Branch and Bound

Searching Strategies

Updated: 12/27/2010

1
Feasible Solution vs.
Optimal Solution
 DFS, BFS, hill climbing and best-first
search can be used to solve some
searching problem for searching a
feasible solution.
 However, they cannot be used to solve
the optimization problems for searching
an (the) optimal solution.

2
The branch-and-bound
strategy
 This strategy can be used to
solve optimization problems
without an exhaustive search in
the average case.

3
 Branch-and-bound strategy
 2 mechanisms:

 A mechanism to generate branches when

searching the solution space

 A mechanism to generate a bound so that

many braches can be terminated

4
 Branch-and-bound strategy
 It is efficient in the average case because
many branches can be terminated very
early.
 Although it is usually very efficient, a very
large tree may be generated in the worst
case.
 Many NP-hard problem can be solved by
B&B efficiently in the average case;
however, the worst case time complexity is
still exponential. 5
A Multi-Stage Graph Searching
Problem.
Find the shortest path from V0 to V3

6
E.G.:A Multi-Stage Graph Searching Problem

7
Solved by branch-and-bound
(hill-climbing with bounds)

A feasible solution is found whose cost is equal to 5.

An upper bound of the optimal solution is first found here.
8
 For Minimization Problems

Upper Bound
(for feasible solutions)
 Usually, LB<UB.
Optimal
 If LBUB, the expanding node can be terminated.

Lower Bound
(for expanding nods)

0
9
 For Maximization Problems

Upper Bound
(for expanding nods)
 Usually, LB<UB.
Optimal
 If LBUB, the expanding node can be terminated.

Lower Bound
(for feasible solutions)
0
10
The traveling salesperson
optimization problem
 Given a graph, the TSP Optimization problem
is to find a tour, starting from any vertex,
visiting every other vertex and returning to
the starting vertex, with minimal cost.
 It is NP-hard.
 We try to avoid n! exhaustive search by the
branch-and-bound technique on the average
case. (Recall that O(n!)>O(2n).)

11
The traveling salesperson
optimization problem
 E.g. A Cost Matrix for a Traveling Salesperson Problem.

j 1 2 3 4 5 6 7
i
1 ∞ 3 93 13 33 9 57
2 4 ∞ 77 42 21 16 34
3 45 17 ∞ 36 16 28 25
4 39 90 80 ∞ 56 7 91
5 28 46 88 33 ∞ 25 57
6 3 88 18 46 92 ∞ 7
7 44 26 33 27 84 39 ∞
12
The basic idea
 There is a way to split the solution space
(branch)
 There is a way to predict a lower bound for a
class of solutions. There is also a way to find
a upper bound of an optimal solution. If the
lower bound of a solution exceeds the upper
bound, this solution cannot be optimal and
thus we should terminate the branching
associated with this solution.

13
Splitting
 We split a solution into two groups:
 One group including a particular arc
 The other excluding the arc
 Each splitting incurs a lower bound and
we shall traverse the searching tree
with the “lower” lower bound.

14
 The traveling salesperson
optimization problem
 The Cost Matrix for a Traveling Salesperson Problem.

Step 1 to reduce: Search each row for the smallest value

j 1 2 3 4 5 6 7 to j
i
1 ∞ 3 93 13 33 9 57
2 4 ∞ 77 42 21 16 34
3 45 17 ∞ 36 16 28 25
4 39 90 80 ∞ 56 7 91
from i
5 28 46 88 33 ∞ 25 57
6 3 88 18 46 92 ∞ 7
7 44 26 33 27 84 39 ∞
15
Step 2 to reduce: Search each column for the smallest value
The traveling salesperson
optimization problem

Reduced cost matrix:
j 1 2 3 4 5 6 7
i
1 ∞ 0 90 10 30 6 54 (-3)
2 0 ∞ 73 38 17 12 30 (-4)
3 29 1 ∞ 20 0 12 9 (-16)
4 32 83 73 ∞ 49 0 84 (-7)
5 3 21 63 8 ∞ 0 32 (-25)
6 0 85 15 43 89 ∞ 4 (-3)
7 18 0 7 1 58 13 ∞ (-26)
reduced:84

A Reduced Cost Matrix. 16

The traveling salesperson
optimization problem
j 1 2 3 4 5 6 7
i
1 ∞ 0 83 9 30 6 50
2 0 ∞ 66 37 17 12 26
3 29 1 ∞ 19 0 12 5
4 32 83 66 ∞ 49 0 80
5 3 21 56 7 ∞ 0 28
6 0 85 8 42 89 ∞ 0
7 18 0 0 0 58 13 ∞
(-7) (-1) (-4)
Table 6-5 Another Reduced Cost Matrix.
17
Lower bound
 The total cost of 84+12=96 is
subtracted. Thus, we know the lower
bound of feasible solutions to this TSP
problem is 96.

18
The traveling salesperson
optimization problem
 Total cost reduced: 84+7+1+4 = 96 (lower bound)
decision tree:

The Highest Level of a Decision Tree.

 If we use arc 3-5 to split, the difference on the lower
bounds is 17+1 = 18. 19
Heuristic to select an arc to split the
solution space
 If an arc of cost 0 (x) is selected, then
the lower bound is added by 0 (x) when
the arc is included.
 If an arc <i,j> is not included, then the
cost of the second smallest value (y) in
row i and the second smallest value (z)
in column j is added to the lower bound.
 Select the arc with the largest (y+z)-x
20
We only have to set c4-6 to be .
For the right subtree
(Arc 4-6 is excluded)
j 1 2 3 4 5 6 7
i
1 ∞ 0 83 9 30 6 50
2 0 ∞ 66 37 17 12 26
3 29 1 ∞ 19 0 12 5
4 32 83 66 ∞ 49 ∞ 80
5 3 21 56 7 ∞ 0 28
6 0 85 8 42 89 ∞ 0
7 18 0 0 0 58 13 ∞

21
 For the left subtree
(Arc 4-6 is included)
j 1 2 3 4 5 7
i
1 ∞ 0 83 9 30 50
2 0 ∞ 66 37 17 26
3 29 1 ∞ 19 0 5
5 3 21 56 7 ∞ 28
6 0 85 8 ∞ 89 0
7 18 0 0 0 58 ∞
A Reduced Cost Matrix if Arc 4-6 is included.
1.4th row is deleted.
2.6th column is deleted.
3.We must set c6-4 to be . (The reason will be clear later.)
22
For the left subtree

The cost matrix for all solution with arc 4-6:
j 1 2 3 4 5 7
i

1 ∞ 0 83 9 30 50
2 0 ∞ 66 37 17 26
3 29 1 ∞ 19 0 5
5 0 18 53 4 ∞ 25 (-
6 0 85 8 ∞ 89 0 3)
7 18 0 0 0 58 ∞
A Reduced Cost Matrix for that in Table 6-6.
 Total cost reduced: 96+3 = 99 (new lower bound)
23
Upper bound
 We follow the best-first search scheme
and can obtain a feasible solution with
cost C.
 C serves as an upper bound of the
optimal solution and many branches
may be terminated if their lower bounds
are equal to or larger than C.

24
 1 2
5
6

4 3
7
Fig 6-26 A Branch-and-Bound Solution of a Traveling Salesperson
Problem. 25
 Preventing an arc
 In general, if paths i1-i2-…-im and j1-j2-…-jn have
already been included and a path from im to j1 is to
be added, then path from jn to i1 must be
prevented (by assigning the cost of jn to i1 to be
)
 For example, if 4-6, 2-1 are included and 1-4 is to
be added, we must prevent 6-2 from being used
by setting c6-2=. If 6-2 is used, there will be a
loop which is forbidden.
26
The 0/1 knapsack problem
 Positive integer P1, P2, …, Pn (profit)
W1, W2, …, Wn (weight)
M (capacity)
n
maximize PX
i i
i1
n
subject to WiXi M Xi =0or 1, i =1, …, n.
i1
Theproblemismodified:
n
minimize PX
i i
i1
27
The 0/1 knapsack problem

Fig. 6-27 The Branching Mechanism in the Branch-and-Bound

Strategy to Solve 0/1 Knapsack Problem.
28
How to find the upper bound?
 Ans: by quickly finding a feasible
solution in a greedy manner: starting
from the smallest available i, scanning
towards the largest i’s until M is
exceeded. The upper bound can be
calculated.

29
The 0/1 knapsack problem
 E.g. n = 6, M = 34

i 1 2 3 4 5 6

Pi 6 10 4 5 6 4

Wi 10 19 8 10 12 8

(Pi/Wi  Pi+1/Wi+1)

 A feasible solution: X1 = 1, X2 = 1, X3 = 0, X4 = 0,
X5 = 0, X6 = 0
-(P1+P2) = -16 (upper bound)
Any solution higher than -16 can not be an optimal solution.
30
How to find the lower bound?
 Ans: by relaxing our restriction from Xi = 0 or 1
to 0  Xi  1 (knapsack problem)
n
Let   Pi X i be an optimal solution for 0/1
i 1
n
knapsack problem and   Pi X i be an optimal
i 1
solution for fractional knapsack problem. Let
n n
Y=   Pi X i , Y =
’
  Pi X i .
i 1 i 1
 Y’  Y
31
 The knapsack problem
 We can use the greedy method to find an optimal solution
for knapsack problem.
 For example, for the state of X1=1 and X2=1, we have
X1 = 1, X2 =1, X3 = (34-6-10)/8=5/8, X4 = 0, X5 = 0, X6 =0
-(P1+P2+5/8P3) = -18.5 (lower bound)
-18 is our lower bound. (We only consider integers, since
the benefits of a 0/1 knapsack problem will be integers.)

32
How to expand the tree?
 By the best-first search scheme
 That is, by expanding the node with the
best lower bound. If two nodes have
the same lower bounds, expand the
node with the lower upper bound.

33
0/1 Knapsack Problem Solved by Branch-and-Bound Strategy 34
 Node 2 is terminated because its lower
bound is equal to the upper bound of
node 14.
 Nodes 16, 18 and others are terminated
because the local lower bound is equal
to the local upper bound.
(lower bound  optimal solution  upper
bound)
35
 The A* algorithm
 Used to solve optimization problems.
 Using the best-first strategy.
 If a feasible solution (goal node) is selected to expand, then it is
optimal and we can stop.
 Estimated cost function of a node n : f(n)
f(n) = g(n) + h(n)
g(n): cost from root to node n.
h(n): estimated cost from node n to a goal node.
h*(n): “real” cost from node n to a goal node.
f*(n): “real” cost of node n
h(n)  h*(n) Estimated further cost should never
 f(n) = g(n) + h(n)  g(n)+h*(n)exceed the………….
= f*(n) real further
(1) cost.
36
 Reasoning
 Let t be the selected goal node. We have f*(t)=f(t)
+h(t)=f(t)+0=f(t)…..(2)
 Assume that t is not the optimal node. There must exist
one node, say s, that has been generated but not selected
and that will lead to the optimal node.
 Since we take the best first search strategy, we have f
(t)f(s)……(3).
 We have f*(t)=f(t)f(s)f*(s) by Eqs. (1), (2) and (3),
which means that s is not the node leading to the optimal
node. Contradiction occurs.
 Therefore, t is the optimal node.
37
 The A* algorithm
 Stop when the selected node is also a goal node. It is
optimal iff h(n)h*(n)
 E.g.: To find a shortest path from node s to node t

38
The A* algorithm
 Step 1.

g(A)=2 h(A)=min{2,3}=2 f(A)=2+2=4

g(B)=4 h(B)=min{2}=2 f(B)=4+2=6
g(C)=3 h(C)=min{2,2}=2 f(C)= 3+2=5

39
The A* algorithm
 Step 2. Expand A

g(D)=2+2=4 h(D)=min{3,1}=1 f(D)=4+1=5

g(E)=2+3=5 h(E)=min{2,2}=2 f(E)=5+2=7
40
The A* algorithm
 Step 3. Expand C

g(F)=3+2=5 h(F)=min{3,1}=1 f(F)=5+1=6

g(G) =3+2=5 h(G)=min{5}=5 f(G) =5+5=10
41
 The A* algorithm
 Step 4. Expand D

g(H)=2+2+1=5 h(H)=min{5}=5 f(H)=5+5=10

g(I)=2+2+3=7 h(I)=0 f(I)=7+0=7
42
The A* algorithm
 Step 5. Expand B

g(J)=4+2=6 h(J)=min{5}=5 f(J)=6+5=11

43
 The A* algorithm
 Step 6. Expand F

I is selected to expand.
The A* algorithm stops,
since I is a goal node.

g(K)=3+2+1=6 h(K)=min{5}=5 f(K)=6+5=11

g(L)=3+2+3=8 h(L)=0 f(L)=8+0=8
44
The A* Algorithm
 Can be considered as a special type of
branch-and-bound algorithm.
 When the first feasible solution is found,
all nodes in the heap (priority queue) are
terminated.
 * stands for “real”
 “A* algorithm” stands for
“real good algorithm”
45
Q&A
46