27.12.2012 Views

Oscillations, Waves, and Interactions - GWDG

Oscillations, Waves, and Interactions - GWDG

Oscillations, Waves, and Interactions - GWDG

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

418 U. Parlitz<br />

Figure 9. Correlation sum Eq. (15)<br />

vs. radius r for the Poincaré cross<br />

section Fig. 4 of the chaotic attractor<br />

of Duffing’s equation computed using<br />

100000 points. The slope of ≈ 1.34<br />

represented by the (black) line segment<br />

is an approximation of the correlation<br />

dimension DC, Eq. (16).<br />

of the correlation sum<br />

C(r) =<br />

2<br />

(N)(N − 1)<br />

ln C(r)<br />

−5<br />

−10<br />

−15<br />

−20<br />

−15 −10 −5 0<br />

ln r<br />

N� �i−1<br />

H(r − �y i − y j �) (15)<br />

i=1 j=1<br />

that counts the relative number of neighbouring points y i <strong>and</strong> y j closer than r (H is<br />

the Heaviside function with H(x) = 1 for x > 0 <strong>and</strong> zero elsewhere). The correlation<br />

dimension<br />

ln C(r)<br />

DC = lim<br />

(16)<br />

r→0 ln r<br />

describes the scaling in the limit r → 0 <strong>and</strong> equals the Rényi dimension D2.<br />

Since any numerical simulation or experimental measurement provides finite data<br />

sets, only, the limit r → 0 cannot be carried out in numerical computations of DC,<br />

but only the corresponding scaling behaviour<br />

ln C(r) ≈ DC ln r (17)<br />

can be exploited to estimate DC as a slope in a log-log-diagram. This approach is<br />

illustrated in Fig. 9 showing ln C(r) vs. ln r for the Poincaré section of the chaotic<br />

Duffing attractor in Fig. 4. The slope in a suitable intermediate range of r gives<br />

the fractal dimension of the Poincaré section of the chaotic Duffing attractor (here<br />

estimated as DC ≈ 1.34).<br />

Each point in the Poincaré cross section corresponds to a one-dimensional trajectory<br />

(segment) starting at that point. Therefore, the dimension estimate in the<br />

Poincaré section has to be increased by one to obtain the (correlation) dimension of<br />

the full chaotic attractor (here DC ≈ 2.34).<br />

The definition of the correlation dimension (16) is based on a given radius r <strong>and</strong><br />

therefore a fixed size approach. As an alternative one may also use a fixed mass<br />

method to estimate the dimension of the attractor as was suggested by Badii <strong>and</strong><br />

Politi [27,28] <strong>and</strong> Grassberger [29]. In this case the k nearest neighbours of each<br />

reference point yn <strong>and</strong> the radius rn = r(k) of this cloud of k points are determined.<br />

For the limit k/N → 0 one obtains, for example, an approximation of the boxcounting<br />

dimension<br />

DB ≈ −<br />

log 1<br />

N<br />

log N<br />

� . (18)<br />

N<br />

n=1<br />

rn

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!