A priori neural networks versus a posteriori MOOD loop: a high accurate 1D FV scheme testing bed (Q2007012)

The authors consider 1D hyperbolic systems of conservation. Such systems cover for instance equations of: scalar linear advection/transport, Burgers, the isentropic Euler, and the so-called M1 model for radiative transfer in transparent media. The computational domain is meshed using a finite volume method with cells of constant size. They present an approach to replace an a posteriori MOOD (multi-dimensional optimal order detection) loop used in a high accurate finite volume (FV) schemes by a trained artificial neural network (NN). The MOOD loop ensures accuracy (essentially non-oscillatory), robustness properties, and preserves some physical features. Shallow NNs made of only two so-called hidden layers and few perceptrons are trained. The strategy which uses such NNs to train is described on several 1D toy models described above. Each toy model brings new difficulties which are enlightened on the obtained numerical solutions. On these toy models, and for the proposed test cases, an artificial NN can be trained and substituted to the a posteriori MOOD loop in mimicking the numerical admissibility criteria and predicting the appropriate polynomial degree to be employed safely. The physical admissibility criteria is dealt with the a posteriori MOOD loop. Constructing a valid training data set is of paramount importance, but once available, the numerical scheme supplemented with NN produces interesting results.