• the Hilbert numbers produced by C_i could be expanded to arbitrary precision in order to distinguish between C_i(p₁ ) and C_i(p₂ )

To use such values as dbm keys, we need them as fixed precision, so we limit expansion (to 4 levels).

Problems:

• gives keys that are 96 bytes long
  (producing very large Index files)

• different v_j can be mapped to same point
  (size of candidate sets is artificially inflated)

Solution:

• use a "prefix" index structure (something like a trie)