Traveling salesman

Given a complete oriented arc-weighted graph find a cycle that visits every vertex exactly once (a Hamiltonian cycle) and that has minimal length.

A dynamic programming algorithm in time \( O(n^2 2^n ) \)

Let w be the arc lengths. Let the vertices be numbered from 0 to n-1. Let vertex n-1 be a source. For every \( S \subseteq \{0,\ldots,n-2\} \) and every \( v\not\in S \) we want to find the shortest path from the source to v that traverses all vertices from S exactly once and only those. Let O[S][v] be this value.

For the base case we have \[ O[\emptyset][v] = w[n-1][v] \]

which is the length of the arc from the source to v. and for non empty set S \[ O[S][v] = \min_{u\in S} O[S \setminus\{v\}][u] + w[u][v]. \]

The goal is to compute \( O[\{0,\ldots,n-2\}][n-1] \).

The claimed complexity follows from the fact that there are \(O(n 2^n )\) variables and each needs time \(O(n)\) to be computed.

Note: in this algorithm we choose n-1 as the source and not say 0, which makes the set definitions easier.

The implementation

For these problems the instances are small, say about n=15. Then we can encode the sets into integers, using the binary decomposition of an integer as the characteristic vector of a set. For example to loop over all non empty sets \( S \subseteq \{0,\ldots,n-2\} \) we write

for S in range(1, (1<<n-1)):

and to test if a vertex u belongs to the set S we write

if (1<<u) & S: