On 3/31/2011 12:59 PM, Christian Froeschlin wrote:
>> I'm trying to find an efficient way to test all points within a set
>> against each other. Quadtrees/octrees seem like the way to go, but...
>> it appears that I need arrays containing vectors with other arrays
>> mixed in. POV doesn't allow that sort of thing.
>
> You might be better off exporting data from another programming
> language. Also, how to proceed might depend on whether your main
> focus is on non-intersection (where you only need to track close
> neighbors) or gravity simulation (where you need to consider
> points that are farther away as well. Better than efficient
> tests is of course if you can avoid having to test each
> point against every other point.

Currently all I want to do is find close neighbors. If I export that 
data I would probably have one array for the point vectors and another 
array as a list of pointers tracking said neighbors. I might even be 
able to use one of the Voronoi libraries, depending on how accessible 
the data is.

> Your scene reminds me a bit of astronomical simulations for
> star/galaxy formation and the likes.

The animation I posted was just an interesting side effect caused by a 
small adjustment. If I take it any further, I would keep the attractive 
force of each point localized, with a radius slightly larger than the 
repulsive force. In this way I could model certain fluids. I would then 
only really need to track the close neighbors...

> Their point counts are now
> over the 10 billion makr (i.e. 10^10) ;) Typically these use
> a hierarchical data structure, i.e. you group points into
> clumps of matter and over longer distances only consider
> the total attraction between increasingly larger clumps,
> ignoring the constituent points.

Wow, that sounds very interesting!

Post a reply to this message