Can you dig through into the code, see what's going on, and suggest a
documentation patch? To get you started, the code for the complex
version of atan2 is in
https://svn.r-project.org/R/trunk/src/main/complex.c
z_atan2 is at line 669 (the first argument is a pointer to the result,
args 2 [csn] and 3 [ccs] are pointers to the arguments of atan2())
In generic cases the computation is
dr = catan(dcsn / dccs);
if(creal(dccs) < 0) dr += M_PI;
if(creal(dr) > M_PI) dr -= 2 * M_PI;
where dcsn, dccs are converted versions of the args.
catan() is *either* taken from system libraries or is defined at line 489.
On my system (Ubuntu), 'man 3 catan' gives documentation on the
function, and says "The real part of y is chosen in the interval
[-pi/2,pi/2]" - but that _could_ be system-dependent.
cheers
Ben Bolker
On 5/18/21 10:39 AM, Jorgen Harmse via R-devel wrote:> The current documentation says that atan2(y,x) is the angle between the
x-axis and the vector from the origin to (x,y), but what does this mean when x
& y are complex? The function seems to pick theta with Re(theta) between -pi
and pi and with tan(theta) (approximately) equal to y/x, but that leaves 2
(sometimes 3) options, and there must be a set (branch region with 3 real
dimensions?) on which the function is discontinuous. Please add details.
>
> Even for real inputs, it might help to spell out the behaviour on the
negative x-axis. It mostly matches the branch-cut rules for the other functions,
but atan2(0,0)==0 is a unexpected.
>
> I also suggest ?See Also? links from trigonometric functions to hyperbolic
functions and from hyperbolic functions to exponential & logarithmic
functions.
>
> Regards,
> Jorgen Harmse.
>
>
>
>> R.version.string
>
> [1] "R version 4.0.4 (2021-02-15)"
>
>
>
>
>
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>