Optimization by trees on simple adaptive landscapes.
We evaluate the optimizing ability (rate of adaptation) of trees on simple adaptive landscapes. At points away from a peak, there is a strong negative relationship between rate of adaptation and tree precision P, a relationship that is independent of the size of the tree. P measures the variability among trial solutions generated by the tree: high precision trees have low variability, low precision trees have high variability. Near a peak, the situation reverses, with high precision trees showing higher rates of adaptation than low precision trees; however, for all trees, the absolute rate of adaptation is uniformly low. On multiple-peak landscapes, the probability of crossing an adaptive valley from a lower peak to a higher peak is also negatively correlated with tree precision. These results suggest that under a wide range of conditions, trees with low precision are, on average, the best optimizers.