Example reduction We reduce the Hamiltonian path problem (HPP) to
the traveling salesman problem (TSP). We show this by an intermediate
reduction for the Hamiltonian cycle/circuit problem (HCP). HCP is almost
the same as HPP, only that we want a cycle now (i.e., the path should start
and end at the same vertex, i.e., one vertex is visited twice). Obviously, HCP
is also an NP problem (why?).
(a) Show HPP ≤p HCP.
Construction:
input: finite graph G = (V, E)
output: finite graph G0 = (V 0 , E 0 ) with
• V 0 = V ∪ {v0 }, v0 ∈
/ V , so we have the old vertices plus one extra
vertex v0
• E 0 = E ∪ {(v, v0 ) | v ∈ V }, so we have the old edges plus an edge
from every old vertex to v0 .
This construction is total (recognizing a proper input is simple) and
computable in polynomial time (we need to add one vertex and |V |
edges, possible in O(|V | + |E|)).
We need to show that G ∈ HPP ⇔ (G0 , k) ∈ HCP.
“⇒” If there is a Hamiltonian path in G, say, from v1 to v2 , then we
can close the cycle by taking the edge (v2 , v0 ) to v0 , followed by taking
the edge (v1 , v0 ) to v1 in G0 .
“⇐” If there is a Hamiltonian cycle G0 , then every vertex is visited ex-
actly once, especially v0 . We find a Hamiltonian path in G by removing
those two transitions from and to v0 on the cycle.
(b) Show HCP ≤p TSP.
Construction:
input: finite graph G = (V, E)
output: tuple (G0 , k), where
• G0 = (V, E 0 ) is a complete graph with E 0 has all edges from E
with weight 1 and all possible edges not in E with weight 2
• k = |V | is a natural number equal to the number of vertices.
5
This construction is total (recognizing a proper input is simple) and
computable in polynomial time (there are O(|V |2 ) possible edges).
We need to show that G ∈ HCP ⇔ (G0 , k) ∈ TSP.
“⇒” If there is a Hamiltonian cycle in G, then there is a tour in G0 of
weight k, namely the same cycle.
“⇐” If there is a tour in G0 of weight k, it means that every edge has
weight 1. This means these edges were already present in G, so we have
a Hamiltonian cycle in G, namely the same as the tour.
(c) We have shown HPP ≤p HCP ≤p TSP. Because HPP is known to be
NP-complete and thus NP-hard, we have shown that HCP and TSP
are NP-hard. Since HCP and TSP are NP problems, they are even
NP-complete.
Some NP-complete problems can be slightly modified and then polynomially
solved. For instance, the Eulerian path problem asks for a path through a
graph which visits each edge exactly once (compare to the Hamiltonian path
problem). This is polynomially solvable (linear in the size of the graph).
A typical way to tackle NP-complete problems in practice is finding good
heuristics. Remember that lower bounds only have to hold for some in-
stances, so many instances might be simple to solve.
There exist important NP-complete problems for which very good approx-
imation algorithms (e.g., the result is always at most 10% away from the
optimum) or probability algorithms (e.g., the result is correct in 90% of all
cases) are known.
§3 The satisfiability problem for Boolean expressions
(pp. 133-138)
We skip this part in the interest of time.
Summary: The “mother of all NP-complete problems”, SAT, is proven to be
NP-complete. The proof is not very hard, but very technical.
