Abstract: | We examine phase transitions in real computational problems using a wide variety of algorithms. These phase transitions resemble those observed with randomly generated problems. Real problems do, however, contain new features (for example, large scale structures rare in random problems) which can make them significantly harder than random problems. Our results suggest a new methodology for benchmarking algorithms. in addition, they help to identify the location of the really hard real problems.
|