On Jan. 14, 2011, Michel Talon wrote:
---------------------------------------------------
>>i have just looked at your manuals, and noted a problem i remarked
>>recently,
that you may have not noted in your Chapter 9, use of bfallroots.
The problem is that if you feed a polynomial with bfloat coefficients (*)
to bfallroots, to get the solution, the first thing it does is to convert
all coeffs to rats, and does that with small precision (something like
eight digits), so you can set fpprec=30 as much as you want the solutions
are not correct to more than 7 digits.
The solution is to also set ratepsilon=30 before beginning the computation,
then you get roots correct to around 30 figures. If this appears somewhere
on your paper i have not seen, please excuse my post.
---------------------------------------------------------------------
Thanks for the information about the behavior of bfallroots. I really
dropped the ball on that section, and will need to revise Ch. 9 with
another look at that function.
--------------------------------------------
>> the example i had was with finding roots for a Bethe ansatz, the
polynomials really has coefficients with a lot of digits, there are huge
order of magnitudes between coeffs, still bfallroots finds correct results
-----------------------------------------------------
Perhaps you could send me a version of that example for possible inclusion
in the next revision of Ch. 9.
Thanks again,
Ted (aka Edwin)