Re: MATLAB 5.3.1 (1999) and latest Octave math discrepancy ..

[Daniel J Sebald]

On 03/22/2013 02:54 PM, Will Flannery wrote:
> Hi, I'm running a long simulation of the Juno space probe that has about
> 4,000,000 iterations, the computer is a DELL laptop running Windows XP, and
> I get different trajectories using MATLAB and Octave.  The math used is the
> 4 basic operations plus squaring with the exponent operator and the sqrt
> function.  The differences occur after 5 significant digits and two million
> iterations, then the physics takes over and the trajectories diverge
> significantly.  I cut and pasted the program from MATLAB to Octave without
> and changes necessary.
>
> Is there any explanation for this?
>
> <http://octave.1599824.n4.nabble.com/file/n4651123/juno1.jpg>
>
> <http://octave.1599824.n4.nabble.com/file/n4651123/Figure13.jpg>

If you'd prefer to not disclose your code, you'll have to be more 
descriptive and help to isolate the problem.  Five significant digits of 
what?  After squaring?  After exponentiation?  After square rooting?

From the figures, it looks like your concern is that the trajectories 
appear to match up to a certain iteration then quickly accuracy is lost. 
 Is that correct?  The odd thing is that it doesn't seem to be the 
magnitude of displacement (distance) that is an issue because the point 
of divergence looks to be as a distance which was similar to that near 
the start.  Are you using absolute time in your algorithm in some way 
(that its value might become so large), or is it just delta time?

Dan