Euler's Partition Function Theory Finished 117
universegeek writes "Mathematician Ken Ono, from Emory, has solved a 250-year-old problem: how to exactly and explicitly generate partition numbers. Ono and colleagues were able to finally do this by realizing that the pattern of partition numbers is fractal (PDF). This pattern allowed them to find a finite, algebraic formula, which is like striking oil in mathematics."
A Partician is: (Score:5, Interesting)
"a partition is a way of representing a natural number n as the sum of natural numbers (ie. for n = 3, we have three partitions, 3, 2 + 1, and 1 + 1 + 1, independent of order). Thus, the partition function, p(n), represents the number of possible partitions of n. So, p(3) = 3, p(4) = 5 (for n = 4, we have: 4, 3 + 1, 2 + 2, 2 + 1 + 1, 1 + 1 + 1 + 1) , etc.."
Very interesting read.
Boy... (Score:3, Interesting)
This pattern allowed them to find a finite, algebraic formula...
Yeah, but looking at the paper, still not that simple. Eventually someone will be able to program it into a function and I'll be able to call it in Matlab, but until then, I'd still be worried about making calculation errors. On the other hand, that may be saying more about my calculation skills than about the work...
Ken Ono is a great guy (Score:5, Interesting)
Couldn't happen to a harder-working guy BTW, or a nicer one. I'll never forget him desperately writing the final draft of his wedding vows on the day of the ceremony.
Re:In English (Score:5, Interesting)
Well, in statistics it's pretty common to fit models to partitions of data, and the partitioning process gets ugly when the data set is large (in terms of classes of data, not in terms of the number of points in the data.) And translating from partition numbers to actual partitions is trivial. Speaking as a statistician who only deals with number theory on the (rare) occasions that it's directly relevant to my work, I have to say that the existing partitioning algorithms, although they work, strike me as inelegant, and I'd be happy to have something cleaner that can deal with an arbitrarily large number of classes of data in "O(something small)" time. I can see this speeding up model selection problems at least somewhat, although most of the computational expense will still be in actually fitting the models and calculating the relevant performance criteria.