Re: problem with acos

[John W. Eaton]

On 29-Jul-2009, zampala wrote:

| This is what I get
| 
| octave:309> imag (P) == 0
| ans =
| 
|    1   1
|    1   1
| 
| I suppose the imaginary part is there, although small.

No, apparently it is exactly zero.  How was P created?

| I'm wondering if setting a very small number to zero is correct and,
| ultimately, this will affect my results on which I have no clue of what they
| should look like

For complex variables (whether the imaginary part is zero or not)
Octave uses a standard C++ library function to compute the complex
inverse cosine.  For real variables, it uses a different C library
function.  For example, note what happens if you do

  octave:1> acos (2)
  ans = NaN
  octave:2> acos (complex (2, 0))
  ans =  0.00000 - 1.31696i
  octave:3> cos (ans)
  ans =  2.0000

What are you trying to acomplish by computing the inverse cosine of a
real number outside the range [-1, 1]?

This discussion brings up an implementation question:  should Octave
switch to complex acos if the some of the arguments are real and
outside the range [-1, 1]?  I suppose that makes the most sense.

jwe