Re:Short algorithm, long-range consequences
by _delirium on Saturday March 02, 2013 @12:13AM
(#5308540)
The term "nearly linear time" is fine. The next part of that sentence is where the problem arises: it suggests that the opposite of "nearly linear", and the status quo, is that running time "increase[s] exponentially". But the existing standard algorithms already scale polynomially, not exponentially.
However, the paper gives a better definition:
When we refer to a nearly linear time SDD system solver we mean an an algorithm that computes such a x in time O(mlog^c(n)logε⁻¹) where m is the number of nonzero entries in A and c ≥ 0 ∈ ℝ is a fixed constant.