Methods for the determination of local dynamical linearization information
from experimental time series data are subject to computational artifacts.
We investigate the artifacts due to observational noise in the data, and gi
ve formulas for the expected values of the reconstructed Jacobian in some s
imple cases. The formulas we derive in the case of realistic noise amplitud
es are quite different from those for the noiseless case. In turn, spurious
Lyapunov exponents in the noisy case are correspondingly different from th
e noiseless case.