11/27/2021, 01:26 PM
(This post was last modified: 11/27/2021, 02:04 PM by Ember Edison.)
Thanks, James! Your solution is valid in broad principle.
But picking g(|u|)=sqrt(abs(u)) is too much of a burden on memory and cpu time: init_OFF(1,1E20) already requires >25GB RAM and several hours of cpu time.
The memory limit is less important for me, I have 128GB RAM and about 8TB HDD and 1TB SSD available for Virtual Memory, but it's just too slow. Not only is init_OFF slow, but beta_off is slow too. This means that for myself g(|u|) better not exceed 1E10.
sqrt(log(3E694127911065419641)) = 1.26E9, so I changed to g(|u|)=sqrt(log(abs(u))), which already solves init_OFF(1,1E25), but fail in init_OFF(1,1E50). A better form of the g function is still about a few weeks away from being determined.
setup g(|u|)=log(abs(u))^6, init_OFF(1E50) need >50GB, I don't think there is much time left before we get the final industrial upper limit of the beta function.
But picking g(|u|)=sqrt(abs(u)) is too much of a burden on memory and cpu time: init_OFF(1,1E20) already requires >25GB RAM and several hours of cpu time.
The memory limit is less important for me, I have 128GB RAM and about 8TB HDD and 1TB SSD available for Virtual Memory, but it's just too slow. Not only is init_OFF slow, but beta_off is slow too. This means that for myself g(|u|) better not exceed 1E10.
sqrt(log(3E694127911065419641)) = 1.26E9, so I changed to g(|u|)=sqrt(log(abs(u))), which already solves init_OFF(1,1E25), but fail in init_OFF(1,1E50). A better form of the g function is still about a few weeks away from being determined.
setup g(|u|)=log(abs(u))^6, init_OFF(1E50) need >50GB, I don't think there is much time left before we get the final industrial upper limit of the beta function.