Introspective sorting and selection revisited

We describe two improvements to introspective sorting and selection algorithms: a simple rule for fine‐grained introspection that detects potential worst‐case performance after only a small constant number of partitioning steps, and the use of remedial randomization as an intervention strategy in order to reduce the performance penalty for false positives. We present experimental results showing that these techniques provide significant improvements in the running time for worst‐case and other troublesome inputs, without sacrificing performance on well‐behaved inputs. Copyright © 2000 John Wiley & Sons, Ltd.