“The alternative is to keep track of each change to values and undo the change when we hit a dead end. This is known as backtracking search. It makes sense when each step is a single change, but is usually too complicated to bother with when each change can lead to many other changes via constraint propagation.”

Which made me question (along with the previous paragraph I quoted about the search algorithm) if he was not having deep branching. i.e. that the choice was a single square being searched on.

I guess as far as he is concerned, it is only called backtracking in the case that you hold a list of the changes and have a master state, rather than having a list of copy states. I would have still called the DFS search backtracking, but I can live with his terminology so long as I’m clear.

Anyways, thanks.

A while ago I wrote myself a sudoku solver in scheme to explore the capabilities of amb (which is quite an elegant closure based implementation of backtracking). But there’s an assumption he (Norvig) makes that I don’t know is trivial.
Why should substituting only one square guarantee a solution? I refer to:
“Simple: first make sure we haven’t already found a solution or a contradiction, and if not choose one unfilled square and consider all its possible values. One at a time, try assigning the square each value, and searching from the resulting position.”

Maybe I’m not understanding what he means but it’s either one of two things:
a) a sudoku puzzle can be solved with only one guess cell. (if so: why?)
b) the DFS search algo he is proposing is just a backtracking algorithm since DFS will eventually ‘pop the stack’ which is essentially what backtracking is all about. It’s just an implementation detail.

Maybe I misread or am completely misunderstanding, anyone care to clarify?

Thank YOU!!!! At least one person understood what I was trying to convey

Ravi