Accuracy of real coefficients

[Jean-Pierre Flori]

Dear all,

It seems that if the "additional magic invisible" digits used for real number computations is less than what the machine float precision, then only machine floats are used, even for the exposed part of real numbers, see:
https://github.com/Singular/Sources/blob/spielwiese/Singular/ipshell.cc#L5293

Note that the current state of that part of the file went through a bunch of changes, inccluding the addition of an assume(float_len < float_len2):
https://github.com/Singular/Sources/commit/046a941e1caf6571a05abbcd7e9dc58a903aa4eb
an addition to ensure this assumption:
https://github.com/Singular/Sources/commit/24bddc389d70082bf25e49397ca06b884dabd5d3
and its removal:
https://github.com/Singular/Sources/commit/0c85f5b430f03235e2be17c77b2811ebf3c1df3f

So what should be the meaning of float_len and float_len2?
If it must be assumed that float_len < float_len2, then maybe a warning should be issued when the precision falls back to machine precision?

Best,
JPF

[han...@mathematik.uni-kl.de]

On Thu, Nov 12, 2015 at 04:54:33PM -0800, Jean-Pierre Flori wrote:
> So what should be the meaning of float_len and float_len2?
> If it must be assumed that float_len < float_len2, then maybe a warning
> should be issued when the precision falls back to machine precision?
>
> Best,
> JPF
>

This was a misinterpretation (maybe unsufficient documentation):
float_len is the visible number of digits, float_len2 the number of
additional, invisible digitss - NOT the number of all digits.

Hans

[Jean-Pierre Flori]

Ok, do you intend to push a fix for that?
Could you post a ref to the commit here?

Thanks!

Hans