Patterns work on the basis of bound fragments/molecules, and a molecule cannot be split up smaller than this - for instance in the examples above water cannot be represented by 'HO(1) H(1)' since this would 'neglect' a bond. However, there is nothing to stop a pattern 'crossing' individual molecules. Consider again the example 1) above. Three further (reiterating the point, equivalent) ways of writing the pattern description are:

Here, we have encompassed individual molecular entities into supermolecular groups, and as long as there are no bonds 'poking out' of the pattern, this is perfectly acceptable. Although this coarse-graining is a rather counter-intuitive way of forming patterns, it nevertheless allows them to be created for awkward systems such as that in 4) above. We may write the following valid patterns for this arrangement of atoms:

Note that, when automatically creating patterns, if Aten stumbles across a situation that confuses it, the default pattern of one supermolecule will be assumed, i.e. X(1) where 'X' is all atoms in the model. This will work fine in subsequent energy calculations etc., but will result in rather inefficient performance.