<div dir="ltr"><div dir="ltr"><br></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Sun, Aug 16, 2020 at 5:09 PM Hans Hagen <<a href="mailto:j.hagen@xs4all.nl" target="_blank">j.hagen@xs4all.nl</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">On 8/16/2020 2:19 PM, luigi scarso wrote:<br>
<br>
> Both binary and decimal uses and epsilon = 1E-52<br>
> buth double has epsilon = 2^-52 and this matches with the binary search.<br>
> Locally I have set epsilon = 2^-52 for both binary and decimal<br>
> and the results are now:<br>
I had been staring at that but wondered if<br>
<br>
@d epsilonf pow(2.0,-52.0)<br></blockquote><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<br>
was to be uised instead.</blockquote><div><br></div><div> epsilonf is used only in the "localized" mp_<number_system>_cross_point</div><div>but in this case there is only one mp_solve_rising_cubic that uses epsilon<br></div><div>which is, for binary and decimal, 10^-52 << 2^-52 .</div><div>Probably I should set </div><div>@d epsilon pow(2.0,-173.0)<br></div></div><div>given that </div><div>1E52*(2^-173) =~0.83523897190381113942<br></div><div><br></div><div><br></div><div><br></div><div><br></div>-- <br><div dir="ltr">luigi<br></div></div>