Link Approximation Performance Ratio and Average Convergence Rate

In practice, evolutionary algorithms are widely used to find good solutions to hard combinatorial optimization problems. In theory, it is necessary to answer the question of how good approximation solutions these algorithms can produce. This paper aims to estimate the approximation performance ratio in each generation. Using the average convergence rate of evolutionary algorithms, an exponential decay is proposed as a theoretical prediction for the approximation performance ratio. A theoretical analysis proves that the prediction is exact for some initialization and this is confirmed by experiments. However, experimental results also reveals that the prediction is not good for other initialization. Therefore, in order to improve the prediction, it is necessary to consider other transition probabilities.