Thanks, Alex. Let me take a one dimensional example to show the situation.

Suppose we have a one dimensional atomic structure:

...---0---1---2---3---4---5---6---...

Initially, 2---3---4 are the defect atoms forming a point defect (let's denote these atoms as set I).

After some deformation, atoms 1---2---3---4---5 become defect atoms (i.e, the number of atoms belonging to a point defect may fluctuate by deformation or local distortion). Of course, there are also completely new point defects formed on this chain. Let's denote all these defect atoms in this deformed configuration as set N.

By the previous discussed procedure (select, freeze and compute property), we can only identify 2---3---4 as the initial point defect (i.e., the intersection of set I and N), thus atom 1 and atom 5 stand out as individual defect atoms (they are not new point defect but still related to the initial point defect).

My question is how to assign these individual atoms such as 1 and 5 to the initial point defect, thus reducing the 'noise' of true new point defects.

My thought was to perform a cluster analysis on the atom set N-I, then those 'noise' atoms should have very small cluster size and the true new point defects should have relatively large cluster size. This may allow us to filter out the 'noise' atoms. But I am not sure if/how this could be done.

In your last reply, did you mean the coordination of atoms 2---3---4 can be directly obtained using 'Compute Property' modifier?

Thanks,

qjli