EVOLUTION TOWARDS ERGODIC BEHAVIOR OF STATIONARY FRACTAL RANDOM-PROCESSES WITH MEMORY - APPLICATION TO THE STUDY OF LONG-RANGE CORRELATIONSOF NUCLEOTIDE-SEQUENCES IN DNA
Mo. Vlad et al., EVOLUTION TOWARDS ERGODIC BEHAVIOR OF STATIONARY FRACTAL RANDOM-PROCESSES WITH MEMORY - APPLICATION TO THE STUDY OF LONG-RANGE CORRELATIONSOF NUCLEOTIDE-SEQUENCES IN DNA, Physica. A, 229(3-4), 1996, pp. 312-342
The possible occurrence of ergodic behavior for large times is investi
gated in the case of stationary random processes with memory. It is sh
own that for finite times the time average of a state function is gene
rally a random variable and thus two types of cumulants can be introdu
ced: for the time average and for the statistical ensemble, respective
ly. In the limit of infinite time a transition from the random to the
deterministic behavior of the time average may occur, resulting in an
ergodic behavior. The conditions of occurrence of this transition are
investigated by analyzing the scaling behavior of the cumulants of the
time average. A general approach for the computation of these cumulan
ts is developed; explicit computations are presented both for short an
d long memory in the particular case of separable stationary processes
for which the cumulants of a statistical ensemble can be factorized i
nto products of functions depending on binary time differences. In bot
h cases the ergodic behavior emerges for large times provided that the
cumulants of a statistical ensemble decrease to zero as the time diff
erences increase to infinity. The analysis leads to the surprising con
clusion that the scaling behavior of the cumulants of the time average
is relatively insensitive to the type of memory considered: both for
short and long memory the cumulants of the time average obey inverse p
ower scaling laws. If the cumulants of a statistical ensemble tend tow
ards asymptotic values different from zero for large time differences,
then the time average is random even as the length of the total time
interval tends to infinity and the ergodic behavior no longer holds. T
he theory is applied to the study of long range correlations of nucleo
tide sequences in DNA; in this case the length t of a sequence of nucl
eotides plays the role of the time variable. A proportionality relatio
nship is established between the cumulants of the pyrimidine excess in
a sequence of length t and the cumulants of the time (length) average
of the probability of occurrence of a pyrimidine. It is shown that th
e statistical analysis of the DNA data presented in the literature is
consistent with the occurrence of the ergodic behavior for large lengt
hs. The implications of the approach to the analysis of the large time
behavior of stochastic cellular automata and of fractional Brownian m
otion are also investigated.