Artificial neural network applied to solid state thermal decomposition

A multi-layer neural network is constructed to describe the thermal decomposition of rhodium acetate. Critical analysis of the residual, trained, interpolated and extrapolated errors, with the number of neurons, indicates the efficiency of the present approach. It was possible, within this framework, to improve the A _n model, with a better correlation between the results. A new value of the activation energy, E _a, and frequency factor, Z, are calculated for the decomposition process. Since the neural network is more precise than a particular model, the calculated values for these quantities are believed to be more precise. The computed values are E _a=194.0 kJ mol^-1 and Z=5.23·1016 s^-1. The neural network eliminates the step to decide, among the available models, the one that best fit the data. An agreement up to four significant figures can be achieved even for data not used in the training process, both in the interpolated and extrapolated regions. This method suggests, therefore, an important alternative tool for the experimentalists. The present approach can also be adapted to other systems and to data in two dimensions. This revised version was published online in July 2006 with corrections to the Cover Date.