## What can we learn from observational stellar time series?
^{1} Institute of Astronomy, Madingley Road, Cambridge CB3
OHA, UK and Institut d'Astrophysique, 5, Avenue de Cointe, B-4000
Liège, Belgium^{2} Observatoire de Paris, 5, Place Jules Janssen, F-92195
Meudon Principal Cedex, France
A synthetic stellar time series simulating the observational radial velocity curve of a variable star is generated as follows. A 2D Lattice Gas (LG) model adapted to mimic the nonlinear nonradial behaviour of a vibrationally unstable plane-parallel stellar atmosphere representative for Long Period Variables is integrated over some 15 cycles. The model exhibits a complex spatio-temporal behaviour, showing simultaneously localised zones of expansion and contraction. The vertical component of the velocity field averaged over the atmosphere defines a noisy, cyclic time signal which is filtered with a Neural Network (NN) device. The analysis of the filtered signal by the method of Global Flow Reconstruction demonstrates that the latter is reproduced by a low dimensional (=4) dynamical system. The experiment is thus indicative that the actual high-dimensional fluid dynamics describing the full 3D behaviour of a real star possesses an approximately autonomous low-dimensional directly observable radial subdynamics which alone is accessible with the conventional global observational procedures. Comments are made on the relation between this observable subdynamics and the observationally hidden full dynamics of the star.
## Contents- 1. Introduction
- 2. Time series from a CA pulsation model
- 3. Neural network analysis and filtering
- 4. Reconstruction of signal dynamics
- 5. Discussion and conclusion
- Acknowledgements
- References
© European Southern Observatory (ESO) 1998 Online publication: June 2, 1998 |