The prediction of future from the past: an old problem from a modern perspective

Abstract

The idea of predicting the future from the knowledge of the past is quite natural when dealing with systems whose equations of motion are not known. Such a long-standing issue is revisited in the light of modern ergodic theory of dynamical systems and becomes particularly interesting from a pedagogical perspective due to its close link with Poincaré's recurrence. Using such a connection, a very general result of ergodic theory - Kac's lemma - can be used to establish the intrinsic limitations to the possibility of predicting the future from the past. In spite of a naive expectation, predictability results to be hindered rather by the effective number of degrees of freedom of a system than by the presence of chaos. If the effective number of degrees of freedom becomes large enough, regardless the regular or chaotic nature of the system, predictions turn out to be practically impossible. The discussion of these issues is illustrated with the help of the numerical study of simple models.

arXiv comment9 pages, 4 figures

References in corpus (4)

• Predictability: a way to characterize Complexity
• Recurrence time analysis, long-term correlations, and extreme events
• Extensive Chaos in the Lorenz-96 Model
• Extracting dynamical equations from experimental data is NP-hard

Cited by in corpus (15)

• Forecasting in the light of Big Data
• The Role of Data in Model Building and Prediction: A Survey Through Examples
• Langevin equation in systems with also negative temperatures
• Nonequilibrium Statistical Mechanics and Optimal Prediction of Partially-Observed Complex Systems
• Revisiting Evidence of Chaos in X-ray Light Curves: The Case of GRS 1915+105
• Artificial Intelligence, Chaos, Prediction and Understanding in Science
• Response and flux of information in extended non-equilibrium dynamics
• Data science and the art of modelling
• Decomposition of conditional probability for high-order symbolic Markov chains
• Data-assisted, physics-informed propagators for recurrent flows
• Correlation properties of the random linear high-order Markov chains
• Isomorphism between Maximum Lyapunov Exponent and Shannon's Channel Capacity
• Network reconstruction may not mean dynamics prediction
• Prediction and Inference: From Models and Data to Artificial Intelligence
• Effective Equations in complex systems: from Langevin to machine learning