Question

I was wondering if it’s possible to print a binary tree in breadth first order while using only O(1) space?

The difficult part is that one have to use additional space to memorize the next level to traverse, and that grows with n.

Since we haven’t place any limitation on the time part, maybe there are some inefficient (in terms of time) ways that can achieve this?

Any idea?

Answer 1

This is going to depend on some finer-grained definitions, for example if the edges have back-links. Then it’s easy, because you can just follow a back link up the tree. Otherwise I can’t think off hand of a way to do it without O(lg number of nodes) space, because you need to remember at least the nodes “above”.

Update

Oh wait, of course it can be done in O(1) space with a space time trade. Everywhere you would want to do a back link, you save your place and do BFS, tracking the most recent node, until you find yours. Then back up to the most recently visited node and proceed.

Problem is, that’s O(1) space but O(n^2) time.

Another update

Let’s assume that we’ve reached node n_i, and we want to reach the parent of that node, which we’ll call wlg n_j. We have identified the distinguished root node n_0.

Modify the breath-first search algorithm so that when it follows a directed edge (n_x,n_y), the efferent or “incoming” node is stored. Thus when you follow (n_x,n_y), you save n_x.

When you start the BFS again from n_0, you are guaranteed (assuming it really is a tree) that at SOME point, you will transition the edge (n_j,n_i). At that point you observe you’re back at n_i. You’ve stored n_j and so you know the reverse edge is (n_i,n_j).

Thus, you get that single backtrack with only two extra cells, one for n_0 and one for the “saved” node. This is O(1)

I’m not so sure of O(n^2) — it’s late and it’s been a hard day so I don’t want to compose a proof. I’m sure it’s O((|N|+|E|)^2) where |N| and |E| are the size of the sets of vertices and edges respectively.