High-Performance A\ast Search Using Rapidly Growing Heuristics

In high-performance A* searching to solve satisfici n g prob lems, there is a c r i t i ca l need to design heur is t ics which cause low t ime-complex i ty . In order for humans or machines to do this effectively, there mus t be an unders tand ing of the domainindependent properties that such heurist ics have. We snow that , contrary to common belief, accuracy is no t c r i t i c a l ; the key issue is w h e t h e r or no t heur is t ic values are concentrated closely near a rapid ly growing "central funct ion." As an applicat ion, we show that , by "mult ip ly ing" heuristics, it is poss ib le to reduce e x p o n e n t i a l average t i m e c o m p l e x i t y to p o l y n o m i a l . Th i s i s c o n t r a r y to conclusions drawn from previous studies. Exper i mental and theoretical examples are given.