Chapter 23

Minimum Spanning Trees

Sun-Yuan Hsieh
謝孫源 教授
成功大學資訊工程學系
Overview

Problem
A town has a set of houses and a set of roads.
A road connects 2 and only 2 houses.
A road connecting houses u and v has a repair cost w(u, v).
Goal: Repair enough (and no more) roads such that
1. everyone stays connected: can reach every house from all
other houses, and
2. total repair cost is minimum.

2
 Overview

Model as a graph:
Undirected graph G = (V, E)
Weight w(u, v) on each edge (u, v)  E
Find T  E such that
1. T connects all vertices (T is a spanning tree), and
2. w(T )   w(u , v) is minimized.
( u ,v )T

3
 Overview

A spanning tree whose weight is minimum over all

spanning trees is called a minimum spanning tree, or
MST.
Example of such a graph [edges in MST are shaded] :
8 8
b d g
10 2
7
e 5
a 9 i
9 3
3
12 11
c f h
1 6

In this example, there is more than one MST. Replace edge (e, f)
by (c, e).
4
 Growing a minimum spanning tree

Some properties of an MST:

It has |V|1 edges.
It has no cycles.
It might not be unique.

Building up the solution

We will build a set A of edges.
Initially, A has no edges.
As we add edges to A, maintain a loop invariant:
Loop invariant: A is a subset of some MST.

5
Growing a minimum spanning tree

Add only edges that maintain the invariant. If A is a subset of

some MST, an edge (u, v) is safe for A if and only if A  {(u,
v)} is also a subset of some MST. So we will add only safe
edges.

Generic MST algorithm

GENERIC-MST(G, w)
1. A  Ø
2. while A is not a spanning tree
3. do find an edge (u, v) that is safe for A
4. A  A  {(u, v)}
5. return A
6
Use the loop invariant to show that this generic
algorithm works.
Initialization: The empty set trivially satisfies the loop
invariant.
Maintenance: Since we add only safe edges, A remains a
subset of some MST.
Termination: All edges added to A are in an MST, so when we
stop, A is spanning tree that is also an MST.

7
Finding a safe edge
How do we find safe edges?
Let’s look at the example. Edge (c, f ) has the lowest weight of
any edge in the Graph. Is it safe for A = Ø?

Intuitively: Let S  V be any set of vertices that includes c but

not f (so that f is in V  S). In any MST, there has to be one
edge (at least) that connects S with V  S. Why not choose the
edge with minimum weight? (Which would be (c, f ) in this
case.)

8
Some definitions: Let S  V and A  E.
A cut (S, VS) is a partition of vertices into disjoint sets S and
VS.
Edge (u, v)  E crosses cut (S, VS) if one endpoint is in S and
the other is in VS.
A cut respects A if and only if no edge in A crosses the cut.
An edge is a light edge crossing a cut if and only if its weight
is minimum over all edges crossing the cut. For a given cut,
there can be > 1 light edge crossing it.

9
 Theorem
Let A be a subset of some MST, (S, VS) be a cut that respects A,
and (u, v) be a light edge crossing (S, VS). Then (u, v) is safe for A.

Proof Let T be an MST that includes A.

If T contains (u, v) , done.
So now assume that T does not contain (u, v) . We’ll construct
a different MST T that includes A  {(u, v)}.

Recall: a tree has unique path between each pair of vertices. Since T is an
MST, it contains a unique path p between u and v. Path p must cross the
cut (S, VS) at least once. Let (x, y) be an edge of p that crosses the cut.
From how We chose (u, v) , must have w(u, v)  w(x, y)
10
[Except for the dashed edge (u, v), all edges shown are in T. A is some
subset of the edges of T, but A cannot contain any edges that cross the cut
(S,
VS), since this cut respects A. Shaded edges are the path p.]

S
u
x

v y
VS

11
Since the cut respects A, edge (x, y) is not in A.
To from T from T:
Remove (x, y) . Breaks T into two components.
Add (u, v) . Reconnects.
So T = T  {(x, y)} {(u, v)}.
T is a spanning tree.
w(T) = w(T) w(x, y) + w(u, v)
 w(T)
since w(u, v)  w(x, y) .
Since T is a spanning tree, w(T)  w(T), and T is an MST, then
T must be an MST.
12
Need to show that (u, v) is safe for A:
A  T and (x, y)  A  A  T
A  {(u, v)}  T
Since T is an MST, (u, v) is safe for A.

13
 GENERIC-MST:

A is a forest containing connected components. Initially,

each component is a single vertex.
Any safe edge merge two of these components into one.
Each component is a tree.
Since an MST has exactly |V|1 edges, the for loop
iterates |V|1 times.
Equivalently, after adding |V|1 safe edges, we’re down
to just one component.

14
 Corollary
If C = (VC, EC) is a connected component in the forest GA = (V, A)
and (u, v) is a light edge connecting C to some other component in
GA (i.e., (u, v) is a Light edge crossing the cut (VC, VVC)), then (u,
v) is safe for A.

Proof Set S = VC in the theorem.

This naturally leads to the algorithm called Kruskal’s

algorithm to solve the Minimum-spanning-tree problem.

15
 Kruskal’s algorithm

G = (V,E) is a connected, undirected, weighted graph.

w: E  
Starts with each vertex being its own component.
Repeatedly merges two components into one by choosing the
light edge that connects them (i.e., the light edge crossing the cut
between them).
Scans the set of edges in monotonically increasing order by
weight.
Uses a disjoint-set data structure to determine whether an edge
connects vertices in different components.

16
 KRUSKAL(V, E, w)
1. AØ
2. for each vertex v  V[G]
3. do MAKE-SET(v)
4. sort E into nondecreasing order by weight w
5. for each (u, v) taken from the sorted list
6. do if FIND-SET(u)  FIND-SET(v)
7. then A  A  {(u, v)}
8. UNION(u,v)
9. return A

17
Run through the above example to see how Kruskal’s algorithm
works on it:
(c, f) : chosen
(g, i) : chosen
(e, f) : chosen 8 8
b d g
(c, e) : reject 10 2
(d, h) : chosen 7
5
(f, h) : chosen a 9 e  9 i
3 3
(e, d) : reject 
(b, d) : chosen 12 11
c f h
(d, g) : chosen 1 6
(b, c) : reject
(g, h) : reject
(a, b) : chosen
18
At this point, we have only one component, so all other
edges will be rejected.
[We could add a test to the main loop of KRUSKAL to stop
once |V|1 edges have been added to A.]

Get the shaded edges shown in the figure.

Suppose we had examined (c, e) before (e, f). Then,

would have found (c, e) safe and would have rejected
(e, f).

19
Analysis
Initialize A: O(1)
First for loop: |V|MAKE-SETS
Sort E: O(E lgE)
Second for loop: O(E) FIND-SETS and UNIONS

20
Assuming the implementation of disjoint-set data structure,
already seen in chapter21, that uses union by rank and path
compression: O((V + E)(V)) + O(E lg E)
Since G is connected, |E|  |V|  1  O(E (V)) + O(E lg E).
(|V|) = O(lg V) = O(lg E)
Therefore, total time is O(E lg E).
|E|  |V|2  lg |E| = O(2 lg V) = O(lg V)
Therefore, O(E lg V) time. (If edges are already sorted,
O(E(V)), which is almost linear.)

21
 Prim’s algorithm

Builds one tree, so A is always a tree.

Starts from an arbitrary “root” r.
At each step, find a light edge crossing cut (VA, VVA),
where VA= vertices that A is incident on. Add this edge to
A. V
A

V  VA

Light edge
[Edges of A are shaded.]
22
How to find the light edge quickly?
Use a priority queue Q:
Each object is a vertex in VVA.
Key of v is minimum weight of any edge (u, v) , where u  VA.
Then the vertex returned by EXTRACT-MIN is v such that
there exists u  VA and (u, v) is light edge crossing (VA, VVA).
Key of v is  if v is not adjacent to any vertices in VA.

23
The edges of A will form a rooted tree with root r:
r is given as an input to the algorithm, but it can be any vertex.
Each vertex knows its parent in the tree by the attribute [v]=
parent of v. [v] = NIL if v = r or v has no parent.
As algorithm progresses, A = {(v, [v]) : v  V{r}Q}.
At termination,
V A  V  Q  , so MST is A  {(v,  [v]) : v  V  {r}}.

24
PRIM(V, E, w, r)
1. QØ
2. for each u  V[G]
3. do key[u]  
4. [u]  NIL
5. INSERT(Q, u)
6. DECREASE-KEY(Q, r, 0)
7. while Q  Ø
8. do u  EXTRACT-MIN(Q)
9. for each v  Adj[u]
10. do if v  Q and w(u, v) < key[v]
11. then [v]  u
12. DECREASE-KEY(Q, v, w(u, v) )
25
Run through the above example to see how Prim’s algorithm works on it:
Vertex c has been chosen as a starting point.
8 8
b d g
10 2
7
5 aifceghdb
a 9 e 9 i
3 3 11
12
10
82∞016358
12 11
bgehda 12
8∞
58
7
6
3
9 12
∞
0 ca
c f h
root 1 6
idg 11
∞
9
7 ∞
9
3 ∞
1 ∞ g
(c, f) : chosen (d, g) : chosen
eb fi
(e, f) : chosen (g, i) : chosen ∞ ∞
(f, h) : chosen (b, d) : chosen
h i
(d, h) : chosen (a, b) : chosen
26
Analysis
Depends on how the priority queue is implemented:
Suppose Q is a binary heap.
Initialize Q and first for loop: O(V lg V)
Decrease key of r: O(lg V)
while loop: |V| EXTRACT-MIN calls
 O(V lg V)
 |E| DECREASE-KEY calls
 O(E lg E)
Total: O(E lgV)

27
Suppose we could do DECREASE-KEY in O(1) amortized time.
Then  |E| DECREASE-KEY calls take O(E) time altogether
 total time becomes O(V lg V + E)
In fact, there is a way to do DECREASE-KEY in O(1)
amortized time: Fibonacci heaps, in chapter20.

28