On Thursday, April 28, 2016 at 8:46:06 AM UTC-4, Steven G. Johnson wrote: > > Out of curiosity, how does the performance and accuracy compare to the > "roots" function in the Polynomials.jl package, which uses the standard > companion-matrix approach? >
Just checked: it is much faster than Polynomials.roots, which I guess is not surprising for a Newton-like method compared to an eigenvalue method, although perhaps the latter is more reliable in some cases? I'm not sure what the state of the art here is, though.
