theoretical Method and Computational Procedure
In this study, the atomic state wave functions (ASFs) have been generated by
the widely used atomic structure package method, the multiconfiguration Hartree-Fock
MCHF method [9].
The theoretical approach employed is sketched below. A more detailed description can
be found in [10]. The relativistic effect is included as a correction to the non relativistic
Hamiltonian by adding the Breit -Pauli operator.
2.1 The Breit - Pauli Wave Functions
In the multiconfiguration approximation, the Breit -Pauli wave functions for a state
labeled γJMJ , where γrepresents the configuration and the set of quantum numbers
required to specify the state, are expanded in terms of configuration state functions
(CSFs)
ψ(γJMJ ) =
M
i=1
ci φ(γi Li Si JMJ )
(1)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 111–122
113
The configuration state functions φ are antisymmetrized linear combinations of the products of the spin-orbitals functions
LML SMS |LSJMJ φ(LML SMS )
(2)
φ(γLSJMJ ) =
ML MS
φnℓmℓ ms = 1r Pnℓ (r)Yℓmℓ (θ, ϕ)ξms (σ)
(3)
In the present method, the radial functions building the CSFs are taken from a nonrelativistic MCHF code and only the expansion coefficients are optimized, this leads to
the matrix eigen value problem
Hψ = Eψ
(4)
where H is the Hamiltonian matrix with elements
Hij = γi Li Si JMJ |HBP |γJ Lj Sj JMJ
(5)
The problem of finding the eigenvalues and eigenfunctions of the Breit –Pauli Hamiltonian
can be reduced to the evaluation of matrix elements between LSJ coupled CSFs and a
matrix diagonalization for each J value.
2.2 The Breit - Pauli Hamiltonian
The Breit - Pauli Hamiltonian is a first order perturbation correction to the non –
relativistic Hamiltonian.
The Breit-Pauli Hamiltonian is given by:
HBP = HN R + HRS + HF S
where HN R is the ordinary non-relativistic many electron Hamiltonian .
N
N
1 2 Z
i
+
HN R =
− ∇i −
2
ri
r
i>j ij
i=1
(6)
(7)
HRS is the relativistic shift operator which commutes with L and S and can be written
as
HRS = HM C + HD1 + HD2 + HOO + HSSC
(8)
where HM C is the mass correction term
HM C
N
1 2 † 2
(∇i ) ∇i
=− 2
α i=1
(9)
and HD1 and HD2 are the one and two – body Darwin terms
HD1 = −
N
α2 Z 2 1
∇( ) ,
8 i=1 i ri
(10)
114
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 111–122
HD2
HOO is the orbit-orbit term
HOO
N
α2 2 1
=
∇( ) .
4 i<j i ri
N
α2 Pi .Pj rij (rij .Pi )Pj
=−
+
3
2 i<j
rij
rij
(11)
(12)
and finally HSSC is the spin-spin contact term
N
8πα2
=−
(si .sj )δ(ri .rj )
3 i<j
HSSC
(13)
The fine - structure operator HF S describes the interactions between the spin and orbital
angular momenta of the electrons, and does not commute with L and S but only with
the total angular momemtum J=L+S.
The fine –structure operator HF S consists of three terms:
HF S = HSO + HSOO + HSS
(14)
where HSO is the spin-orbit term
HSO
N
α2 Z 1
=
ℓi .si
2 i=1 ri3
(15)
HSOO is the spin-other-orbit term
HSOO
N
α2 rij × Pi
(si + 2sj )
=−
3
2 i<j
rij
(16)
and HSS is spin –spin term
HSS
N
(si .rij )(sj .rij )
1
=α
si .sj − 3
3
2
r
rij
ij
i<j
2
(17)
2.3 Electric Quadrupole Decay Rates: E2
The transition probabilities are calculated using the following formula
AE2 =
1.120 × 1018
SE2
gi λ5
in(sec−1 )
(18)
where λ is the transition wavelength in Å, gi is the statistical weight of the initial state
and SE2 is the line strength given by
SE2 = |Ψf |OE2 | Ψi |2
(19)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 111–122
115
OE2 is the electric quadrupole operator
OE2 =
ri2 Cq(2) (i)
(20)
i
Because this operator has even parity, the electric quadrupole transitions involve no
change in parity . The selection rules on the total angular momentum are
ΔJ=0, ±1, ±2 J+J’≥ 2
where J and J’ are the total angular momentum for the initial and final states, respectively.
2.4 Magnetic Dipole Decay Rates: M1
The transition probabilities are calculated using the following formula
AM 1 =
2.697 × 1013
SM 1
gλ3
(sec−1 )
(21)
where λ is the transition wavelength in Å, gi is the statistical weight of the initial
state and SM 1 is the line strength given by:
SM 1 = |ψf |OM 1 | ψi |2
3.
Results and Discussion
The MCHF method has been used with the Breit-Pauli relativistic corrections to
calculate the transition energies, wavelengths and the electric quadrupole and magnetic
dipole transition rates for the Ti XIX ion of the Be-isoelectronic sequence and that among
each of the configurations 2s2p, 2p2 and the transitions between 2s2 and 2p2 .
It should be mentioned here that there are two alternative forms of the electric dipole
matrix element, namely the length form and the velocity form. The two forms are equivalent for exact solutions of the Hamiltonian. It is customary to compute both forms and
use the agreement between the two results as one of the quality criteria for the calculation. Moreover that the agreement between the length and velocity form is a necessary
condition for accurate wavefunctions. These wavefunctions have been used to calculate
the magnetic dipole and the electric quadrupole transitions between the n=2-2 levels.
It should be noticed that the dominant correlation effect is in the n=2 intrashell
correlation using the three configuration basis composed of 2s2 , 2s2p and 2p2 . To achieve
a better agreement between experiment and theory for the transition energies between the
n=2-2 levels, additional correlation functions have to be included in the basis set including
most of the n=3 configurations. No attempt has been made to include configurations with
n=4 and n=5 in the basis functions in the present work.
In these ab-initio calculations, the wavefunctions with configuration interactions (CI)
between the levels with the same parity and the same total angular momentum have been
obtained. The even configurations considered are (2s2 , 2p2 , 2s3s , 2s3d , 2p3p and 3p2 )
116
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 111–122
while the odd configurations are (2s2p ,2s3p ,2p3s ,2p3d , 3s3p and 3p3d), outside the1s2
core.
Tables (I), (II) and (III) present the transition energies ΔE, the wavelengths (λ),
the electric quadrupole transitions rates A(E2) and the magnetic dipole transitions rates
A(M1) for the transitions among the 2s2p, 2p2 and the transition from 2s2 to 2p2 configurations. The present calculated values are referred to (a) while (b) refers to Glass results
[11] which presented very sophisticated relativistic intermediate-coupling calculations of
the wavefunctions and (c) stands for the calculations of Bhatia et al [12] which also used
relativistic intermediate coupling wavefunctions, while (d) stands for the semiempirical
wavelengths of Edlén [14], (e) gives the NIST values [16] and (f) are the values for the
multiconfiguration Dirac –Fock (MCDF) of Froese-Fischer [15].
The NIST values are followed by a capital letter denoting the uncertainties in the
atomic transition probability data which is C for uncertainties within 25%, D for uncertainties within 50% and E for uncertainties greater than 50%.
3.1 M1/E2 Transitions Among the 2s2p Levels
The 2s2p configuration of Ti XIX has 4 fine structure levels 1 P1 and 3 P0,1,2 . These levels
are separated into two groups by the different multiplets. The possible E2 transitions
among these 2s2p levels include the 2s2p (1 P1− 3 P1,2 ), 2s2p(3 P1− 3 P2 ) and (3 P0− 3 P2 ) lines,
while for the possible M1 transitions they include the 2s2p (1 P1− 3 P0,1,2 ), 2s2p(3 P0− 3 P1 )
and 2s2p (3 P1− 3 P2 ) lines.
As seen from table I, the calculated transition energies from the singlet state to
the triplet states are in good agreement with the calculations (b) of Glass and (f) of
Froese- Fischer, both calculations were done performing purely relativistic calculations,
the relative percentage is a maximum of 1% with Glass and NIST values and up to 0.2%
with Froese –Fischer, the discrepancy with the calculation of Bathia et al. is slightly large,
it is up to 3.4% relative percentage difference, while with the semi –empirical values of
Edlén, the agreement is good and is less than 1% relative percentage difference.
For the triplet-triplet transitions, the (3 P0− 3 P1 ) line show the biggest discrepancy,
the difference is 787.5 cm−1 which is 4.8% relative percentage difference as compared to
Glass and 670.13 cm−1 or 4% as compared to Froese–Fischer. For all the other presented
data, the relative percentage difference with the other calculations does not exceed 0.2%.
The electric quadrupole E2 and the magnetic dipole M1 transition probabilities show
good agreement with the calculations of Froese-Fischer, it is 3.4% maximum of the relative
percentage difference and 13.7% maximum of the relative percentage difference with all
other calculations.
3.2 M1/E2 transitions among the 2p2 levels
The 2p2 configuration of Ti XIX has 5 fine structure levels 3 P0,1,2 , 1 D2 , 1 S0 , which are
separated into three groups by the different multiplets. The possible E2 transitions among
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 111–122
117
the 2p2 levels include seven transitions, while the M1 transitions include five transitions
which are presented in table II.
Our calculated values for the transitions between 2p2 levels are close to the semi
empirical values of Edlén and the (MCDF) calculations of Froese-Fischer. They have up
to 1% relative percentage with Edlén and up to 0.2% with the calculations of FroeseFischer. The present calculated transition energies among the triplet states show a large
difference with that of Glass and Bathia et al. for the (3 P1 −3 P2 ) transition, the difference
is 710 cm−1 or 2.5% with Glass and 2268 cm−1 or 7.5% with Bathia et al., but it is 116.2
cm−1 or 0.4% with Edlén and 41.04 cm−1 or 0.14% with Froese-Fischer. The value of the
E2 decay is weak among the triplet states, in these transitions the M1 decay is about
four to five orders of magnitude faster than the E2 decay.
In the singlet D and triplet P transitions, our calculated transition energies show
the best agreement with the (MCDF) calculations of Froese-Fischer, it is 156.8 cm−1 or
0.1%, 78.8 cm−1 or 0.07% and 119.9 cm−1 or 0.13 % for the transitions (1 D2 −3 P0,1,2 ),
respectively.
The transitions to the singlet S show a good agreement with all the calculations as well
as with the semi empirical values of Edlén and the NIST values, it is 1% of the relative
percentage difference for all transitions except for the (3 P1− 1 S0 ) there is a difference of
7852.1 cm−1 or 2.5% of the relative percentage difference and for (3 P2− 1 S0 ) a difference
of 5577.7 cm−1 or 2% with the calculations of Bathia et al.. For the transition rates,
the A(E2) agree to within 15% and within 9% for the A(M1) of the relative percentage
difference with all other calculations.
3.3 M1/E2 transitions between the 2s2 − 2p2 levels
In table III the even-even transitions between 2s2 − 2p2 levels arise in two transitions
(1 S0 −3 P2 ) and (1 S0 −1 D2 ) for the electric quadrupole, they have prominent transition
probabilities A(E2), and have one transition (1 S0 −3 P1 ) for the magnetic dipole. In these
transitions the relative percentage difference for the transition energies agrees with all
other calculations and with the semi empirical values of Edlén as well as with the NIST
values, there is less than 9% as maximum of relative percentage difference.
The quadrupole transition rates present a relative percentage of 50 % and 45 %
with the values of Glass and Bathia et al., respectively but show a better agreement
with that of Froese-Fischer, it is 16% with the same transition (1 S0 −1 D2 ) and for the
(1 S0 −3 P2 ) transition the relative percentage difference is 39%, 32% and 40% with that
of Glass, Bathia et al and the NIST, respectively, but it is close to the calculation of
Froese-Fischer, it is 7.5% of a relative percentage difference.
The magnetic dipole decay rates for (1 S0 −3 P1 ) transition show a good agreement
between all calculations, it has a difference of 4.65% with that of Glass, 8.7% with Bathia
et al, 6.8% with the NIST values and 1.75% with Froese-Fischer.
118
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 111–122
Conclusion
In summary, the multi-configuration Hartree-Fock (MCHF) method has been used
to study the E2 and M1 transitions among some configurations of TiXIX, the inclusion of
the configuration interaction and relativistic corrections by using the Breit-Pauli approximation in the calculation of the wave functions yielded satisfactory results as compared
with the available theoretical data in literature.
It should be mentioned here that while progress is being made at the theoretical
level, there is a lack of experimental data to check this material. Experimental measured
transition energies and transition probabilities are highly desirable for the highly ionized
atomic system.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 111–122
119
References
[1] Jonsson, P., Froese Fischer, C. and Träbert, E., J. Phys. B: At. Mol. Opt. Phys. 31,
3497(1998).
[2] Biemont, E. and Zeippen, C. J., Physica Scripta T 65, 192(1996).
[3] Beauchamp, A., Wesemael, F., Bergeron, P. and Liebert, J., Astrophy. J. 44 L
85(1995).
[4] Liebert, J., Beaver, E. A., Robertson, J. W. and Strittmatter, P. A., Astrophy. J. 204
L 119(1976).
[5] Tachier, G., Fischer,C. F., J. Phys. B: At. Mol. Opt. Phys. 32, 5805(1999).
[6] Kingstone, A. E., Hibbert, A., J. Phys. B: At. Mol. Opt. Phys. 33,693(2000).
[7] Kingstone, A. E., Hibbert, A., J. Phys. B: At. Mol. Opt. Phys. 34,81(2001).
[8] Koc, K., J. Phys. B: At. Mol. Opt. Phys. 36,L93(2003).
[9] Fischer, C. F., Comput. Phys. Commun. 128, 635(2000).
[10] Fischer, C. F., Brage, T., and Per JÖnsson, “ Computational Atomic Structure: An
MCHF Approach,” edited by Institute of Physics, Bristol, 1997.
[11] Glass, R., Zeitschrift Fur Phys. A. 320, 545(1985).
[12] Bhatia, A. K., Feldman, U. and Doschek, G. A., J. Appl. Phys. 51,1464(1980).
[13] Tachier, G., Fischer, C. F., ADNDT. 87, 1(2004)
[14] Edlén, B. Phys.Scr. 20,129(1979).
[15] http://atoms.vuse.vanderbilt.edu
[16] Martin, G. A., Fuhr, J. R., and Wiese,W. L., “ Atomic Transition Probabilities
Scandium through Manganese” in Journal of Physical and Chemical Reference Data
,volume 17, 1988 supplement No 3.
120
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 111–122
Table 1 Transition energies (cm−1 ), wavelength (Å) and E2,M1 transition probabilities (sec−1 )
within the 2s2p configuration of Ti XIX
Transition
1P − 3P
1
0
1P
1
1P
1
− 3 P1
− 3 P2
3P
3
0− P1
3P
3
0− P2
3
3P
1− P2
a
b
c
d
e
f
a
b
c
d
e
f
a
b
c
d
e
f
a
b
c
d
f
a
b
c
d
f
a
b
c
d
f
∆Eij (cm −1 )
303821
303683
312470
30152
301504.5
303110.9
286691
286779
295394
285095
285095.2
286651.03
244459
244678.9
253267
242453
242453.6
244025.31
17130
16342.5
17073.5
16433.8
16459.87
59362
58997
59206.6
59066.7
59085.59
42232
42662.1
42122.9
42643.9
42625.72
λ ( Å)
329.14
329.29
320.03
331.65
331.67
329.91
348.8
348.02
338.53
350.76
350.76
348.85
409.06
408.7
394.84
412.45
412.45
409.79
5837.71
6119
5857
6085
6075.32
1684.6
1695
1689
1693
1692.44
2367.8
2344
2374
2345
2345.98
A (E2)
9.69E+00
9.87E+00
9.90E+00
9.80E+00
1.89E+00
1.49E+00
1.84E+00
A (M1)
5.03E+03
4.46E+03
4.94E+03
4.50E+03 D
4.93E+03
3.21E+03
2.81E+03
3.10E+03
2.80E+03 D
3.17E+03
3.32E+03
2.92E+03
3.29E+03
2.90E+03 D
3.25E+03
8.20E+01
7.81E+01
8.87E+01
7.93E+01
3.86E-02
3.65E-02
3.77E-02
3.77E-02
1.67E-02
1.60E-02
1.65E-02
1.04E+03
1.04E+03
9.99E+02
1.03E+03
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 111–122
121
Table 2 Transition energies (cm−1 ), wavelength (Å) and E2,M1 transition probabilities (sec−1 )
within the 2p2 configuration of Ti XIX
Transition
3P
3
0− P1
3P
3
0− P2
3P
3
1− P2
1D
2
− 3 P0
1D
2
− 3 P1
1D
2
− 3 P2
3P
1
1− S0
3P
1
2− S0
1D
2
− 1 S0
a
b
c
d
e
f
a
b
c
d
f
a
b
c
d
e
f
a
b
c
d
f
a
b
c
d
e
f
a
b
c
d
e
f
a
b
c
d
e
f
a
b
c
d
f
a
b
c
d
f
∆Eij (cm −1 )
29069
28776.98
28563.27
29044.44
29088.37
29001.24
56695.8
57110.2
58445.35
56561
56668.59
27626.8
28336.6
29895.3
27510.3
27527.73
27667.35
142806
144408.5
149423.2
141799.7
142657.99
113737
115625.7
120863.4
112756.1
112688.8
113656.75
86110.2
87260.03
90991.81
85251.49
85171.62
85989.4
301837.74
304053
309.4693
299320.5
299275.8
301596.32
274210.94
275710
279790.7
271805.6
273928.97
188100.74
188423.3
188828.9
186560.2
187939.57
λ ( Å)
3440.09
3475
3501
3443
3437.8
3448.09
1763.8
1751
1711
1768
1764.63
3619.7
3529
3345
3635
3632.7
3614.33
700.25
692.48
669.24
705.22
700.97
879.22
864.86
827.38
886.87
887.4
879.83
1161.3
1146
1099
1173
1174.1
1162.92
331.3
328.89
322.9
334.09
334.14
331.57
364.68
362.7
357.41
367.91
365.05
531.63
530.72
529.58
536.02
532.08
A (E2)
A (M1)
3.44E+02
4.19E+02
4.08E+02
4.19E+02 C
4.25E+02
3.54E-02
3.32E-02
3.79E-02
3.20E-02
1.55E-03
1.89E-03
1.64E-03
3.30E-03
1.48E-03
2.37E-03
7.72E-03
2.44E-01
2.43E-01
2.62E-01
3.90E-01
3.74E-01
4.09E-01
2.45E+02
2.77E+02
3.23E+02
2.77E+02 C
2.50E+02
2.14E+03
2.14E+03
2.50E+03
2.10E+03 D
2.42E+03
2.45E+03
2.48E+03
2.87E+03
2.50E+03 D
2.77E+03
3.90E+04
4.01E+04
4.27E+04
3.80E+04 D
4.17E+04
9.48E+01
8.71E+01
9.84E+01
1.01E+02
2.36E+02
2.14E+02
2.19E+02
2.05E+02
122
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 111–122
Table 3 Transition energies (cm−1 ), wavelength (Å) and E2,M1 transition probabilities (sec−1 )
within the 2s2 - 2p2 transitions for Ti XIX
Transition
1S − 3P
0
2
1S
0
1S
0
− 3 P1
− 1 D2
a
b
c
d
e
f
a
b
c
d
e
f
a
b
c
d
f
∆Eij (cm −1 )
832671.8
833263.9
834794.2
832362.2
832431.5
832488.52
805045
804893.8
804893.8
804829
804893.8
804821.16
918782
920556
925754.5
917599.6
918477.92
λ ( Å)
120.09
120.01
119.79
120.14
120.13
120.12
124.21
124.24
124.24
124.25
124.24
124.25
108.83
108.63
108.02
108.98
108.87
A (E2)
6.74E+02
4.84E+02
5.10E+02
A (M1)
4.80E+02 E
7.28E+02
2.05E+03
2.15E+03
2.25E+03
2.20E+03 E
2.09E+03
8.46E+03
5.62E+03
5.83E+03
7.28E+03
EJTP 3, No. 11 (2006) 123–132
Electronic Journal of Theoretical Physics
Some Properties of Generalized Hypergeometric
Thermal Coherent States
Dušan Popov
∗
University ”Politehnica” of Timişoara, Department of Physics,
Piaţa Regina Maria No. 1, 300004 Timişoara, Romania
Received 1 April 2006 , Accepted 2 June 2006, Published 25 June 2006
Abstract: The generalized hypergeometric coherent states (GHCSs) have been introduced by
Appl and Schiller [1]. In the present paper we have extended some considerations about GHCSs
to the mixed (thermal) states and applied, particularly, to the case of pseudoharmonic oscillator
(PHO). The Husimi’s Q distribution function and the diagonal P - distribution function, in the
GHCSs representation, have been deduced for these mixed states. The obtained distribution
functions were used to calculate thermal averages and to examine some nonclassical properties
of the generalized hypergeometric thermal coherent states (GHTCSs), particularly for the PHO.
We have also defined and calculated the thermal analogue of the Mandel parameter and the
thermal analogue of the second-order correlation function. By particularizing the parameters p
and q of the hypergeometric functions, we recover the usual Barut-Girardello coherent states and
their main properties for the PHO from our previous paper [9]. All calculations are performed
in terms of the Meijer’s G-functions [2], which are related to the hypergeometric functions. This
manner provides an elegance and uniformity of the obtained results and so the GHCSs become a
new field of application for these functions. Moreover, this mathematical approach can be used
also for other kind of coherent states (e.g. Klauder-Perelomov, Gazeau-Klauder or nonlinear
coherent states [10], [12]).
c Electronic Journal of Theoretical Physics. All rights reserved.
Keywords: Coherent states, Hypergeometric Functions, Pseudoharmonic Oscillator
PACS (2006): 03.65C, 03.65.-w, 02.30, 42.50.Ar
1.
Introduction
By considering the so-called generalized lowering and raising operators
∗
dusan popov@yahoo.co.uk
124
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 123–132
p Uq =
∞
pf
q
(n)|λ;n >< λ;n + 1|,
(1)
n=0
+
p Uq
=
∞
pf
q
(n)|λ; n + 1 >< λ; n|
(2)
n=0
where |λ; n > are the vectors of an orthonormal basis (usually, the Fock basis indexed by a
real parameter λ which can play the role of the degeneracy parameter), Appl and Schiller
have defined the generalized hypergeometric coherent states (GHCSs) as the eigenvalues
of the lowering operator [1]:
p Uq |p; q; λ; z
>= z|p; q; λ; z > .
(3)
The numbers p and q are natural and the positive functions p f q (n) were defined as
follows:
pf
q
(n) =
(n + 1)
(b1 + n)(b2 + n) . . . (bq + n)
.
(a1 + n)(a2 + n) . . . (ap + n)
(4)
Consequently, the expansion of the GHCSs in the Fock basis is [1]:
|p; q; λ; z >=
∞
− 12
zn
2
$
|λ; n >
N
(|z|
;
λ)
p q
ρ
(n;
λ)
p
q
n=0
(5)
where the strictly positive parameter functions of the discrete variable n are defined as:
'q
j=1 (bj )n
p ρq (n; λ) = Γ(n + 1) 'p
i=1 (ai )n
(6)
and where (x)n = Γ(x + n)/Γ(x) is the Pochhammer’s symbol [2].
The appellation ”generalized hypergeometric coherent states” becomes from the normalization function which is given by generalized hypergeometric functions:
2
2
p Nq (|z| ; λ) = p Fq (a1 , a2 , . . . , ap ; b1 , b2 , . . . , bq ; |z| ) =
∞
∞
(a1 )n (a2 )n . . . (ap )n (|z|2 )n
1
2 n
=
n=0
(b1 )n (b2 )n . . . (bq )n
n!
=
ρ (n; λ)
n=0 p q
(7)
(|z| ) ,
where, generally, we consider that ai = ai (λ) and bj = bj (λ) are the complex functions.
In the cited paper [1] were examined the main properties of the GHCSs, including the
resolution of unity. Let us here we express the weight function of the integration measure
through the Meijer’s G-functions (whose definition and main properties can be found,
e.g. in [2]):
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 123–132
125
d2 z
2
p wq (|z| ; λ) =
π
⎞
⎛
⎞
⎛
{1 − ap } ; /
d2 z 1,p ⎜
⎟ q+1,0 ⎜ 2 /; {ap − 1} ⎟
=
Gp,q+1 ⎝−|z|2
|z|
G
⎠
⎠ p,q+1 ⎝
π
0, {bq − 1} ; /
0 ; {1 − bq }
dμ(p; q; λ; z) =
(8)
where, in order to simplify the formulae writing, we have used the following notation:
{cl } ≡ c1 , c2 , . . . , cl .
Below we have also used the connection between the generalized hypergeometric functions and the Meijer’s G-functions [2]:
⎛
⎞
'q
⎜ {1 − ap } ; / ⎟
j=1 Γ(bj )
1,p
'
F
({a
};
{b
};
x)
=
G
−x
⎝
⎠.
p q
p
q
p
p,q+1
i=1 Γ(ai )
0 ; {1 − b }
(9)
q
2.
GHCSs for Thermal States
In [1] were examined some properties of the GHCSs for the case of pure quantum
states. One of the aims of present paper is to extend this examination also to the case of
the mixed quantum states. As a typical example of mixed states we consider the thermal
states described by the normalized canonical density operator:
∞
1 −β En,λ
ρλ =
e
|λ; n >< λ; n|
Zλ n=0
(10)
where En,λ are the eigenvalues of the Hamiltonian operator of the examined quantum
system and the normalization constant Zλ = Zλ (β) is the partition function.
In the present paper we will limiting only to the case of quantum systems with linear
energy spectra with respect to the energy quantum number n and we will to describe
the corresponding method for examine some properties of generalized hypergeometric
thermal coherent states (GHTCSs). For systems with more complicated energy spectra
it must elaborate specifical methods (e.g. for the Morse oscillator, see [3]).
Let us we assume that the energy of linear spectra (where n = 0, 1, . . . , ∞ and ω is
the angular frequency) is
En,λ = E0,λ + nω
(11)
and then the partition function becomes:
Zλ =
∞
e−β En,λ = e−β E0,λ (< n > +1)
n=0
where we have used the expression of the Bose-Einstein distribution function:
(12)
126
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 123–132
1
.
(13)
−1
The first characteristic of the GHTCSs we want to examine is the thermal Husimi’s
distribution function:
< n >=
eβω
∞
2 n
1
1
−β En,λ (|z| )
Q|p;q;λ;z> (ρλ ) ≡< p; q; λ; z|ρλ |p; q; λ; z >=
.
e
Zλ p Fq ({ap }; {bq }; |z|2 ) n=0
p ρq (n; λ)
(14)
For the above examined linear spectra this expression becomes:
<n>
2
1
p Fq ({ap }; {bq }; <n>+1 |z| )
=
Q|p;q;λ;z> (ρλ ) =
< n > +1
F ({a }; {bq }; |z|2 )
⎛ p q p
⎞
{1 − ap } ; /
⎜ <n>
⎟
2
G1,p
|z|
−
⎝
⎠
p,q+1
<n>+1
0 ; {1 − bq }
1
⎞ .
⎛
=
< n > +1
{1 − ap } ; /
⎟
⎜
2
G1,p
⎠
p,q+1 ⎝−|z|
0 ; {1 − bq }
(15)
Using the integration measure expression and the properties of the integral of Meijer’s
G-functions products [2], it is not difficult to prove that the thermal Husimi’s distribution
function is positive, normalized to unity with the measure dμ(p; q; λ; z) and provides a
two-dimensional probability distribution over the complex z = |z| exp (iϕ) plane:
dμ(p; q; λ; z)Q|p;q;λ;z> (ρλ ) = 1.
(16)
After these considerations let us we perform the diagonal expansion of the density
operator ρλ over the projector of GHTCSs:
ρλ = dμ(p; q; λ; z)|p; q; λ > p Pq (|z|2 ; λ) < p; q; λ; z|
(17)
where the function p Pq (|z|2 ; λ) is called P-function or P-distribution function, even if is in
fact a quasi-probability distribution, because of their negative values on certain domains.
By inserting the expressions for the GHTCSs and the integration measure, performing
the angular integration and using the notation x = |z|2 , we lead to the following equation:
⎞
⎛
'p
2
∞
Γ(ai ) |λ; n >< λ; n| R
⎜ /; {ap − 1} ⎟
ρλ = 'qi=1
dx xn Gq+1,0
x
⎠ p Pq (x; λ)
⎝
p,q+1
p ρq (n; λ)
0
j=1 Γ(bj ) n=0
0, {b − 1} ; /
q
(18)
where R is the convergence radius of the GHTCSs [5].
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 123–132
127
If we perform the function change
⎡
⎛
⎞⎤−1
⎢ q+1,0 ⎜ /; {ap − 1} ⎟⎥
P
(x;
λ)
=
G
x
⎣
⎝
⎠⎦ p hq (x; λ)
p q
p,q+1
0, {bq − 1} ; /
(19)
and consider the expression of strictly positive parameters functions p ρq (n; λ) and the
energy En,λ , in order to obtain the expression (10) for the normalized canonical density
operator ρλ , we must solve the following equation where p hq (x; λ) is, for the moment, an
unknown function which must be determined:
0
R2
'q
1
1
j=1 Γ(bj + n)
e−β E0,λ βω n Γ(n + 1) 'p
.
dx xn p hq (x; λ) =
Zλ
(e )
i=1 Γ(ai + n)
(20)
The above problem is just the moment power problem (if R is a finite quantity we
have the Haussdorff moment problem, while if R is infinite we have the Stieltjes moment
problem [4]) and, in order to solve it, we extend the real values n to the complex ones
s = n + 1 [5] and so we lead to the definition of the Meijer’s G-functions [2]. In this stage
the problem can be solved.
Finally, for the P-function we obtain:
⎞
/; {ap − 1}
⎟
q+1,0 ⎜ <n>+1
Gp,q+1
⎠
⎝ <n> |z|2
0, {bq − 1} ; /
1
2
⎛
⎞
p Pq (|z| ; λ) =
<n>
/; {ap − 1}
⎜ 2
⎟
Gq+1,0
⎠
p,q+1 ⎝|z|
0, {bq − 1} ; /
⎛
(21)
Using the properties of Meijer’s G-functions [2] it is not difficult to prove that the
P-function is normalized to unity:
dμ(p; q; λ; z)p Pq (|z|2 ; λ) = 1.
(22)
By particularizing the parameters p and q of the hypergeometric functions, it must
examine the positivity of the weight function of the integration measure p wq (|z|2 ; λ) (see,
also [1]) and of the Husimi’s function Q|p;q;λ;z> (ρλ ). Implicitly from their definition, the
Husimi’s function Q|p;q;λ;z> (ρλ ) is always positive, while the P-function can bring also
negative and singular values (for non-classical fields or states). For these reasons, the
P-function is sometimes called the quasi-distribution function. If the P-function becomes
negative for a certain state, then this state has a non-classical character [6].
One of the practical utility of the P-function of a density operator (which can be
measured in experiments) consists in its role in the calculation of thermal averages of an
observable A which characterize the quantum system:
128
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 123–132
< A >p;q;λ = T r(ρλ A) =
dμ(p; q; λ; z)p Pq (|z|2 ; λ) < p; q; λ; z|A|p; q; λ; z > .
(23)
On the other hand, if the P-function has non-classical character (e.g. for a paircoherent state [7]), then the state is entangled. This will be a good test for the inseparability of the quantum states, particularly for the coherent states.
In many cases the calculations in the GHTCSs representation are much simpler that
in other representations (e.g. in the coordinate or in the momentum representations),
which constitute a good reason for using the GHTCSs formalism.
An important class of observables is represented by the diagonal operators in the Fock
Basis |λ; n >. As an example we examine the thermal averages of integer powers of the
number operator < N s >, where s = 1, 2 . . . In this reason we adopt an original ansatz.
If we calculate the average of the operator eεN in a pure GHTCS |p; q; λ; z >, where ε is
a small positive parameter, i.e.
⎛
< eεN
⎞
{1 − ap } ; /
⎜ ε 2
⎟
G1,p
⎠
p,q+1 ⎝−e |z|
0
;
{1
−
b
}
ε
2
q
p Fq ({ap }; {bq }; e |z| )
⎞ ,
⎛
>|p;q;λ;z> =
=
2
p Fq ({ap }; {bq }; |z| )
{1 − ap } ; /
⎟
⎜
2
G1,p
−|z|
⎠
⎝
p,q+1
0 ; {1 − bq }
(24)
we observe that
s
< N >|p;q;λ;z> = lim
ε→0
∂
∂ε
s
< eεN >|p;q;λ;z> .
(25)
Then the corresponding thermal averages are
s
< N >=
1
lim
dμ(p; q; λ; z)p Pq (|z| ; λ) < N >|p;q;λ;z> =
< n > ε→0
2
s
∂
∂ε
s
1
,
− eε
(26)
where we have used the integral properties of the Meijer’s G-functions and also the
particular expression for the function G11
11 (x| . . . ) [2].
We can observe that these thermal averages are independent of the parameters p, q and
λ, which was to be expected, due to the fact that the thermal averages are independent
on the representation.
The first two power thermal averages can be expressed through the Bose-Einstein
distribution function (13):
< N >=< n >,
< N 2 >=< n > (2 < n > +1).
<n>+1
<n>
(27)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 123–132
129
With these thermal averages we can define and calculate the thermal second-order
correlation function g (2) and the thermal Mandel parameter Q(< n >) [3], which are
defined as the thermal analogues of the corresponding quantities for the pure GHCS
|p; q; λ; z > [6]:
g (2) =
< N2 > − < N >
= 2,
(< N >)2
Q(< n >) =< N > g (2) − 1 =< n > .
(28)
So, all GHTCSs, independently of the values of integer parameters p, q and λ, have
the same (constant) thermal second-order correlation function g (2) and the same thermal
Mandel parameter Q(< n >) (which are dependent of the equilibrium temperature T ,
through the quantity β = (kB T )−1 ).
3.
Particularization for the Pseudoharmonic Oscillator
In the following, we particularize the GHTCSs formalism for the case of the pseudoharmonic oscillator (PHO), whose effective potential is [8]
mred ω 2 2
r0
VJ (r) =
8
r
r0
−
r0
r
2
+
1
2
J (J + 1) 2
2mred
r
(29)
where mred is the reduced mass of the quantum system (e.g. the diatomic molecule), r0
is the equilibrium distance between the diatomic molecule nuclei and J = 0, 1, 2, . . . is
the rotational quantum number, while n will be the vibrational quantum number.
The importance of the PHO consists in the fact that this potential also admits the
exact analytical solution of the rotational-vibrational Schrödinger equation, being in a
certain sense an intermediate potential between the three-dimensional harmonic oscillator
potential (HO-3D)(an ideal potential) and other much anharmonic oscillator potentials
(the more realistic potentials, e.g. Pöschl-Teller, or Morse [3]).
Earlier we have showed [9] that the dynamical group associated with the PHO is
SU (1, 1) quantum group, whose lowering operator is
K− =
∞
$
(n + 1)(2k + n)|k; n >< k; n + 1|
(30)
n=0
where k is the Bargmann index which labels the irreducible representations of this group.
By comparing this expression with that of generalized hypergeometric operator p Uq
and also with the positive functions p f q (n) [1], we observe that the above expression is a
particular case, if we take p = 0, q = 1, b1 = 2k and λ = k. So, the GHCSs denoted as
|0; 1, k; z >= |k; z >, defined as the eigenstates of the lowering operator K− , i.e.
K− |k; z >= z|k; z >,
(31)
are called the Barut-Girardello coherent states for the PHO and their expansion over the
Fock vector basis |k; n > is [9]
130
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 123–132
∞
|z|2k−1
zn
$
|k; n > .
I2k−1 (2|z|) n=0 n!Γ(2k + n)
|k; z >=
(32)
In the following we look over all the quantities which characterize the CSs of the PHO,
previously obtained in [9], but now regarded only as the particular case of the GHCSs,
with the above mentioned values of the parameters p, q, b1 and λ.
The weight function of the integration measure finally is expressed as a product of
two Bessel functions of the first and second kind:
2
0 w1 (|z| ; λ)
2,0
2
2
= G1,0
0,2 (−|z| |0; 1 − 2k) G0,2 (|z| |0, 2k − 1) = 2I2k−1 (2|z|) K2k−1 (2|z|). (33)
On the other hand, the PHO energy eigenvalues are
mred ω 2 2
r0 + ω n ≡ E0,k + ω n.
(34)
2
while the connection between the Bargmann index k and the rotational quantum number
J is [9]
En,k = ω k −
1 1
k = k(J) ≡ +
2 2
"
1
J+
2
2
+
m
red ω 2
r0
2
2
# 12
.
(35)
Accordingly to this, the partition function for a fixed quantum number J becomes
Zk ≡ ZJ =
∞
e−βEn,k = e−βE0,k (< n > +1) .
(36)
n=0
Similarly, the total partition function (by considering also the degeneration of the
rotational states) is
Z=
∞
β
(2J + 1)ZJ = e
mred ω 2 2
r0
4
(< n > +1)
J=0
∞
(2J + 1)
J=0
<n>
< n > +1
k(J)
.
(37)
The total partition function is a quantity of exceptional information importance because it enables the calculations of all statistical properties of a PHO quantum canonical
gas (for details, see [9]).
For the thermal Husimi’s distribution function for the PHO it follows:
Q|0;1;k;z> (ρk ) =
1
< n > +1
=
G1,0
0,2
1
< n > +1
<n>
|z|2 0;
− <n>+1
1 − 2k
=
2 0; 1 − 2k
G1,0
−|z|
0,2
<n>
k− 12 I2k−1 2|z|
<n>+1
< n > +1
<n>
I2k−1 (2|z|)
(38)
.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 123–132
131
On the other hand, for the P-function of the PHO we obtain:
2k
−
1
1
2
=
0 P1 (|z| ; k) =
<n>
2 0, 2k − 1
|z|
G2,0
0,2
<n>+1
k− 21 K
2k−1 2|z|
<n>
1
< n > +1
.
=
<n>
<n>
K2k−1 (2|z|)
G2,0
0,2
<n>+1
|z|2 0,
<n>
(39)
In both equations, in order to express the particular values of the Meijer’s G-functions,
we have used the book of Mathai and Saxena [2].
Because, as we have seen, the thermal averages of the integer powers of the number
operator are independent of the parameters p and q of the hypergeometric functions, the
indicated values in the previous section are identical both for the CSs of the PHO and for
the GHCSs with arbitrarily parameters p and q . This property also pass to all statistical
averages of the diagonal operators in the Fock basis.
4.
Concluding Remarks
In the paper we have shown that the formalism of the GHCSs, previously introduced
by Appl and Schiller for the pure quantum (coherent) states [1], can be extended also to
the mixed (thermal) quantum states and in this manner it can be connected with more
practical, quantum statistical, problems.
We have showed that by particularizing the parameters of the hypergeometic functions
(p = 0 and q = 1) and also for the particularly value of parameter λ (i.e. λ = k and b1 =
2k) we recover the Barut-Girardello coherent states (BG-CSs) for the pseudoharmonic
oscillator (PHO), deduced in a previous paper [9], with all statistical properties evinced
therein.
It is important to point out that we have performed all calculations in terms of the
Meijer’s G-functions [2], which are more generally functions and whose particular cases
are also the hypergeometric functions. This manner provide an elegance and uniformity
in the use of the GHTCSs formalism and, implicitly, indicates a new field of applications
of these functions in theoretical physics.
Moreover, the mathematical approach of the GHTCSs can be used also for other kind
of thermal coherent states, e.g. Klauder-Perelomov, Gazeau-Klauder (for PHO these CSs
were deduced in [10]) or for nonlinear CSs [11]. This approach can be also applied in the
theory of quantum information [12].
132
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 123–132
References
[1] Appl T and Schiller D 2004 J. Phys. A: Math. Gen. 37 2731-2750; arXiv: quantphys/0308013 (2003)
[2] Mathai A M and Saxena R K 1973 Generalized Hypergeometric Functions with
Applications in Statistics and Physical Sciences (Lecture Notes in Mathematics vol.
348) (Berlin: Springer)
[3] Popov D 2003 Phys. Lett. A 316 369-381
[4] Quesne C 2000 Phys. Lett. A 272 313-321
[5] Klauder J R Penson K A and Sixderniers J-M 2001 Phys. Rev. A 64 013817
[6] Mandel L, Wolf E 1995 Optical Coherence and Quantum Optics (Cambridge:
University Press)
[7] Agarwal G S and Biswas A 2005 Quantitative measures of entanglement in pair
coherent states, arXiv: quant-ph/0501012
[8] Sage M, Goodisman M J 1985 Am. J. Phys 53 350-355
[9] Popov D 2001 J. Phys. A: Math. Gen. 34 5283-5296
[10] Popov D 2004 Density Matrix - General Properties and Applications in the Physics
of Multiparticle Systems (in Romanian) (Timişoara: Editura Politehnica)
[11] Dodonov V V 2002 J. Opt. B: Quant. Semiclass. Opt. 4 R1
[12] Fujii K 2002 Coherent States and Some Topics in Quantum Information Theory:
Review, arXiv: quant-ph/0207178
EJTP 3, No. 11 (2006) 133–142
Electronic Journal of Theoretical Physics
Space-Filling Curves for Quantum Control
Parameters
Fariel Shafee
∗
Department of Physics,
Princeton University,
Princeton, NJ 08544, USA
Received 28 April 2006 , Accepted 8 June 2005, Published 25 June 2006
Abstract: We consider the use of space-filling curves (SFC) in scanning control parameters
for quantum chemical systems. First we show that a formally exact SFC must be singular in
the control parameters, but a finite discrete generalization can be used with no problem. We
then make general observations about the relevance of SFCs in preference to linear scans of the
parameters. Finally we present a simple magnetic field example relevant in NMR and show
from the calculated autocorrelations that a SFC Peano-Hilbert curve gives a smoother sequence
than a linear scan.
c Electronic Journal of Theoretical Physics. All rights reserved.
Keywords: Quantum Control, Space-Filling Curves, Peano-Hilbert Curve, Control Parameters
PACS (2006): 02.30.Yy , 95.30.Ft, 98.38.Bn, 46.25.Cc, 02.70.Ns
1.
Introduction
A quantum system is represented by a state vector in Hilbert space, and the system
has a Hamiltonian operator that determines how the system evolves by itself. The system
may be in the ground state, in which case the evolution is simply a phase change with
time of the state vector. If it is in an excited state then sooner or later it will drop to a
lower state with either simply an emission of radiation, or more fundamental changes in
the system. In a second quantized theory there is provision for considering annihilation
and creation of particles; in a first quantized theory the transition of the dynamical system in terms of changes in its Hamiltonian description is more complicated. Feynman’s
path integral method is second quantized and yet it avoids explicit use of creation and
annihilation operators. The transition in that case is implemented by a similar Hamil∗
f shafee@yahoo.com
134
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 133–142
tonian as the second quantized case, and though an obvert operatorial interpretation is
avoided the path integral actually achieves the same purpose of a discrete change in time
from one set of fields to another. In particle physics, at relativistic energies this is a
simple procedure.
In quantum chemistry usually we do not worry about changes in the nuclei, and
hence with particle creation or destruction. The system is described by fixed nuclei and
a number of electrons, all with the constant masses and charges. It is fairly easy to set
up a Schrodinger equation, though not to solve it. If a large molecule dissociates, or if
two molecules synthesize into a bigger one, no new particles are created or destroyed,
they all correspond to the same Hamiltonian with the same nuclei and electrons as the
components. However, the states vary. Like ionization being the free state of an electron,
dissociation is also a new state of the system for the same Hamiltonian.
But it is possible to change the environmental factors of a system. That can be discrete
or continuous. We introduce a control Hamiltonian that changes the overall development
of the system. Populations can be inverted from the original ground state to some other
desired state. Or other properties of the original state may be changed, e.g. expectation
values of an observable. If this control Hamiltonian is a function of two variables, and
these variables are continuous, then the control Hamiltonian space may be described by
a two dimensional surface. But the process of control takes place as a temporal sequence
and any variation of these parameters must also be a temporal sequence. If we try to
map a two-dimensional surface by a one-dimensional sequence, we run into the concept
of space-filling curves.
There is another way space-filling curves can enter quantum chemical systems. If the
state space is continuous and described by two parameters, like the parameters of the
control Hamiltonian, then too we have a surface of states. In certain cases it may also be
possible to relate a two-dimensional state space with a two parameter control space, i.e.
two relate the two sets of parameters, So that by changing the control space parameters
we achieve an equivalent change in the state space. Evolution of the system in state space
can be ill-defined when we try to view the process in a continuous way, but using the
concept of Bell’s “beables” Rabitz et al [1] have shown that a satisfactory picture can be
formed in a statistical sense. Then the concept of SFC can apply to the state space too
in terms of “beables”.
In this work we shall first show that in the mathematical sense a true SFC must a
singular object in terms of the control parameters. However, its discrete finite versions
do not pose the problems of the singularity. We shall then consider a comparison of
different ways of scanning the parameter space, including a finite representation of the
Peano-Hilbert curve.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 133–142
2.
135
SFC and Singular Parameters
A SFC must have a Hausdorff (fractal) dimension more than one; otherwise it cannot
fill a two-dimensional space even partly. Let us take a small segment of the curve
f (x, y) = c
(1)
where x , y are the (control) parameters.
So anywhere on the curve the tangent is given by
dy/dx = −[∂f /∂x]/[∂f /∂y]
(2)
Now if we consider only a small section of the curve
ds = dx
However, for a fractal of dimension n
$
1 + (dy/dx)2
ds = ds0 (Λ0 /Λ)n Λ
(3)
(4)
So this goes to infinity as scale parameter Λ goes to zero. Hence dy/dx must be singular
at every point of the curve.
Hence either ∂f/∂y = 0 or ∂f/∂x is singular at every point. The first option makes
the curve trivial. The second option simply states that physically it is impossible to span
a two-dimensional parameter space completely using only a single parameter [e.g. time]
which remains within a finite, and hence practically realizable range.
Hence a practical control experiment involving a two-parameter space must involve
discretization of the space, such as choosing a lattice [in parameter space]. The experiment
then would correspond to a finite number of steps through the lattice, and any particular
sequence will correspond to a specific trajectory, and the choice of trajectory should be
such as to optimize the output expected or desired.
3.
Control Parameter Space and Sfc Approximations
In quantum control processes the effect of a sequence of laser pulses needed to achieve
any given program of operations has been studied by Ramakrishna, Rabitz et al. [2].
They considered a one-dimensional model with a one-dimensional amplitude. In general
a randomly polarized signal, or one with a given polarization, will have two components
of the electrical vector, so the field space would be two-dimensional. We can proceed in a
similar fashion and obtain formulas which are generalizations involving both components.
The unitary operator which gives the evolution of the system can now be described by
similar unitary operators
U I = exp[iΓ]Vk ... V1
(5)
136
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 133–142
with k being the pulse number in the sequence,
⎞⎤
⎛
⎡
sin φ ⎟⎥
⎜0
⎢
Vk = exp ⎣Ck ⎝
⎠⎦
− cos φ 0
(6)
in the two-dimensional vector space formed by |m> and |m+1>
and
tk+1
μ
Ck =
dt E.
tk
E and the dipole µ being two-dimensional vectors.
Now, with an unknown orientation of the dipole vector μ it may be necessary to try
out different components of E to get the optimal Ck and hence it may be necessary to
find the best algorithm to hit on the right value. If it is (the discrete practical limit of) a
SFC it may offer certain advantages over a random choice or a linear scan, as we shall see
later with a particular numerical example. The point to remember is that on a SFC two
dimensional neighborhood values [with small variations of both components] are tried in
sequence before moving out of a region, and physically that may offer the possibility of
locating the right Ck quicker than varying one parameter at a time as in a linear scan,
because such a trajectory is likely to have a better autocorrelation [and hence a smoother
variation] than any other option.
We may also be interested in a system with an arbitrarily large number of neighboring
states each attainable by small changes in control parameters. For example a simple Bohr
atom has energy levels going like ∼ 1/n 2 , where n is the principal quantum number. So
if we want to jump from any state |n 1 ¿ to another |n 2 ¿, we need a frequency
ω(n 1 , n2 ) ∼ 1/n 22 − −1/n21
and this will be nondegenerate for different n1 and n2 , but almost continuous for n1
and/or n2 large.
Now supposing we need a particular sequence of jumps (n1 → n 2 ) in an ensemble of
such atoms, we would need to use ω( n 1 , n2 ) in the given sequence, i.e. we would need to
trace out a trajectory in the (1/n21 , 1/n22 )space (we use the reciprocals to emphasize the
approximation to a continuum).
There is a solved problem in mathematics, called the ETSP [Euclidean traveling
salesman problem] which states that the optimum joining a very large random sequence
of cities within a disc is given in the infinite limit by the SFC called MNPeano [3], which
is similar to the SFC Peano curve [Figure 1]. The path length is given by [in the units
for A below]
√
Loptimum = 0.72 N A
(7)
where N is the number of states to be excited and A is the area of the control parameter
space. This expression is similar to a random walk problem, except for the coefficient,
which is somewhat smaller than 1.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 133–142
137
So the quickest way to make a large sequence of quantum jumps for an ensemble
of Bohr atoms would be given by the discrete approximation of such a MNPeano curve
in the control parameter space which would need to be somehow translated into the
corresponding frequency of the exciter. Similar situations of greater practical utility
may be found involving electron energy bands of materials and continuous spectra of
molecules. The former may be relevant to electronic devices.
4.
Magnetic Environment: A Numerical Example
NMR spectroscopy is a very useful tool in augmenting our knowledge of the structure
of large molecules when X-ray diffraction data or analyses are insufficient. The magnetic
environment of a nucleus with nonzero magnetic moment, such as a proton, or P31 may
be affected by the electron orbits in the neighborhood contributing their own magnetic
fields. As a result nuclear magnetic resonance would occur for a shifted field compared
to the lone nucleus, which may be expressed as a chemical shift in the Lande g factor.
N
HI = −g s . Hμ
(8)
where s is the nuclear spin, H is the magnetic field at the nucleus, and μN is the nuclear
magneton. In nuclear magnetic imaging used in medical diagnosis one displaces a component of the magnetic field spatially as a raster scan and observes the absorption of the
electromagnetic waves for the resonant change of spin state.
In theory it is possible to combine these two uses of nuclear resonance to get a spatial
picture of the magnetic environment in a given space of the molecule, e.g. the active site
of an enzyme. This is apparently a more direct and theoretically more convenient way,
involving no ill-posed inversion problem, than another novel method using the Ehrenfest
theorem and a classical correspondence proposed by Rabitz and Vivie-Riedle [4].
Let us take a crystal of the molecule and inundate it slow neutrons. Let us use two
orthogonal narrow strips [one along x and the other along y say] of movable magnetic
fields in the z−direction [Figure 2] such that their combined value, which occurs at the
intersection of the two strips and hence corresponds to a unique (x, y) point is near the
resonance value for the r.f. radiation used for the neutron. Then, if the r.f. is varied over
a relevant range for each intersection of (x, y) and the absorbed frequency observed, we
get the magnetic field at that point. This may allow us to draw a magnetic map inside
the molecule and hence can give us useful ideas about its structure. The problem we ask
now is what should be the trajectory of the control parameters (x, y) for an optimal scan.
Again, it would appear that the trajectory with the highest autocorrelation would be the
best, because it would identify the two-dimensional region in space with smoothness and
hence reliability of measurement.
We performed a simulation with four dipoles, all with equal x, y components at four
corners of a square, with the field measured in a smaller [with half the sides] square inside
138
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 133–142
the larger one. The field is of course
3
=
H
[μ/(4π)][3 m
i . ri ri /ri2 − mi ]/ri
(9)
i
where r i is the unit vector from the magnetic moment vector m i to the observation point
and ri is the magnitude of the distance vector etc.
We calculated the autocorrelations for the i−th value in the sequence with the (i +k)th value, both for Hx and Hy , using a linear scan and then using a Hilbert-Peano scan,
which is a SFC [finite approximation at a given scale, in this case with 32 points each
way].
Table I gives the autocorrelations for Hx (Hy is similar). It is obvious that the HilbertPeano trajectory gives the much better correlated sequence than a linear scan.
5.
Conclusions
We have argued in general terms about the relevance of SFCs in choosing trajectories
of control parameters, and have also shown by a simple simulation example that a SFC
is better autocorrelated and hence smoother when magnetic data are sequenced. This is
similar to the finding in image processing [5] but much more pronounced. More realistic
examples and details are being investigated. Quantum control has previously [2] been
identified as useful in three contexts: (a) transferring population from an initial state to a
given state, (b) making transitions through a given sequence of states, and (c) forcing the
system to have a given expectation value for an observable. We add here the possibility
of also using quantum control not to make transitions (the resonant nucleon returns to
ground) but to probe the system as it is.
The author thanks Professor H. Rabitz for useful discussions and Ignacio Sola and
Andrew Tan for encouragement.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 133–142
139
References
[1] Denis E. and H. Rabitz, Beable trajectories for revealing quantum control mechanisms,
quant-ph/0208109 (2002)
Meier C., Mixing quantum and classical dynamics using Bohmian trajectories, ITAMP
Workshop paper (May 9-12, 2002) Universite Paul Sabatier Preprint
[2] Schirmer S.G., A Greentree, V. Ramakrishna and H Rabitz, Quantum control using
sequence of simple control pulses, quant-ph-0105155 (2001)
[3] Norman M.G., and P Moscato, The Euclidean Traveling Salesman Problem and a
Space-Filling Curve (CeTAD, Universidad Nacional de La Plata preprint)
[4] Kurtz L, H Rabitz and Regina de Vivie-Riedle, Optimal use of time-dependent
probability density data to extract potential energy surfaces, physics/0109043 (2001)
[5] Dafiner R, D Cohen-Or and Y Matias, Context-based Space-Filling Curves, submitted
to EUROGRAPHICS 2000 (vol. 19, No. 3)
140
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 133–142
TABLE I
Autocorrelation <Hx (i) Hx (i+k)>
k
Linear scan
Hilbert-Peano
0
1.00
1.000
1
0.86
0.995
2
0.73
0.990
3
0.60
0.984
4
0.49
0.980
5
0.38
0.973
6
0.29
0.967
7
0.19
0.960
8
0.11
0.955
9
0.04
0.949
10
-0.01
0.943
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 133–142
Fig. 1 Peano-Hilbert curve [finite approximation]
141
142
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 133–142
Fig. 2 using two orthogonal fields as control to locate point of NM resonance.
EJTP 3, No. 11 (2006) 143–149
Electronic Journal of Theoretical Physics
The Spectrum of the Lagrange Velocity
Autocorrelation Function in Confined
Anisotropic Liquids
Sakhnenko Elena I∗ . and Zatovsky Alexander V.
Department of Atmosphere Physics,
Odessa State Environmental University,
15 Lvovskaya Str., Odessa, 65016,
Ukraine
Received 6 May 2006 , Accepted 5 June 2006, Published 25 June 2006
Abstract: The results of our further analysis of the thermal hydrodynamic fluctuations in an
anisotropic liquid under heterogeneous conditions are represented. The heterogeneity is modeled
in the form of a plane-parallel layer, the liquid is considered is taken to be incompressible, and the
rapid processes of the angular momentum relaxation to equilibrium are ignored. The extended
system of hydrodynamics equations is linearized for small deviations from the equilibrium values.
For the case of spontaneous fluctuation fields being present in the system of equations for the
velocity and inertia tensor components, the boundary problem solution is found in the form of
an expansion in the harmonic functions. The spectral densities of the fluctuation correlation
functions are obtained by using the fluctuation dissipation theorem (FDT). A special attention
is paid to the correlation functions (CFs) for the velocity field in the anisotropic liquid. The
spectrum of the Lagrange velocity autocorrelation function (LVACF) and the collective part
of the self–diffusion coefficient of the molecules are determined as functions of the coordinate
normal to the confining planes.
c Electronic Journal of Theoretical Physics. All rights reserved.
Keywords: Confined Liquid, Anisotropic Liquid, Velocity Fluctuations,
Autocorrelation Function, Self-Diffusion Coefficient, Lagrange Particle
PACS (2006): 61.25-f , 82.70-y
1.
Spectrum,
Introduction
The investigation of thermal hydrodynamic fluctuations in heterogeneous systems is
an urgent question at present [1-8]. The heterogeneity can be caused by different external
∗
geophys@ogmi.farlep.odessa.ua
144
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 143–149
factors. Under natural conditions, such factors as geometrical restrictions on the system
and imposition of external electromagnetic and gravitation fields are most frequently
encountered. Despite the different nature of these factors, their influences are identical
and reduce to some orientation ordering of the liquid’s anisotropic molecules. In this
connection, the need for studying heterogeneous anisotropic liquids, including confined
ones, arises.
The influence of a geometrical restriction on the dynamic properties of the system
is taken into account by supplementing the system of hydrodynamics equations with
proper boundary conditions. The anisotropy of the liquid’s molecules can incorporated
by introducing into the above system the anisotropy tensor, which serves as an additional
field variable and characterizes the deviations of the anisotropic molecule axes within a
liquid volume element from the isotropic distribution. Such an approach was first offered
and justified in [9], and then refined in [10]. Recently, with the participation of one of us
[11], a complete system of hydrodynamics equations was developed in which along with
the local traverse velocity, the internal angular momentum is taken into account and the
inertia tensor is considered as the anisotropy tensor. The equations were used for the
study of the space–time CF and the spectral density of the inertia tensor fluctuations
of the liquid Lagrange particle in a homogeneous liquid under equilibrium and nonequilibrium conditions – against the background of a liquid flow with constant velocity
gradient [11, 12].
In this report, the results of our further analysis of the thermal hydrodynamic fluctuations in an anisotropic liquid are presented. The fluctuations are studied for the
heterogeneous conditions modeled, for the sake of simplicity, in the form of a planeparallel layer. The extended system of hydrodynamics equations is linearized for small
deviations from the equilibrium values, the liquid is taken to be incompressible, and the
rapid processes of the angular momentum relaxation to equilibrium are ignored. The solution of the boundary problem with arbitrary initial conditions is found for the system
of linked equations for the velocity and inertia tensor components by developing it in an
expansion in the harmonic functions. The spectral densities of the fluctuation CFs for
the expansion amplitudes are obtained by using FDT. The results obtained for the field
variable CFs for the heterogeneous anisotropic liquid are compared with those for the
bulk case. The LVACF spectrum for the anisotropic liquid and the collective part of the
molecule self–diffusion coefficient are determined as functions of the coordinate normal
to the restrictive planes. The transformation from the Euler to the Lagrange variables is
carried out on the basis of the Malomuzh approach [13-15].
2.
Analysis
The complete system of hydrodynamics equations for the anisotropic liquid was
obtained in [11]. Let us assume that the liquid occupies the region between two parallel
planes separated by distance d, and the z axis of the Cartesian coordinate system to be
directed along the normal to the planes, so that 0 ≤ z ≤ d. For the sake of simplicity,
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 143–149
145
we consider the anisotropic liquid to be incompressible and ignore the rapid processes of
the internal angular momentum relaxation to equilibrium [11]. Now, let us linearize the
equation of motion for the velocity, pressure, and inertia tensor fluctuation fields, and
add to them spontaneous sources of the fluctuations. As a result, we obtain the system
of linked differential equations
1
∂
υα = − ∇α p + ν Δυα + ν12 ∇β δ Iαβ + σα ,
∂t
ρ0
1
∂
δ Iαβ = − δ Iαβ + D Δδ Iαβ + ν21 (∇α υβ + ∇β υα ) + Rαβ ,
∂t
τ
div υ = 0, νab = ηab/ρ0 , τ = ρ0/η2 , D = μ2/ρ0 .
(1)
Here ρ0 is the equilibrium density, ν is the kinematics viscosity, τ is the relaxation time,
D is the diffusion coefficient for the inertia tensor, δI is the deviation of the inertia tensor
from its equilibrium value, and σ and R are the spontaneous fluctuation sources. The
velocity field and the inertia tensor components, υα and δIαβ , vanish at the restrictive
planes:
υ (t, x, y, z = 0) = 0,
δIαβ (t, x, y, z = 0) = 0,
υ (t, x, y, z = d) = 0,
δIαβ (t, x, y, z = d) = 0.
(2)
In equations (1) and boundary conditions (2), let us pass to the Fourier–transforms with
respect to the time and the transversal radius-vector components:
Φ(t, r⊥ , z) =
e−i(ωt−k⊥r⊥ ) Φωk (z).
(3)
dω
4
dk⊥
Retaining the former designations for the amplitudes of the velocity and the inertia tensor
fields, and taking into account that the Laplace operator in representation (3) is of the
form
∂2
2
2
,
k⊥
= kx2 + ky2 ,
(4)
Δ = 2 − k⊥
∂z
after simple transformations we obtain these independent inhomogeneous equations:
1
fˆυα (z) = ν12 ∇β Rαβ + (−iω + − DΔ)σα ,
τ
Here Fαβ
fields
(5)
1
1
fˆ(−iω + − DΔ)δIαβ= = fˆRαβ + (Fαβ + Fβα ).
τ
2
is the result of transformation of the following combination of the spontaneous
Fαβ (t, r) = ν12 ν21 ∇β ∇β ′ Rαβ ′ + ν21 (
1
∂
+ − DΔ)∇α σβ ,
∂t τ
(6)
and the operator
ν12 ν21
1
Δ].
fˆ = [(−iω − νΔ)(−iω + − DΔ) −
τ
2
(7)
146
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 143–149
It is convenient to represent the solution of the ordinary differential equation system (5)
in the form of expansions in the harmonic functions. In view of the properties of these
functions and boundary conditions (2), these expansion assume the form
υα (z) =
Aαm sin μm z,
(8)
m
δIαβ (z) =
Bαβm sin μm z,
m
where the eigenvalues are μm = πm/d. Let us represent the components of the spontaneous sources in equations (5) in the form of similar expansions with the expansion
0
. With the orthogonality property and the normality condition for
coefficients A0αm , Bαβm
the harmonic functions, the amplitudes in (8) are simply expressed from (5) through the
expansion amplitudes for the spontaneous fields. Based on these expressions and using
the FDT, we find the spectral densities for the amplitudes of the velocity and inertia
tensor fluctuations:
2
+ μ2m )]−1 ,
A∗αm Aα′ m′ ω = Θδαα′ δmm′ Re[−iω + νωλ (k⊥
where
(9)
. ∗
1
2
+ μ2m )]−1 ,
Bαβm Bα′ β ′ m′ ω = 2Θδαα′ δββ ′ δmm′ Re[−iω + + D(k⊥
τ
k ? T0
1
ν12 ν21
2
[−iω + + D(k⊥
, νωλ = ν +
+ μ2m )]−1 .
(10)
2
2π
2
τ
The spectral density of the amplitudes for the inertia tensor fluctuations in (9) is obtained
under the assumption that the velocity field fluctuations and the inertia tensor field
fluctuations are independent. The difference of the result (9) from the those for the bulk
case is only in the discrete values of the wave number k and the appearance of a frequency
dependence of the kinematics viscosity coefficient.
The spectral densities of the thermal fluctuations of the Euler velocity and inertia
tensor hydrodynamic fields are respectively given by
Θ=
< υα∗ (r⊥ , z, t)υα′ (r′ ⊥ , z ′ , t′ ) >ω =
1
(2π)3
∗
(r⊥ , z, t)Iα′ β ′ (r′ ⊥ , z ′ , t′ ) >ω =
< Iαβ
&
1
(2π)3
′
dk⊥ eik⊥ (r⊥ −r ⊥ )
A∗αm Aα′ m′ ω sin μm z · sin μm′ z ′ ,
&
∗
′
Bαβm Bα′ β ′ m′ sin μm z · sin μm′ z ′ .
dk⊥ eik⊥ (r⊥ −r ⊥ )
m m′
m m′
ω
(11)
Let us explore the spectral density of the autocorrelation function for the velocity thermal fluctuations in more detail. In the limiting case of coincident spatial arguments, the
expression for the velocity CF from (11) reveals the logarithmic divergence ∼ ln ρ (ρ =
ρ→0
′
|r⊥ − r⊥
|) and, because of this, cannot be used to estimate the self-diffusion coefficient.
However, we can analyze the spectrum of the velocity autocorrelation function by changing from the Euler variables to the Lagrange ones. Moreover, since the velocity of the
Lagrange particle is usually considered to equal the collective component of the velocity
of the molecule [13, 14], the realization of such a change is actually equivalent to obtaining
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 143–149
147
the hydrodynamic asymptotic for the spectrum of the velocity autocorrelation function
for the velocity of the molecule.
To change to the LVACF, we take advantage of the Malomuzh approach [13], whose
essence consists in taking into account nonlocal relation between the Lagrange and Euler
variables. The derivation of the relation between the Lagrange and Euler velocity CFs is
given in [13]. It is considerably simplified if we ignore small drifts of the centre of mass
of the Lagrange particle, as in [12]; in this case, it assumes the form
1
(12)
ϕL (t) = 2 dr1 dr2 < υ (r(t) + r1 , t)υ (r(0) + r2 , 0) > .
VL
VL
VL
where VL is the volume of the Lagrange particle. The spectral density of the Lagrange
velocity hydrodynamic field has, according to (11), (12), and in view of the presence of
the delta-functions in the spectral densities of the velocity fluctuation amplitudes, the
form
∞
RL4 J12 (k⊥ RL )
2
ϕLω (z) = 2
< A∗αm Aαm >ω S̄m
(z)dk⊥ .
(13)
VL m
k⊥
0
Here J1 (y) is the Bessel function, RL =
2
(z)
S̄m
1
= 2
RL
3
V
4π L
1/3
, and
RL
RL
dz1 dz2 sin μm (z + z1 ) sin μm (z + z2 ) =
0
(14)
0
= (μm RL )−2 · (cos μm z − cos μm (z + RL ))2 .
After integration over the k⊥ and in view of (9), we get:
ϕLω (z) =
9Θ I1 (RL γml )K1 (RL γml ) 2
al
· S̄m (z).
16π 2 ν m l=1,2
(RL γml )2
(15)
Here I1 (y) and K1 (y) are the modified Bessel functions, and al and γml are the solutions
of these simple algebraic equations:
2
γm1
= μ2m + C1 ,
2
= μ2m + C2 ,
γm2
C1 = λ − λ2 +
C1 + C2 = 2λ,
a1 =
a2 =
ω 2 +iω/τ
,
Dν
λ=
C2 −C3
,
C2 −C1
C3 −C1
,
C2 −C1
−iω(D+ν)+ν/τ +ν12 ν21 /2
,
2Dν
(16)
C3 = D−1 (1/τ − iω) .
Since RL ≪ d, using the series expansion for the functions I1 (y) and K1 (y), result (15)
is represented in the form
9Θ
1
1
1 RL γml
−2
2
ϕLω (z) =
CL −
· S̄m
ln
+ (RL γml ) +
(z), (17)
al
32π 2 ν m l=1,2
2
2
2
4
148
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 143–149
where CL is the Euler constant.
After carrying out the summation, we finally derive:
/
0
1
s (2Cl )
2
9Θ
1
RL2 Cl
2s (Cl )
1
ϕLω (z) =
+
+ 3 +
CL −
ln
al
,
+
128π 2 ν l=1,2
RL
4
2RL
4
RL2 (2Cl )1/2 RL Cl RL4 Cl3/2
(18)
where
√
√ 2
√
s (x) = cth (d x) (ch ((z + RL ) x) − ch (z x)) +
√
√ 2
√
√
√
+ (sh ((z + RL ) x) + ch (z x)) − ch (RL x) (ch ((2z + RL ) x) + sh ((2z + RL ) x)) .
(19)
The derived spectrum of the LVACF allows us to find the expression for the Lagrange
self-diffusion coefficient, which is defined as
1
DL (z) = ϕL,ω=0 (z).
3
(20)
In the limit ν12 ν21 → 0, when the fluctuations of the inertia tensor and velocity fields are
independent, with the use of (15), (16) we obtain:
9Θ d
DL (z) =
2
64π 2 ν RL
/
z
d
/
1+o
RL
d
2 0
−
z 2
d
/
1+o
RL
d
2 0
RL
+
d
37 1
8πd
− ln
12 2
RL
RL
+o
d
(21)
Expression (21) is given with an accuracy of up to the second order in the small ratio
RL /d; depending on the distance d between the parallel planes, this relation takes on
values from −∞ to 10−1 .
Conclusion
The spectrum of the Lagrange velocity hydrodynamic field as a function of the
coordinate z normal to the restrictive planes is derived. It represents the hydrodynamic
asymptotic form for t → ∞ of the spectrum of the velocity autocorrelation function
for the anisotropic molecules of a liquid confined by two parallel planes. The distance
between the planes is a parameter for this dependence and can be taken to range from
d ∼ 10RL to d → ∞. With the use of the expression for the spectrum, the Lagrange
self-diffusion coefficient is determined as a function of z and dimensionless relation RL /d.
It is in good agreement with results [6], where the effect of confinement on the modecoupling contribution to the self-diffusion coefficient (in the direction parallel to the walls)
in a fluid slab was computed. In [6], these finite-size corrections are shown to reduce the
bulk diffusion constant near the walls by an amount of − (σ/d) log (σ/d), where d is the
thickness of the fluid slab and the physical interpretation of σ is given, based on intuitive
considerations, as the atomic size.
Similar finite-size corrections near the walls with RL instead of σ are obtained for the
Lagrange self-diffusion coefficient in the present paper. In the limit d → ∞, the results
obtained transform into the self-diffusion coefficient for the bulk case.
2 0
.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 143–149
149
References
[1] J. M. Ortiz de Zarate, L. M. Redondo (2001), Finite-size effects with rigid boundaries
on nonequilibrium fluctuations in a liquid. Eur. Phys. J., 21, 135-144.
[2] T. G. Sokolovska, R. O. Sokolovskii, M. F. Holovko (2000), Orientational ordering in
fluids with partially constrained molecule orientations. Phys. Rev., E62, N5, 6771-6779.
[3] I. Pagonabarraga, M. H. J. Hagen, With. P. Lowe, and D. Frenkel (1999), Shorttime dynamics of colloidal suspensions in confined geometries. Phys. Rev., E59, N4,
4458-4469.
[4] I. Pagonabarraga, M. H. J. Hagen, With. P. Lowe, and D. Frenkel (1998), Algebraic
decay of velocity fluctuations near a wall. Phys. Rev., E58, N6, 7288-7295.
[5] J. Teixeira, J.-M. Zanotti, M.-C. Bellissent-Funel, S.-H. Chen (1997), Water in confined
geometries. Physica, B234/236, 370-374.
[6] L.Bocquet, J.-L. Barrat (1996), Hydrodynamic properties of confined fluids. J. Phys:
Condens. Matter., 8, 9297-9300.
[7] K. P. Travis, B. D. Todd, D. J. Evans (1997), Departure from the Navier-Stokes
hydrodynamics in confined liquid. Phys. Rev., 55, N4, 4288-4295.
[8] F. Benmouna and D. Johannsmann (2002), Hydrodynamic interaction of AFM
cantilevers with solid walls: An investigation based on AFM noise analysis. Eur. Phys.
J., E9, 435-441.
[9] M.A. Leontovich (1941), Relaxation in liquids and scattering of light. J. Phys. USSR,
4, 499-518.
[10] I.L. Fabelinski (1997), Spectra of molecular scattering of light. Progress in Optics,
37, 97-184.
[11] A.V. Zatovsky, A.V. Zvelindovsky (2001), Hydrodynamic fluctuations of a liquid with
anisotropic molecules. Physica A, 298, 237-254.
[12] O.I. Sakhnenko (2005), The spectrum of the correlation function for fluctuatios of
the anisotropy tensor of a Lagrangian liquid particle. Ukr. J. Phys., 50, N7, 714-719.
[13] T. V. Lokotosh, N. P. Malomuzh (2000), Lagrange theory of thermal hydrodynamic
fluctuations and collective diffusion in liquids. Physica A., 286, 474-488.
[14] T. V. Lokotosh, N. P. Malomuzh (2001), Manifestation of the Collective Effects in
the Rotational Motion of Molecules in Liquids. J. Mol. Liq., 93, N1-3, 95-108.
[15] T. V. Lokotosh, N. P. Malomuzh, K. S. Shakun (2002), Nature of oscillations for the
autocorrelation functions for translational and angular velocities of a molecule. J. Mol.
Liq., 96/97, 245-263.
EJTP 3, No. 11 (2006) 151–158
Electronic Journal of Theoretical Physics
On the Quantum Correction of Black Hole
Thermodynamics
Kourosh Nozari and S. Hamid Mehdipour
∗
Department of Physics, Faculty of Basic Sciences,
University of Mazandaran,
P. O. Box 47416-1467, Babolsar, IRAN
Received 20 May 2006 , Accepted 7 June 2006, Published 25 June 2006
Abstract:
Bekenstein-Hawking Black hole thermodynamics should be corrected to
incorporate quantum gravitational effects. Generalized Uncertainty Principle(GUP) provides a
perturbational framework to perform such modifications. In this paper we consider the most
general form of GUP to find black holes thermodynamics in microcanonical ensemble. Our
calculation shows that there is no logarithmic pre-factor in perturbational expansion of entropy.
This feature will solve part of controversies in literatures regarding existence or vanishing of
this pre-factor.
c Electronic Journal of Theoretical Physics. All rights reserved.
Keywords: Quantum Gravity, Generalized Uncertainty Principle, Black Holes Thermodynamics
PACS (2006): 04.70.-s, 04.70.Dy
1.
Introduction
Quantum geometry, string theory and loop quantum gravity all indicate that measurements in quantum gravity should be governed by generalized uncertainty principle[15]. As a result, there is a minimal length scale of the order of Planck length which
can not be probed. In the language of string theory, this is related to the fact that a
string can not probe distances smaller than its length. Therefore, it seems that a reformulation of quantum theory to incorporate gravitational effects from very beginning is
necessary to investigate Planck scale physics. Introduction of this idea, has drawn considerable attentions and many authors have considered various problems in the framework
of generalized uncertainty principle[6-20]. Quantum gravitational induced corrections to
black hole thermodynamics as a consequence of GUP are studied with details in liter∗
knozari@umz.ac.ir
152
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 151–158
atures. Adler and his coworkers[21] have argued that contrary to standard viewpoint,
GUP may prevent small black holes total evaporation in exactly the same manner that
the ordinary uncertainty principle prevents the Hydrogen atom from total collapse. They
have considered these black holes remnants as a possible source of dark matter. Medved
and Vagenas[22], have recently formulated the quantum corrected entropy of black holes
in terms of an expansion and have claimed that this expansion is consistent with all
previous findings. Bolen and Cavaglia, have obtained thermodynamical properties of
Schwarzschild anti-de Sitter black holes using GUP [23]. They have considered two limits
of their equations, quantum gravity limit and usual quantum mechanical regime and in
each circumstances they have interpreted their results. Action for the exact string black
hole has been considered by Grumiller and he has found exact relation for entropy of a
string black hole[24]. Existence or vanishing of logarithmic prefactor in the expansion of
black hole entropy has been considered in details by Medved. He has argued in [25] that
”the best guess for the prefactor might simply be zero” regarding to mutual cancelation of
canonical and microcanonical contributions. But later, considering some general considerations of ensemble theory, he has argued that canonical and microcanonical corrections
could not cancel each other to result in vanishing logarithmic pre-factor in entropy[26].
Meanwhile, Hod has employed statistical arguments that constrains this prefactor to be a
non-negative integer[27]. There are other literatures considering logarithmic corrections
to black hole entropy[28,29], but there is no explicit statement about the ultimate value
of this prefactor.
Here, using generalized uncertainty principle in its most general form as our primary
input, we find explicit perturbational expansion of black hole entropy in microcanonical
ensemble. By computing the coefficients of this expansion, we will show that there is no
logarithmic prefactor in expansion of microcanonical entropy.
2.
Generalized Uncertainty Principle
Usual uncertainty principle of quantum mechanics, the so-called Heisenberg uncertainty principle, should be re-formulated regarding to non-commutative nature of spacetime. It has been indicated that in quantum gravity there exists a minimal observable
distance on the order of the Planck length which in the context of string theories, this
observable distance is referred to GUP[1-5],[30-33]. A generalized uncertainty principle
can be formulated as
+ const.Gδp,
(1)
δx ≥
2δp
which, using string theoretical arguments regarding the minimal nature of lp [4], can be
written as
δp
+ α2 lp2
(2)
δx ≥
2δp
2
The corresponding Heisenberg commutator now becomes,
[x, p] = i(1 + α′ p2 ).
(3)
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 151–158
153
Note that commutator (3) is not the direct consequence of relation (2), but can be
considered as one of its consequences[11]. α is positive and independent of δx and δp but
may in general depend to the expectation values of x and p. In the same manner one can
consider the following generalization,
β2
2
δxδp ≥
(4)
1 + 2 (δx) ,
2
lp
which indicates the existence of a minimal observable momentum. It is important to note
that GUP itself can be considered as a perturbational expansion[11]. In this viewpoint,
one can consider a more general statement of GUP as follows
α2 lp2
β2
2
2
(5)
δxδp ≥
1 + 2 (δp) + 2 (δx) + γ ,
2
lp
where α, β and γ are positive and independent of δx and δp but may in general
depend
to the expectation values of x and p. Here, Planck length is defined as lp = G
. Note
c3
that (5) leads to nonzero minimal uncertainty in both position (δx)min and momentum
(δp)min . In which follows, we use this more general form of GUP as our primary input
and construct a perturbational framework to find thermodynamical properties of black
hole and their quantum gravitational corrections. It should be noted that since GUP
is a model independent concept[6], the results which we obtain are consistent with any
promising theory of quantum gravity.
3.
Black Holes Thermodynamics
Consider the most general form of GUP as equation (5). A simple calculation gives,
"
#
2 (γ + 1)
lp2 δp
δx 2 1 ± 1 − β 2 α2 + 2
.
(6)
β
lp (δp)2
Here, to achieve standard values (for example δxδp ≥ ) in the limit of α = β = γ = 0,
we should consider the minus sign. One can minimize δx to find
(δx)min ±αlp
(1 + γ)
.
1 − α2 β 2
(7)
The minus sign, evidently has no physical meaning for minimum of position uncertainty.
Therefore, we find
(δx)min αlp
(1 + γ)
.
1 − α2 β 2
(8)
This equation gives the minimal observable length on the order of Planck length. Since in
our definition, α and β are dimensionless positive constant always less than one(extreme
quantum gravity limit), (δx)min is defined properly. Equation (5) gives also
"
#
2 (γ + 1)
l
δx
p
δp 2 2 1 ± 1 − α2 β 2 +
.
(9)
α lp
(δx)2
154
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 151–158
Here to achieve correct limiting results we should consider the minus sign in round bracket.
From a heuristic argument based on Heisenberg uncertainty relation, one deduces the
following equation for Hawking temperature of black holes[21],
TH ≈
δpc
2π
(10)
Based on this viewpoint, but now using generalized uncertainty principle in its most
general form, modified black hole temperature in GUP is,
"
#
2 (γ + 1)
l
cδx
p
THGU P ≈
.
(11)
1 − α2 β 2 +
2 2 1−
(δx)2
2πα lp
Now consider a quantum particle that starts out in the vicinity of an event horizon and
then ultimately absorbed by black hole. For a black hole absorbing such a particle with
energy E and size R, the minimal increase in the horizon area can be expressed as [34]
(ΔA)min ≥
8πlp2 ER
,
c
(12)
then one can write
8πlp2 δpcδx
,
(13)
c
where E ∼ cδp and R ∼ δx. Using equation (9)(with minus sign) for δp and defining
A = 4π( δxmin
)2 , we find
2
(ΔA)min ≥
(ΔA)min
"
8A
2 1−
α
1 − α2
πlp2 (γ + 1)
β2 +
A
#
.
(14)
Now we should determine δx. Since our goal is to compute microcanonical entropy of a
large black hole, near-horizon geometry considerations suggests the use of inverse surface
gravity or simply twice the Schwarzschild radius for δx. Therefore, δx ≈ 2rs and defining
4πrs2 = A and (ΔS)min = b = constant, then it is easy to show that,
(ΔS)min
dS
dA
(ΔA)min
"
8A 1 −
bα2
2
1 − α β2 +
πlp2 (γ+1)
A
# .
(15)
Three point should be considered here. First note that b can be considered as one bit
of information since entropy is an extensive quantity. Considering calibration factor of
Bekenstein as ln2
, the minimum increase of entropy(i.e. b), should be ln2. Secondly,
2π
(∆S)min
dS
note that dA (∆A)min holds since this is an approximate relation and give only relative changes of corresponding quantities. As the third remarks, our approach considers
microcanonical ensemble since we are dealing with a Schwarzschild black hole of fixed
mass. Now we should perform integration. There are two possible choices for lower limit
of integration, A = 0 and A = Ap . Existence of a minimal observable length leads
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 151–158
155
2
to existence of a minimum event horizon area, Ap = 4π (δx)2min . So it is physically
reasonable to set Ap as lower limit of integration. This is in accordance with existing
picture[21]. Based on these arguments, we can write
S
A
Ap
bα2
1 − α2 β 2 +
"
8A 1 −
πlp2 (γ+1)
A
# dA.
(16)
Integration gives,
#
"
−2$A(ζA + η) + A(ζ + 1) + η $
η + 2ζA + 2$ζA(ζA + η)
$
S μ ln $
+ ζ ln
η + 2ζAp + 2 ζAp (ζAp + η)
−2 Ap (ζAp + η) + Ap (ζ + 1) + η
(17)
where,
b
μ = 2,
η = −πα2 lp2 (γ + 1),
ζ = 1 − α2 β 2 ,
8β
Ap =
πα2 lp2 (1 + γ)
(1 − α2 β 2 )
(18)
This is the most general form of the black hole entropy which can be obtained from
perturbational approach based on GUP.
Expansion of (17) gives
∞
Dn (A − Ap )n .
(19)
S
n=1
The coefficients of this expansion have very complicated form. The first coefficient is
D1 = μ
/
− √ 2ζAp +η
Ap (ζAp +η)
+ζ +1
$
+
−2 Ap (ζAp + η) + Ap (ζ + 1) + η
$
√2ζ
2 A +ζη
p
ζAp (ζAp +η)
+ 2ζ
0
. (20)
ζ $
2 ζAp (ζAp + η) + 2ζAp + η
The matter which is important in our calculations is the fact that expansion (19) has no
logarithmic term. In other words, since expansion (19) contains only integer power of
A − Ap , we conclude that in microcanonical ensemble, there is no logarithmic corrections
due to quantum gravitational effects for thermodynamics of black holes. Adler et al have
found vanishing entropy for remnant in their paper[21]. In other words, their result for
entropy vanishes when one considers Planck mass limit. In our framework, when A = Ap ,
one finds S = 0 and therefore remnant has zero entropy. A result which physically can
be acceptable since small classical fluctuations are not allowed at remnant scales because
of the existence of the minimum length.
4.
Summary
In this paper, using generalized uncertainty principle in its most general form as
our primary input, we have calculated microcanonical entropy of a black hole. We have
156
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 151–158
shown that in perturbational expansion there is no logarithmic pre-factor, which has been
the source of controversies in literatures. Actually in calculation of entropy we should
compute the number of possible microstates of the system and there are two possible
choices for corresponding ensemble: canonical and microcanonical ensemble. We have
shown that the contribution of microcanonical ensemble itself is vanishing. If there is
any contribution related to canonical ensemble, it cannot cancel vanishing contribution
of microcanonical ones. This argument resolves part of controversies regarding mutual
cancelation of two contributions as have been indicated in introduction.
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 151–158
157
References
[1] G. Veneziano, Europhys. Lett. 2 (1986) 199;
[2] D. Amati, M. Ciafaloni and G. Veneziano, Phys. Lett. B216 (1989) 41;
D. Amati, M. Ciafaloni and G. Veneziano, Phys. Lett. B197 (1987) 81; Int. J. Mod.
Phys. A3 (1988) 1615; Nucl. Phys. B347 (1990) 530.
[3] D.J. Gross and P.F. Mende, Phys. Lett. B197 (1987) 129; Nucl. Phys. B303 (1988)
407.
[4] K. Konishi, G. Paffuti and P. Provero, Phys. Lett. B234 (1990) 276;
R. Guida, K. Konishi and P. Provero, Mod. Phys. Lett. A6 (1991) 1487;
M. Kato, Phys. Lett. B245 (1990) 43.
[5] S. Capozziello, G. Lambiase, G. Scarpetta, Int.J.Theor.Phys. 39 (2000) 15-22.
[6] S. Hossenfelder et al, Phys. Lett. B 575 (2003) 85-99
[7] L. J. Garay, Int. J. Mod. Phys. A10 (1995) 145.
[8] M. Maggiore, Phys. Lett. B304 (1993) 65.
[9] C. Castro, Found.Phys.Lett. 10 (1997) 273-293 .
[10] M. Maggiore, Phys. Lett. B319 (1993) 83.
[11] A.Kempf, et al., Phys. Rev. D52 (1995) 1108
[12] M. Maggiore, Phys. Rev. D49 (1994) 5182.
[13] M. R. Setare, Phys.Rev. D70 (2004) 087501.
[14] S. Kalyana Rama, Phys.Lett. B519 (2001) 103-110 .
[15] A. Camacho, Gen.Rel.Grav. 35 (2003) 1153-1160.
[16] P. Chen, R. J. Adler, Nucl.Phys.Proc.Suppl. 124 (2003) 103-106.
[17] F. Scardigli, R. Casadio, Class.Quant.Grav. 20 (2003) 3915-3926.
[18] S. Hossenfelder, Mod.Phys.Lett. A19 (2004) 2727-2744.
[19] S. Hossenfelder, Phys.Rev. D70 (2004) 105003
[20] S. Hossenfelder, Mod.Phys.Lett. A19 (2004) 2727-2744
[21] R. J. Adler, P. Chen, D. I. Santiago, Gen. Rel. Grav. 33 (2001) 2101.
[22] A. J. M. Medved and E. C. Vagenas, Phys.Rev. D70 (2004) 124021
[23] B. Bolen and M. Cavaglia, arXiv: gr-qc/0411086 v1
[24] D. Grumiller, arXiv: hep-th/0501208, v3, Liepzig University-ITP-2005/001
[25] A. J. M. Medved, arXiv: gr-qc/0411065
[26] A. J. M. Medved, Class.Quant.Grav. 22 (2005) 133-142
[27] S.Hod, Class. Quant.Grav., Letter to the Editor. Vol. 21, L97 (2004)
[28] G. Gour and A. J. M. Medved, Class. Quant. Grav. 20 (2003) 3307
[29] A. J. M. Medved, Class. Quant. Grav. 20 (2003) 2147; Class. Quant. Grav. 19 (2002)
2503.
158
Electronic Journal of Theoretical Physics 3, No. 11 (2006) 151–158
[30] M. Lubo, Phys.Rev. D61 (2000) 124009
[31] S. Benczik et al, Phys.Rev. D66 (2002) 026003
[32] L. N. Chang et al, Phys.Rev. D65 (2002) 125027
[33] I. Dadic et al, Phys.Rev. D67 (2003) 087701
[34] D. Christodoulou and R. Ruffini, Phys. Rev. D4 (1971) 3352