... function3.1
In [Grassberger, 1986,Broggi, 1988], $k$ is kept constant so that the correction function is a constant. It should be investigated how the correction function can be incorporated.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... unclear3.2
Somorjai [Somorjai, 1986] suggests to use expressions derived by Fukunaga and Hostetler [Fukunaga and Hostetler, 1973] for the choice of $\alpha$ and $\beta$. However, here we do not use these expressions since for $D=2$, $\alpha=0$ and the dimension of the Rössler attractor is very close to 2.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... found5.1
With data from an analog-to-digital converter (ADC) one could use $r_{min} = 2^{\mbox{-number of bits}}$.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... spectrum7.1
The power spectrum was computed using NAG [Numerical Algorithms Group, 1983] routine G13CAF, with mean correction, split cosine bell taper coefficient 0.1 (as in the NAG example), the Parzen window such that the spectral density estimates have approximately 28 degrees of freedom, which is close to the recommended value of 32 in [Beauchamp and Yuen, 1979].
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... conditions7.2
This problem may be avoided by using eq. (5.61) instead of eq. (5.86).
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... matrix7.3
We used the Euclidean norm here in accordance with [Albano et al., 1988].
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... themselves7.4
Note that the number of distances is too large compared to the length of the time series (see section 5.6.1). This problem can be alleviated somewhat by multiplying the confidence intervals by a factor $\sqrt(N/M)$, where $N$ is the number of distances drawn and $M$ is the number of distances one is allowed to draw (see section 5.6.2). This will result in conservative confidence intervals.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.