PARTICLES
V.M.SOMSIKOV
Ionosphere institute, Alma-Ata, Kazakhstan
It is short explained, how it is possible to expand the classical mechanics within the limits of its laws
that it was applicable for the description of irreversible processes in the nature.
ҚҰРЫЛЫМДЫ БӨЛШЕКТЕР МЕХАНИКАСЫНДАҒЫ
ДЕТЕРМИНДАЛҒАН ҚАЙТЫМСЫЗДЫҚ
В.М.Сомсиков
Ионосфера институты, Алматы, Қазақстан
Қысқа әрі нұсқа классикалық механиканы сол өзінің заңдарын қолдана отырып табиғаттағы
қайтымсыз процесстерді сипаттауға қалай қолдануға болатындығы түсіндірілген.
Журнал проблем эволюции открытых систем
Вып. 12, Т.1, 2010 8
CRITICAL STATISTICAL ENSEMBLES IN ELECTRON NANOSYSTEMS
AT THE LOCALIZATION TRANSITION
I.Kh. Zharekeshev
Al-Farabi Kazakh National Universiiy, Almaty
The statistical properties of spectra in the electron nanostuctures are studied with
and without magnetic field. It is shown that the spectral correlations exhibit
universal scale-independent behaviour characteristic of critical statistical
ensembles
Introduction
The crossover from integrability to chaos
is one of the important issues in physics of
complex systems. Non-ergodicity and nonlinear
behavior of open systems out of equilibrium is
often caused by the influence of the external
conditions, for example, by the adiabatic
connection to the environment or by imposing
strong external fields. The description of
dynamical aspects of time-dependent evolution
of statistical parameters can be based on the
formalism of canonical ensembles (when the
number of particles N in an open nanosystem is
fixed) and grand-canonical ensembles (when N
is large and not fixed). As a counterpart for the
transition
between
full
integrability
and
complete chaos in quantum electronic systems
(e.g. quantum dots coupled to a bath reservoir)
can serve a delocalization-localization transition.
On the other hand, the latter is
characterized by a sharp crossover of the
electronic conductivity from the metallic regime
to the insulating one. A striking signature of this
metal-insulator transition is a presence of the
criticality,
meaning
that
a
set
of
thermodynamical quantities exhibit critical
behavior. It turns out, that the distributions of
strongly fluctuating parameters at the critical
point of the transition obey generic common
laws and can be analyzed by the finite-size
scaling scenarios. Moreover, a new set of the
statistical ensembles, named critical ones, have
been introduced especially for characterizing the
critical point.
The problem of localization of quantum
particles in a disordered nanosystem has
attracted great attention during last decades,
triggered by the discovery of new quantum
phenomena in condensed matter physics. After
the first formulation of the modern theory of
solids in the 1930’s it was thought for a long
time that the effect of disorder on the state of
electrons in solid structures can be described in
terms of the perturbations of low orders. Indeed,
the
standard
quantum-mechanical
implementation
of
Boltzmann
transport
approach
appeared
to
work
perfectly
everywhere, expect for certain unexplainable
observations like, for example, the negative
magnetoresistance of doped semiconductors in a
weak magnetic field. It was only in 1958 that
Philipp Anderson suggested the currently well-
known concept of quantum localization of
electrons by the potential disorder [1]. He
proposed that in a tight-binding model of
electrons on a lattice with chaotically varying
site energies V the electrons of a given energy E
would become localized if the spread of the on-
site energies V meaning the disorder degree is
sufficiently large.
In the other words, the behavior of the
electronic states would change drastically from
extended to localized behavior. In the former
case the disorder manifests itself mainly through
a decay of phase coherence in the averaged one-
particle propagator, while in the latter case the
probability amplitude decreases exponentially as
one goes away from the centre of localization. It
is not surprising, since even in a classical system
the disorder may cause the localization of
particles. However the quantum nature of
electrons make them harder to overcome narrow
passages and channels, despite the fact that
Журнал проблем эволюции открытых систем
9 Вып. 12, Т.1, 2010
quantum particles can tunnel under barriers, i.e.
through classically forbidden regions. Hence,
quantum particles in nanoclusters tend to be
localized more easily than their classical
counterparts.
Quantum transport in open nanosystems
The first qualitative consequence of the
quantum nature of particles for transport in
disordered open nanosystems is reflected in the
fact that the mean free path l cannot become
shorter that the wavelength of the electrons λ
F
.
For electrons in the centre of the band λ
F
is of
the order of the lattice spacing a, but for energies
near the band edge λ
F
may be larger than lattice
constant a. In a classical system the shortest
possible mean free path is always given by the
average distance between the scattering centers,
irrespectively of the particle energy. If the
disorder increases beyond the point where l≈ λ
F
or else, if the energy of the electrons decreases
for a fixed disorder, the nature of the electronic
states is expected to change from extended to
localized. As a consequence, the electrical
conductivity of the nanosystem or mobility of
carriers is expected to vanish to zero.
This scenario, where a change in the
electron energy induces a metal-insulator
transition, was explored early by Mott [2]. He
introduced the term ‘mobility edge’ for the
critical energy separating extended and localized
states. These two types of states are not likely to
coexist at a given energy, since any small change
in the potential would cause admixtures of
extended states with a localized state, and would
thus delocalize it. On the basis of qualitative
considerations and experimental data Mott
concluded that the Anderson transition in a
three-dimensional
system
should
be
discontinuous, the conductivity jumps from a
finite
value,
called
minimum
metallic
conductivity downwards to zero. The advent of
the
computer,
which
made
the
exact
diagonalization of finite-size systems possible,
and the advances in electrical transport
measurements near the metal-insulator transition
at low temperatures have changed this picture.
It is well established now that there is, in
fact, no minimum metallic conductivity and that
the transition is continuous, much like a
continuous
phase
transition
in
usual
thermodynamics. A continuous phase transition
is necessarily associated with a characteristic
length ξ, localization length, which tends to
infinity as the transition is approached. At the
transition a natural unit of length does not exist
anymore and the system is therefore considered
as being scale-invariant. The ensuing scaling
behavior was discovered by Thouless [3] who
noticed that the conductance of a finite-size
block scales with the size in a universal way.
In this paper we also consider the
electronic spectra rather than the conductivity.
Its statistics also undergo the phase transition
similar to the conductance. Mostly concentrated
on the critical point of the metal-insulator
transition, the various classes of the ensembles
are numerically investigated depending on the
presence of the external magnetic field. It has
earlier shown that the probability function of the
neighboring spacings P(s) exhibits finite-size
scaling and also becomes scale-invariant exactly
at the transition point for both the orthogonal
and the unitary symmetry. Interestingly, our
results demonstrate that there is a essential
difference between the spectral correlations for
critical orthogonal ensemble (COE) and for the
Gaussian orthogonal ensemble (GOE) [4]. The
same is observed for the deviations between
critical unitary (CUE) and Gaussian unitary
ensembles (GUE).
Critical orthogonal ensemble
For
the
orthogonal
case,
which
corresponds to the system without magnetic field
and with spinless particles, the level statistics at
non-vanishing disorder of random potential
exhibits critical behaviour starting from the three
dimensions.
As
an
example,
we
have
numerically calculated the level statistics of a
Журнал проблем эволюции открытых систем
Вып. 12, Т.1, 2010
10
sample with time-reversal symmetry ( φ= 0) for L =
6 for different disorders. Figure 1 demonstrates how
the distribution P(s) changes from the GOE-result
to the Poisson distribution, when the disorder W
increases. One observes a continuous change of the
data between the two limits, the slope at small s
being always linear for arbitrary disorder W in
agreement with the time-invariant symmetry:
,
)
(
s
B
s
P
o
o
(1)
with β = 1. In the metallic regime ( W < W
c
) it
diminishes towards B
o
= π
2
/6 with decreasing
disorder or/and increasing the size, i.e. with
increasing the conductance g, unless the system is
in the ballistic regime. In the insulating regime the
slope B
o
increases to infinity when both the disorder
W and the linear size of the system L tend to
infinity.
Fig.1.The level spacing distribution P(s) for a
cubic nanosystem with orthogonal symmetry for
various disorder strength W=5, 16.5, 30 и 100
(box distribution). Taken is a nanocube of linear
size L=16. Solid line is the GOE result and
dashed line is Wigner surmise. Dotted line:
Poisson distribution P
p
( s) = exp(- s). Inset shows
an enlarged area near s = s
0
≈2.0, the crossing
point of the Poisson and the Wigner
distributions. The number of spacings is about
10
8
.
At the transition point W = W
c
= 16.5 of the
`box' model the level spacing distribution function
corresponds to the critical orthogonal probability
function P
o
c
( s), which is not sensitive to the change
of the system size L. The prefactor is found to be
B
o
c
= 2.12
0.06, i.e. B
o
c
≈1.29 B
o
, according to the
Eq. (1). The scale-invariance of the critical level
statistics has been justified in many works [5-8].
We analyze here the overall functional form of the
critical P
o
c
( s), in more detail using results of large-
scale computations, particularly concentrating on
the asymptotic behaviour for large s.
We have computed also the level spacing
distribution function (in form of a histogram) for the
critical disorder W
c
= 20.9 of the Gaussian
distribution of the on-site energies. The eigenvalues
were taken from the interval centred at E = 0
containing approximately 10% of the spectrum.
Within the numerical error-bars the data of P(s) are
the same for different L and coincide with those of
P
o
c
( s) for the box distribution, justifying the
independence on the model of diagonal disorder. By
other words, it is proved that the critical statistics
are universal irrespective of the microscopic details
of the system.
Of particular interest is the region of spacings
around s
0
= 2.002, where the Wigner surmise
P
W
( s)= π
2
/2 exp(- π
2
s
2
/4), and P
P
( s) = exp(- s),
intersect. It has been suggested by Shklovskii [5]
that independently of the disorder degree W, all
empirical curves P(s) including the critical one,
P
o
c
( s) should intersect at the same point s
0
, which
would then play the role of a universal energy. The
existence of such a universal point would mean that
the system possess a hidden symmetry. The
underlying reason for this symmetry, however, is
not known yet at present.
Focusing on a region close to s
0
, we have
performed detailed calculations with the large
number of realizations. One should again take into
account that, in fact, the Wigner surmise only
approximates the exact RMT result for P(s), albeit
quite well (within 5%). Thus P
GOE
( s)≠ P
W
( s). The
true intersection of P
P
( s), and P
GOE
( s), lies at s
0
=
2.019. Careful analysis of our data does not,
however, show any common crossing point for
various disorder degrees W (see inset of Fig. 1).
The computed value of P( s
0
) for disorder W = 16.5
(also for W = 30 and 100) differs from P
P
( s
0
) by a
magnitude which exceeds the numerical errors. No
unique point so has been observed also for the
unitary case.
Журнал проблем эволюции открытых систем
]
11
Вып. 12, Т.1, 2010
In the vicinity of the transition point W
c
the
level spacing distribution exhibits critical behaviour
similar to that of the level number variance,
discussed in papers [6-8]. Using the finite-size
scaling analysis for the distribution of neighbouring
spacings one can construct the disorder dependence
of the localization length ξ( W) and extract the
critical exponent ν. The one-parameter scaling
scenario for the function P(s) has been corroborated
in a large number of computer simulations [8-10].
Therefore we provide here only the summary of
numerical results available on calculations of the
critical exponent for various physical situations and
do not further focus on this topic.
One can show that the critical exponent as a
characteristic of the symmetry class, is almost the
same within statistical uncertainty for different
models of the diagonal disorder (for the uniform
and the Gaussian distributions of on-cite energies
ε
n
) and of the quantum percolation. As expected, it
is also not sensitive to the anisotropy of the system.
Generally, all the data for the critical exponent of
the localization length are centred about the value ν
= 1.45 ± 0.1, as an acceptable estimate for the
orthogonal case. It turns out that the value of the
critical exponent evaluated from the level number
variance is consistent with that obtained from the
analysis of P(s). Notably this value is somewhat
smaller than that found by the high-precision
transfer-matrix calculations ν
TM
= 1.57 ± 0.02. The
reason of such a slight, but resolvable discrepancy
is not known yet.
Critical unitary ensemble
We now consider the level statistics for
systems with broken time reversal symmetry. Such
a situation can be realized by applying an external
Aharonov-Bohm (AB) magnetic flux through a
system forming a ring geometry. In order to achieve
the maximal effect of the change of the symmetry,
the AB-fluxes of the equal magnitude are applied
along all three perpendicular directions in a three-
dimensional cubic lattice (the three-component
flux). Performing diagonalization for different
magnitudes of the flux ranging from φ = 0 to φ =
1/4, we found the critical statistics to be sensitive to
the flux. As a function of φ, the distribution P(s) at
the critical point W
c
= 16.5 changes smoothly from
the critical orthogonal P
o
c
( s) to the critical unitary
form P
u
c
( s) at φ = 1/4, which is known as the COE-
CUE crossover [8] (see figure 2).
Fig.2.The level spacing distribution P(s) for the
unitary case at the critical disorder W=16.5.
Upper panel: at different values of the
Aharonov-Bohm flux φ for a system with fixed
linear size L = 5. Lower panel: at the fixed flux φ
= 0.2 for different sizes L = 5 (+); 10 (□); 20 (●).
Solid and dashed lines are the critical P(s) for
two limiting phases φ = 1/4 and φ = 0,
respectively. Inset: asymptotic behaviour of P(s)
at the limiting AB-phase φ = 1/4 for linear sizes
L = 5 (+) and 20 (●). Solid line – P
u
c
( s) [Eq.
(3)], dashed line - P
P
( s) and dotted line – P
u
W
( s)
all correspond to critical, insulating and metallic
regimes, respectively.
This flux-controlled crossover of the critical level
statistics
repeats
periodically,
resuming
the
orthogonal form at φ=0.5, since a half of the flux
quantum corresponds to the real Hamiltonian
defined with antiperiodic boundary conditions
Журнал проблем эволюции открытых систем
Вып. 12, Т.1, 2010
12
(`false' T-invariance). For fixed flux, all P(s) at W
c
proved to be insensitive to variations of the size of
the system L = 5, 10 and 20, as shown in figure 2
(left panel) for φ = 0.2. The same L-independent
behaviour has been observed for other values of
flux φ = 0.05, 0.1 and 0.25. As expected for the T-
symmetry broken, we observe a quadratic
behaviour P
u
( s) ~ s
2
for small s at all W, as soon as
L is finite. It has recently been suggested that the
flux-driven crossover of the critical unitary
statistics can be explained on the basis of the
analogy to the semi-classical limit. For n = 0 both
the Mehta parameter [9] for the unitary critical
statistics J
c
u
= 0.685±0.003 and, consequently, the
spacing variance var s = 2 J
c
u
-1 ≈ 0.344 are larger
than those for the GUE, where J
u
for RMT is equal
to 0.590.
The extreme form of the unitary critical
P
u
c
( s) corresponding to the AB-phase φ = 1/4
coincides with that found in the presence of the
strong magnetic field, as has been shown in Ref.
[10]. In the latter case, the COE-to-CUE crossover
is discontinuous, unlike to the application of an
AB-flux. The behaviour of P
u
c
( s) for small
spacings s is described by the power-law
,
)
(
s
B
s
P
u
с
u
с
,
1
.
0
5
.
8
u
с
B
(2)
with the `repulsion parameter'
=2. This result is
in direct contradiction to another numerical work
[11], which has claimed the linear growth of P
u
c
( s)
at small s, i.e.
= 1, similar to the critical
orthogonal case.
The asymptotic form of the size-invariant
P
u
c
( s) for large s can be well approximated by the
simple exponential decay [11]
),
exp(
)
(
s
A
D
s
P
u
c
u
с
u
с
,
1
.
0
85
.
1
u
с
A
(3)
that is slower compared to the Gaussian tail
characteristic of the RMT-result. On the other
hand, it is similar to the Poisson decay valid in the
strongly localized regime, although the decay rate
A
u
c
is certainly larger than unity. Note that the
critical prefactor B
u
c
= 2.59 B
u
RMT
is also markedly
larger than that of the GUE. In fact, the value A
u
c
≈1.85 should be considered as a lower bound for
the exponential decrease of P
u
c
( s) because obtained
range of s is not sufficiently large. It was
demonstrated
in
[12]
that
the
similar
asymptotic behaviour holds also for systems
with time-reversal symmetry.
Достарыңызбен бөлісу: |