Building healthy Lagrangian theories with machine learning

The existence or not of pathologies in the context Lagrangian theory is studied with aid Machine Learning algorithms. Using an example framework classical mechanics, we make a proof concept, that construction new physical theories using machine learning possible. Specifically, utilize fully-connected, feed-forward neural network architecture, aiming to discriminate between ``healthy'' and ``non-healthy'' Lagrangians, without explicitly extracting relevant equations motion. network, after training, used as fitness function concept genetic algorithm healthy Lagrangians are constructed. These different from contained initial data set. Hence, searching for possessing number pre-defined properties significantly simplified within our approach. employed this work can be explore more complex theories, such generalizations General Relativity gravitational physics, constructions solid state which standard procedure laborious.