DATA

STRUCTURES AND
ALGORITHMS

CONTENT

• General diagram of branch and bound

• The traveling salesman problem

DATA STRUCTURES AND

3 4
 GENERAL DIAGRAM GENERAL DIAGRAM

• Branch and Bound: One of the methods to solve combinatorial optimization problems • Consider the problem of finding the
minimum of the objective function in
• Use the backtracking technique to list all options, thereby retaining the best option
which the solution is represented by a set
• Use bound evaluation (upper bound for the problem of finding max and lower bound for the
of variables X = (X1, X2, . . ., Xn).
problem of finding min) to cut down the search space during the listing process.
• The Try(k) function is used to try the
value for variable Xk during the listing
process Evaluate a lower bound
of the objective
function of the solution
• Symbol f*: objective function, the value evolving from this point

of the best solution found

5 6

GENERAL DIAGRAM THE TRAVELING SALESMAN PROBLEM

• After assigning the value v to Xk, we  Problem statement:

evaluate the lower bound g of the  A tourist wants to visit n cities 1, 2, . . ., n

objective function of the development  A itinerary (journey) is a way to start from city 1, go through all the remaining
cities, each city exactly once, and then return to starting city 1.
options continuing from point 13.
 Know that c(i, j) is the cost of traveling from city i to city j (i, j = 1, 2,..., n)
• If g is greater than or equal to f*, do not
 Find the itinerary with the smallest total cost.
continue developing the solution from
Some comments:
point 13 Evaluate a lower bound
of the objective • The number of traveler's itineraries is (n-1)!
function of the solution Sir William Rowan Hamilton [1]
developing from this • We have a one-to-one correspondence between a tourist's itinerary: (1805 – 1865)
point
• 1 ->x[2] -> x[3] ->...-> x[n]-> 1 with a permutation x = (x[2], x[3],..., x[n]) of n -1
Skip
natural number 2, 3,..., n.
• Journey cost: f(x) = c(1, x[2]) + c(x[2], x[3]) +... + c(x[n-1],x[n] ) + c(x[n] ,1)

7 8
 THE TRAVELING SALESMAN PROBLEM THE TRAVELING SALESMAN PROBLEM

 Solve using searching in the whole space: Init: Solve using branch and bound
• f* = +∞; f = 0; x[1] = 1;
 Itinerary : x = (1, x[2], x[3],..., x[n], 1) • for (int v = 2; v<=n; v++) visited[v]=0; Calculate bound:
• Let cmin = min { c(i, j) , i, j = 1, 2, ..., n, i  j } be the minimum traveling cost between cites
void Try(int k) {
try(k){//try values assignable to x[k]
for (int v = 2; v<=n; v++) { • Need to estimate the itinerary cost for the current branch corresponding to the part (1, u2, . . .,
for v in candidates(k) do { if (!visited[v]) {
x[k] = v; uk) passing through: 1  u2  . . .  uk-1  uk
if (check(v,k)) then {
visited[v] = 1;
x[k] = v; f = f + c(x[k-1],x[k]); • If the lower bound g(1, u2, . . ., uk) ≥ f* then skip (1, u2, . . ., uk)
Determine:
if (k == n) { //Update record
[Update the data structure D] 1) candidates(k)
ftemp = f + c(x[n],x[1]);
2) check(v, k)
if (k == n) then solution(); if (ftemp < f*) f∗ = ftemp;
}
else try(k+1); else Try(k + 1);
[Recover the data structure D] f = f − c(x[k-1],x[k]);
visited[v] = 0;
} }
} }
}
}

9 10

THE TRAVELING SALESMAN PROBLEM THE TRAVELING SALESMAN PROBLEM

• Need to estimate the itinerary cost for the current branch corresponding to the part (1, u2, . . ., • Function Try(k) finds the optimal solution to
uk) passing through: 1  u2  . . .  uk-1  uk the tourist problem using branch and bound Try(k) {
for v = 2 to n do {
• The cost for the partial itinerary (1, u2, ..., uk) is: techniques if not visited[v] {
x[k] = v;
 = c(1,u2) + c(u2, u3) + ... + c(uk-1, uk) visited[v] = true;
f = f + c(x[k-1],x[k]);
• For a full itinerary: if k = n then { //Update record
ftemp = f + c(x[n],x[1]);
1  u2  . . .  uk-1  uk  uk+1  uk+2  …..  un  1 if (ftemp < f*) f∗ = ftemp;
}
• We need n-k+1 more stages, each stage has a cost at least cmin, hence the minimum cost for else {
g = f + (n−k+1)∗cmin;
the remaining itinerary : (n-k+1) cmin Main(){ if g < f∗ then Try(k+1);
//Init: }
• If the partial itinerary is (1, u2, ..., uk) then the full itinerary has the cost at least g(1, u2, ..., uk) =  f* = +∞; f = 0; x[1] = 1; f = f − c(x[k-1],x[k]);
for v = 2 to n do visited[v] = false; visited[v] = false;
+ (n-k+1) cmin Try(2); }
print(f*); }
} }

11 12
THANK YOU !

13