Re: roots(poly(r)) != r

[Quentin Spencer]

Torben Mikael Hansen wrote:
> I have been playing around with poly and roots, don't understands this
>
> a=rand(100,1);
> r=exp(I*.1*pi*2*[a;-a]);
> e=max(sort(abs(roots(poly(r))))-sort(abs(r)));
>
> Shouldn't this give an "e" that close to zero. I always get something in the 
> order of 10.
>   

Not really. Computing the roots of a polynomial is known to be sensitive 
to roundoff errors. This is easily demonstrated by looking at the output 
of "roots(poly(1:20))". The errors only get worse as the order of the 
polynomial increases.

Quentin