Speaker
Description
Computational efforts for the calculation of chemical reactions are about 30% of the total resource requested to run simulations involving climate models. Finding alternatives to speed up the calculation of the chemistry module is then a crucial task.
Recent studies show that the calculation of the Jacobian matrix is the most computationally demanding part of the related ODEs and then solutions have been sought to overcome this problem.
In this poster results from KPP and ICON-ART (in a box model version) for the stiff H2O2 chemistry and the air-pollution Verwer systems, compared with neural network corresponding results are shown. The H2O2 chemistry mechanism consists of 4 reactions (3 species), while the Verwer system is made of 25 reactions (21 species). The simulations have been initialized with a fixed and a random set of values. These results form the basis to subsequently train the neural network.
