Algorithms Question

[noahgibbs]

If you're not into programming, you're probably happier if you skip this.

Say you've got a graph. It's not a tree (i.e. it has cycles), and it's fully-connected (no subsets of it that aren't connected to the main graph). You've got the list of nodes and the list of connections.

You'd like to find the longest shortest path. That sounds silly, so lemme 'splain. For any two nodes there's a shortest path between them. If you were to look at every pair of nodes and find the shortest path, you'd like the longest of those.

So if you had a graph like this:

  A--o--o--o--o--o--o--o
           |           |
           o--o--o--o--o
                 |
           B--o--o

The longest shortest path would be between A and B, so length 9. The longest shortest path would *still* go the short way 'round the cycle (because it's a shortest path), so it would be 9 and not 13.

You're not guaranteed that there *are* any leaf nodes (remember, this isn't a tree, there are cycles), so a simple Minimum Spanning Tree won't help so much...

This stuff is for a nifty network connectivity problem that a friend is working on. I've written a little statistical simulator for these things, and I'm trying to find the longest shortest path in a 30,000-node graph in less than O(n^2) time...

Comments

Re:

[r-transpose-p.livejournal.com]

Also, note, i lied when i said n²

"E" is the total number of edges in the graph, not the outgoing number per node. So make that n²log(n) for your 3 outgoing edges per node problem. (the log is from the use of the heap for a priority queue in BFS, and the E is a nastily approximate upper bound of the number of things that can be in the priority queue at a given time)