A mathy construction like in the article is probably important for the full conjecture, but isn't this concrete case just an instance of the bin-covering problem? Your discrete items are the log of each prime factor (included according to its multiplicity), set the lower threshold to log(100k), and if you get any solution with 300k or more factors, you can redistribute the extra factors arbitrarily.
As impressed as I am with this solution, it didn't get to be the first to solve the problem.
I did't notice it. It's using a different method, but it's interesting that it also uses the primes in N! ordered from bigger to smaller.
the primes bigger than ~sqrt(t) have an "obviously" best matching so it's not too surprising.
I like how Tao put out the call, and you answered it. Very cool! I was waiting for someone on here with engineering and math chops to attempt devising a more efficient approach teased in that article. Super cool to see you do it! You might be co-author on his paper, I think that's what he said, hahaha ! :)
Hey! This guy is using the fancy new treelist [1] package which implements RRB trees. These are super cool data structures: a cache-friendly, multiway tree structure to store a sorted collection with O(log_{32} n) read, functional insert/update/append/prepend. Really a fantastic data structure for functional programming.