In a previous work, those of the luminosity measurements of the pre-white dwarf PG 1159-035 which are available online yielded estimates for the optimal embedding dimension, for the dimensionality of the phase space reconstructed from these observations, and for the maximal Lyapunov exponent $\lambda$: the result $\lambda = (9.2 \pm 1.0 ({\rm stat.}) \pm 2.7 ({\rm syst.})) \cdot 10^{-2}~\Delta \tau^{-1}$ ($\Delta \tau=10$ s is the sampling interval in the measurements) was obtained, suggesting that the physical processes, underlying the variation of the luminosity of PG 1159-035, are chaotic. An improved approach is employed in the present work in relation to the database of embedding vectors: instead of assigning each of the input time-series arrays either to the training or to the test set, the new approach features the creation of a dynamical database, i.e., of one which depends on the choice of the input test file. Although the size of the database is thus increased by a factor of about $2$ (compared to the previous study), the impact of this change on the important results is found to be insignificant. The estimate of this work for the maximal Lyapunov exponent ($\lambda = (8.9 \pm 0.7 ({\rm stat.}) \pm 1.9 ({\rm syst.})) \cdot 10^{-2}~\Delta \tau^{-1}$) is in very good agreement with the result of the earlier study.

The study of the luminosity measurements of the pre-white dwarf PG 1159-035 has established the properties of the rich power spectrum of the detected radiation and, derived thereof, the physical properties of this celestial body. Those of the measurements which are available online are analysed in this work from a different perspective. After the measurements were band-passed, they were split into two parts (of comparable sizes), one yielding the training (learning) set (i.e., the database of embedding vectors and associated predictions), the other the test set. The optimal embedding dimension $m_0=10$ was obtained using Cao's method; this result was confirmed by an analysis of the correlation dimension. Subsequently, the extraction of the maximal Lyapunov exponent $\lambda$ was pursued for embedding dimensions $m$ between $3$ and $12$; results were obtained after removing the prominent undulations of the out-of-sample prediction-error arrays $S (k)$ by fitting a monotonic function to the data. The grand mean of the values, obtained for sufficient embedding dimensions ($10 \leq m \leq 12$), was: $\lambda = (9.2 \pm 1.0 ({\rm stat.}) \pm 2.7 ({\rm syst.})) \cdot 10^{-2}~\Delta \tau^{-1}$, where $\Delta \tau=10$ s is the sampling interval in the measurements. On the basis of this significantly non-zero result, it may be concluded that the physical processes, underlying the variation of the luminosity of PG 1159-035, are non-linear. The aforementioned result for $\lambda$ was obtained using the $L^\infty$-norm distance; a larger, yet not incompatible, result was extracted with the Euclidean ($L^2$-norm) distance.