Reviewing the complexity of endogenous technological learning for energy system modeling

Energy system components like renewable energy technologies or electrolyzers are subject to decreasing investment costs driven by technological progress. Various methods have been developed in the literature to capture model-endogenous technological learning. This review demonstrates the non-linear relationship between investment costs and production volume, resulting in non-convex optimization problems and discuss concepts to account for technological progress. While iterative solution methods tend to find future energy system designs that rely on suboptimal technology mixes, exact solutions leading to global optimality are computationally demanding. Most studies omit important system aspects such as sector integration, or a detailed spatial, temporal, and technological resolution to maintain model solvability, which likewise distorts the impact of technological learning. This can be improved by the application of methods such as temporal or spatial aggregation, decomposition methods, or the clustering of technologies. This review reveals the potential of those methods and points out important considerations for integrating endogenous technological learning. We propose a more integrated approach to handle computational complexity when integrating technological learning, that aims to preserve the model's feasibility. Furthermore, we identify significant gaps in current modeling practices and suggest future research directions to enhance the accuracy and utility of energy system models.