In this thesis the potential of modulating solvent composition during chromatographic separation is investigated theoretically and experimentally. Hereby main focus is set on evaluating the potential of nonlinear gradient profiles in preparative liquid elution chromatography. As a case study was analysed the isolation of the second eluting component from a ternary mixture. Based on an experimental investigation, the changing thermodynamic equilibria and the effect of the gradient profiles on the shape of the elution profiles were studied theoretically. A reversed phase system was used with binary solvent mixtures of water and methanol to form the gradients. Thereby simulation and experimental verification of applying nonlinear gradients for the separation of ternary mixtures were performed. To quantify the isolation of components in the middle of an elution train, a careful selection of the cut times is required. In order to fulfil this task, a suitable procedure was developed in this study. The separation of the middle component of a ternary mixture resembles a more general separation problem of multi-component mixtures, where the target component needs to be separated from neighbouring components. Thus, the results of this study can be easily extended to optimize separations of multi-component mixtures. In the course of the work, at first adsorption isotherm parameters were estimated for a ternary mixture of three cycloketones considered as a model system. The effect of solvent compositions on these parameters was described mathematically. Gradient profiles were described mathematically as a function of time and a gradient shape factor. Four cases, differing by the number of free parameters, were considered to investigate the potential of nonlinear solvent gradients. The Craig equilibrium stage model was used to predict the band profiles and to quantify and compare different modes of operation (isocratic and various variants of gradient elution). Optimal operating conditions were identified theoretically for the production of cyclohexanone. The strong impact of the shape of gradients on process performance was elucidated. In the optimizations an artificial neural network method was used successfully. Finally, selected predictions were validated experimentally for optimal cases.