Evaluation of Equilibria with a Use of Artificial Neural Networks (ANN): I. Artificial Neural Networks and Experimental Design as a Tool in Electrochemical Data Evaluation for Fully Inert Metal Complexes

Abstract

The use of artificial neural networks (ANN) and experimental design (ED) for refinement of experimental data obtained in a polarographic metal-ligand equilibrium study of fully inert (kinetically slow) metal complexes of highly acidic metal ions at low pH is described. Three metal-ligand systems are discussed: i) evaluation of log K1 of an inert complex ML when hydrolysis of a metal ion M is negligible, ii) simultaneous evaluation of log K1 and log K, iii) evaluation only of log K1 in the presence of metal hydrolysis and when the stability constant log K is not known and is of no interest. It is shown that one can estimate the above stability constants with satisfactory accuracy (with a relative error below ± 0.2 % or a standard deviation in the calculated stability constant below ± 0.02 log unit on error-free data) by ANN and ED approach. High rigidity of ANN towards errors that were significantly larger than expected from an experiment was demonstrated. Methodology described allows one to study metal-ligand equilibria by polarography at any temperature and ionic strength without prior requirement of establishing the hydrolysis constant at the experimental conditions employed. Several structures of ANN and ED were tested and optimized.