Directed acyclic graph (DAG) is an essentially important model to represent terminologies and their hierarchical relationships, such as Disease Ontology. Due to massive terminologies and complex structures in a large DAG, it is challenging to summarize the whole hierarchical DAG. In this paper, we study a new problem of finding k representative vertices to summarize a hierarchical DAG. To depict diverse summarization and important vertices, we design a summary score function for capturing vertices' diversity coverage and structure correlation. The studied problem is theoretically proven to be NP-hard. To efficiently tackle it, we propose a greedy algorithm with an approximation guarantee, which iteratively adds vertices with the large summary contributions into answers. To further improve answer quality, we propose a subtree extraction based method, which is proven to guarantee achieving higher-quality answers. In addition, we develop a scalable algorithm k-PCGS based on candidate pruning and DAG compression for large-scale hierarchical DAGs. Extensive experiments on large real-world datasets demonstrate both the effectiveness and efficiency of proposed algorithms.