Re: About -0 and type complex

[Daniel J Sebald]

On 10/09/2012 12:22 PM, Michael D. Godfrey wrote:
> I would just like to ask if the following behavior seems right
> to you. And, is there a way to set the imaginary part of
> a complex to -0?
>
> octave:46> -(1-1)
> ans = -0
> octave:47> -(1-1) +0i
> ans = 0
> octave:48> -(1-1) -0i
> ans = -0
> ==================
> The obvious question is: why does the sign of the imaginary
> part change the sign of the real part?

Well, what you've written isn't manipulating the real/imaginary part, 
from what I gathered of the discussion about a month back related to 
this.  The commands above are adding two different numbers and how the 
addition rules act might be slightly different than what you have in 
mind, which is

octave-cli:7> real(complex(-0,0))
ans = -0
octave-cli:8> real(complex(0,0))
ans = 0
octave-cli:9> real(complex(-0,0))
ans = -0
octave-cli:10> real(complex(0,-0))
ans = 0
octave-cli:11> real(complex(-0,-0))
ans = -0

Let's try some of the calculations you wrote:

octave-cli:12> x = -(1-1)
x = -0
[I think that was 1-1 is 0 and minus 0 is -0]
octave-cli:13> x = -(1-1) + complex(0,0)
x = 0
[I think that was -0 + 0 which is probably 0]
octave-cli:14> x = -(1-1) + complex(0,-0)
x = 0
[I think that was -0 + 0 again, which is 0]
octave-cli:15> x = -(1-1) - 0i
x = -0
[I think that was -0 - 0, which is -0]
octave-cli:16> x = -(1-1) - complex(0,0)
x = -0
[Same as the last case, the sign of the complex portion not having any 
significance.]

Dan


>  The math definitions for
> arg(z) and |z| treat the real and imaginary parts symmetrically.
>
> I have not found a standard for +-0 for complex, but it would
> seem reasonable to treat the real and imaginary parts according
> to the same rule. And, is it a good idea to reduce a complex
> with +-0 imaginary part to real? Saves space, I know, but the
> sign is lost.
>
> Currently, for complex the rule is:
> real imag result
> -0 -0 -0
> -0 0 0
> 0 -0 0
>
> In effect, there is no -0 imaginary part even though -0i affects
> results.
>
> I know that this seems pretty academic, but it is getting in the way
> of my figuring out the correct corner values of atan2(y,x) for complex
> arguments (another pretty academic pursuit. I know).
>
> If you think this is good enough, I will just add it to the documentation
> about -0. At least, users should know what is supposed to happen.
>
> Michael

--

Dan Sebald
email: daniel(DOT)sebald(AT)ieee(DOT)org
URL: http://www(DOT)dansebald(DOT)com