The soft heap: an approximate priority queue with optimal error rate

Reviewer: Susan M. Merritt

This paper describes a simple variant of a priority queue, an approximate priority queue, called a soft heap. The soft heap supports the operations: create, insert, delete, meld, and findmin. The authors show that the complexity of a soft heap is better than the logarithmic bound of a heap in a comparison-based model. The soft heap is implemented with pointers, with no arrays and no assumptions about the keys. The soft heap may increase the value of certain keys, which are then referred to as corrupted. Corruption is done by the data structure and the user has no control over it. Findmin, for example, finds the current minimum, which may or may not be corrupted. The benefit of the approach is speed: during heap operations items travel together in something akin to "carpooling" to save time. The authors prove that the soft heap has optimal time bonds in a comparison-based model. They indicate that the soft heap was designed specifically for computing minimum spanning trees; it is a key part of the current fastest deterministic algorithm. They show that soft heaps support dynamic maintenance of percentages in constant amortized time. They explain how soft heaps offer an alternative method for computing medians in linear time. They demonstrate that soft heaps can also do two kinds of approximate sorting. The paper is very well written. The authors describe in detail the soft heap data structure, which is a binomial tree with subtrees possibly missing. They include the actual C code for implementing a soft heap. The set of references is apparently complete. The paper is a good exposition of a very interesting and ingenious data structure and its applications. It is accessible to those who know some computer science.