0.0 versus 0.00000000000
Jeroen Demeyer
there is some discussion about whether or not the number 0.00000000
(with many zeros) should have a higher precision than 0.0. Currently,
0.0 and 0.0000000000000000 have the same precision of 53 bits.
In my opinion, the current behaviour is what makes the most sense
mathematically, but other disagree.
Jason Grout
Another issue is the following two cases:
000000000000000000.0
and
0.000000000000000000
Should those both have the same precision? As Robert points out on the
ticket, this is related to what the definition of 'trailing zero' is.
Thanks,
Jason
Jeroen Demeyer
> On 6/8/11 2:04 PM, Jeroen Demeyer wrote:
>> On Trac ticket #8896 (http://trac.sagemath.org/sage_trac/ticket/8896),
>> there is some discussion about whether or not the number 0.00000000
>> (with many zeros) should have a higher precision than 0.0. Currently,
>> 0.0 and 0.0000000000000000 have the same precision of 53 bits.
>>
>> In my opinion, the current behaviour is what makes the most sense
>> mathematically, but other disagree.
>
> Another issue is the following two cases:
>
> 000000000000000000.0
>
> and
>
> 0.000000000000000000
>
> Should those both have the same precision?
Since a decimal point simply indicates a change of exponent in the
floating-point representation, I think these two certainly should have
the same precision. Think of the above numbers as
0000000000000000000e-1
and
0000000000000000000e-18
Jeroen
Jason Grout
But we already have the convention that zeros after the decimal indicate
precision:
sage: a=4.000000000000000000000000000000
sage: a.prec()
103
sage: b=4.00000
sage: b.prec()
53
So there is already something special about where the decimal point is
placed.
Jason
Jeroen Demeyer
> So there is already something special about where the decimal point is
> placed.
4.000000000000000000000000000000
and
4000000000000000000000000000000.
and
4000000000000000000000000000000e100
should have the same precision.
I did not make a claim about
4.00000
Nils Bruin
> But we already have the convention that zeros after the decimal indicate
> precision:
> placed.
have a special role. I don't think the decimal point plays a role in
this:
sage: parent(4.00000000000000000000)
Real Field with 70 bits of precision
sage: parent(400000000000000000.000)
Real Field with 70 bits of precision
sage: parent(0.400000000000000000000)
Real Field with 70 bits of precision
sage: parent(0.0000000000000000400000000000000000000)
Real Field with 70 bits of precision
sage: parent(00000000000000400000000000000000.000)
Real Field with 70 bits of precision
(in each example there are 20 zeros trailing the 4)
That's also the reason why the system has problems guessing the right
precision for representations of 0. I think providing a way to imply
precision for 0 by specifying extra leading/trailing zeros (how do you
tell the difference?) is always going to be a hack.
Robert Bradshaw
> On 2011-06-08 21:23, Jason Grout wrote:
>> So there is already something special about where the decimal point is
>> placed.
> Not really. My claim is that
> 4.000000000000000000000000000000
> and
> 4000000000000000000000000000000.
> and
> 4000000000000000000000000000000e100
> should have the same precision.
Which they do, that's not the question here.
The question is really whether
0.00000000000000000000000000000000
should have the same precision as
1.00000000000000000000000000000000
I think it should, as the only semantic information trailing zeros
could possibly carry is an attempt to increase precision. I find the
fact that adding zeros to the latter increases precision, but adding
zeros to the former does not, to be surprising. To me, trailing zeros
are the unnecessary zeros to the right of a decimal expansion (whether
or not they come after a non-zero digit), and are used to denote
precision.
Whether 000000000000000000.0 and 0.000000000000000000 have the same
precision is a smaller question, but I'm inclined to say yes.
- Robert