Competitive Analysis for Dynamic Multiperiod. Uncapacitated Routing Problems Enrico Angelelli and M. Grazia Speranza, страница 6

increasing function of x and L1|2.

When L2 L1, the optimal strategy is to serve the customers in∗(I) = L1C,11||22 +in the first time period, resulting in a total costL2 √= L1|2 + L2. z

If L1,1|2 ≤ (1 + 2)L1, then Algorithm IMMEDIATE is applied and it finds the optimal solution:√ rSMART(I) = 1.

       IfL1,1|2 > (1+         2)L1,thenAlgorithmDELAYisapplied.

We derive

rSMART(I) = z(DELAY,I) = L1 ++L1|2,2 = L1 ++L1|2 z∗(I)            L1|2           L2               L1|2           L2

L1

  In conclusion, we have≤ √                                                         rSMART=(I) ≤+√√2 for all instances

I. Thus, rSMART      2 when p              1             2. The tightness of the bound and the optimality of the algorithm come from

Theorem 1.                                                                               ■

Next, we consider what happens when we restrict ourselves to instances where customer locations are on the entire real line. Obviously, the lower bound provided in Theorem 1 is still valid.

Theorem 5.√   An optimal algorithm with competitive ratio

2 exists for instances with customer locations on the real line (RR+).

Proof. Observe that the vehicle has to pass the depot when going from a customer located on R− to a customer locatedonR+ andviceversa.Therefore,thelengthofanytour is the sum of the distances traveled on R− and R+. It follows that it is optimal to partition each customers set− and a subset ofC1, C1|2 and

C2 into a subset of customers located on R customerslocatedonR+ (customerslocatedattheoriginmay be included in either one of the sets), to compute the length of an optimal tour starting and ending at the depot and visiting

all customers in each subset, and to apply SMART√ (1 + √2)

on R− and SMART(1 +        2) on R+.                                      ■

3.2. The Euclidean Plane

A  lower bound on the competitive ratio of any algorithmfor DMPRP with customer locations in the Euclidean plane (R2) is given in the next theorem.

Theorem 6. The competitive ratio of any algorithm for DMPRP with customer locations in the Euclidean plane is greater than or equal to .

Proof. Consider the following instance. The depot is located at the origin O. Customers are located on three line segments emanating from O. The endpoints A, B, and C of the three line segments lie on the circle with radius . The distance between A and B as well as the distance between

B  and C is  (distance between A and C is approximately  small). The customers in C1, with |C1| = n, are distributeduniformlyonthelinesegmentOA.Halfofthecustomers in C1|2, with |C1|2| = 2n, are distributed uniformly on the line segment OB and the other half of the customers are distributed uniformly over the line segment OC. For large enough n (depends on ), the customers on a line segment are always visited consecutively in an optimal vehicle route, i.e., it never pays to zig-zag between customers on different line segments.

Consequently, any algorithm for DMPRP has three choices. The first choice is to serve only the customers in C1 in the first time period. In that case, let the customers in C2, with |C2| = n, be distributed uniformly on the line segment OA. The algorithm produces a solution with cost  AO plus OA  and OC + CO).

The optimal off-line solution has cost 2  BO plus OA ). The ratio tends to  tends to

0. The second choice is to serve the customers in C1 C1|2 in the first time period. In that case, let the customers in C2, with |C2| = n, be distributed uniformly on the line segment OC. The algorithm produces a solution with cost

(OA + AO and OB  plus OC + CO). The optimal off-line solution has cost 2BO plus OC+CO). The ratio tends to  when tends to 0. The final choice is to serve the customer in C1 and the customers in C1|2 on the line segment OB in the first time period. In that case, let

half of the customers in C2, with |C2| = 2n, be distributed uniformly on the line segment OA and the other half of the customers be distributed uniformly on the line segment OB. The algorithm produces a solution with cost (1