A central limit theorem for measurements on the logarithmic scale and its application to dimension estimates

We show consistency and asymptotic normality of certain estimators for expected exponential growth rates under i.i.d. observations. These statistical functionals are of the formT(F)=∫log∫h(x,y)F(dx)F(dy)T(F)=∫log∫h(x,y)F(dx)F(dy)and are applicable to dimension estimates (information dimension), entropy estimates and estimations of the growth rate of “generating” functions. We also give an affirmative answer to a question posed by Keller in 1997 [A new estimator for information dimension with standard errors and confidence intervals, Stochastic Process. Appl. 71(2):187–206] whether this estimator, specialized for dimension, is an alternative to standard procedures.