Introduction
============

kdcount is a simple API for brute force pair counting, there is a C interface
and a Python interface. It uses a KDTree to prune
the K-D spatial data; for each pair within a given distance D, a callback
function is called; the user-defined callback function does the actual counting. 

Periodic boundary is supported, and it is non-trivial.

The python interface is more complicated, and powerful:

 * paircounting via :py:mod:`kdcount.correlate`
 * clustering via Friend-of-Friend algorithm; :py:mod:`kdcount.cluster`

The calculation can be made parallel, if :py:mod:`sharedmem` is installed.

The time complexity is :code:`O[(D / n) ** d]`, 
where n is number density. Each pair is opened. 

Note that
smarter algorithms exist (more or less, O(D / n log Dn), I may remembered it
wrongly See Gary and Moore 2001. 
The smarter algorithm is internally implemented, but not very much tested, and undocumented;
do not use it.

Unfortunately in cosmology we usually want to project the pair separation along
parallel + perpendicular direction relative to a given observer -- in this case,
the smarter algorithm become very difficult to implement. 

The spatial complexity is a constant, as we make extensive use of callback functions