[r-t] 147 TDMM

[Mark Davies]

I've studied Richard's description of the new search algorithm with 
interest. There seem to be no completely new innovations here, but the 
key breakthrough appears to be the separation of the tree search into 
two parts: first, the partitioning of the space into sets of 
mutually-true leads, and second, the linkage search, seeking to join the 
members of each set. Is this a fair summary?

I think this approach has been used before, in custom searches (Stedman 
Triples springs to mind), but I wonder if it could be more widely 
applied. It is obviously most useful when working on complete extents, 
but may have applications to non-extent searches too. I'm imagining a 
music-driven search in TD Major, for example, where the "plans" would be 
driven from the forced inclusion of certain musical nodes, and exclusion 
of others, with the majority of less-musical plans omitted. The 
difficulty would be establishing, in each case, whether the time taken 
to create N plans, and to run N linkage searches on each one, is less 
than the time taken to run a single conventional search.

Some interesting possibilities, though.

MBD