Re: Different numerical results between sparse and full matrices

[fotios]

On 02/17/2012 02:41 PM, Michael Goffioul wrote:
> Hi,
>
> While investigating some the remaining test errors I have when
> building octave with MSVC, I found a behavior difference between
> sparse and full matrices, which boils down to the following example
> (see result of "sparse(x)./sparse(0)"):
>
> ========
>
> octave.exe:26>  x = 1+1i
> x =  1 + 1i
>
> octave.exe:27>  x/0
> warning: division by zero
> ans = Inf + Infi
>
> octave.exe:28>  sparse(x)./sparse(0)
> ans = Compressed Column Sparse (rows = 1, cols = 1, nnz = 1 [100%])
>
>    (1, 1) ->  NaN + NaNi<<<<<<<<  THIS IS WHERE THE PROBLEM IS
>
> octave.exe:29>  sparse(x)./0
> warning: division by zero
> ans = Compressed Column Sparse (rows = 1, cols = 1, nnz = 1 [100%])
>
>    (1, 1) ->  Inf + Infi
>
> octave.exe:31>  sparse(real(x))./sparse(0)
> ans = Compressed Column Sparse (rows = 1, cols = 1, nnz = 1 [100%])
>
>    (1, 1) ->  Inf
>
> ========
>
> In other words, complex division by 0 in sparse matrices gives NaN
> instead of Inf. Does anybody know where I should look at to find where
> this could come from?
>
> Michael.

octave:9> ver
----------------------------------------------------------------------
GNU Octave Version 3.6.0
GNU Octave License: GNU General Public License
Operating System: Linux 3.0.0-16-generic-pae #28-Ubuntu SMP Fri Jan 27 
19:24:01 UTC 2012 i686
----------------------------------------------------------------------
no packages installed.
octave:10> x = 1 + 1i
x =  1 + 1i
octave:11> sparse(x)./sparse(0)
ans = Compressed Column Sparse (rows = 1, cols = 1, nnz = 1 [100%])

  (1, 1) -> Inf + Infi

/Fotis