Okay. From this post I think that my original post did not explain the method I was trying to outline very well. Also, looking back at itubs reply, he also seems not to have fully understood what I was getting at.

I've put together a crude simulation using some randomly generated data, and it seems that using my method will process: 10,000 atoms, in around 17 seconds using < 6 MB memory. 20,000 atoms, in around 112 seconds using < 11 MB 30,000 atoms, in around 350 seconds using < 14 MB.

From these crude tests, I think it would handle 100,000 atoms in around 2 1/2 hours using 40 MB ram.

Update Ignore the figures above. I discovered a stupidity in my testcase that meant I was doing way more work than I needed to. I can now processing 100,000 atoms using 50MB in under 4 minutes.

This is without any attempt to really optimise things. I have no real feel for how this compares with other methods as I don't know what other methods are used.

I don't know for sure that I am processing the data correctly.

If you have a smallish (1000 or 2000 atom) dataset (with the desired results for a specified cutoff), that you could make available to me somehow?

Posting this here is probably not a good idea:), but I could let you have my email id.

Then I could see if what I am doing even approximates to giving correct results before I go posting triumphantly.