Geometric Considerations of Search Behavior

Abstract

In this article, behavior of a heuristic search with a heuristic function divided into two parts is discussed based on a geometric analysis, and a method for improving the heuristic function is proposed. One part of the divided heuristic function is called short-term forecast. Another part is called long-term forecast. Based on the discussion, it is suggested there exists a possibility that the difference of accuracy of both forecasts leads the search to an incorrect path. Since there exists a possibility that the search becomes efficient and that the search does not select an incorrect path by improving accuracy of a long-term forecast, a method to update the value of a long-term forecast is proposed. It is shown that the proposed method is effective under the tree structure which cost of the edge observed at deep location is greater or equal than that observed at shallow location.