Competitive Analysis for Dynamic Multiperiod. Uncapacitated Routing Problems Enrico Angelelli and M. Grazia Speranza, страница 9

Theorem 12. The competitive ratio of Algorithm DELAY for instances with customer locations on the non-negative real line is 2 for any length T of the planning horizon.

     Proof.    Similar to the proof of Theorem 11.                   ■

Finally, we turn to the performance analysis of SMART(p). We consider the case T = 3. We start by examiningoptimaloff-linealgorithms.Whentheplanninghorizon has three time periods, there are two decision points, since in the third time period we have to serve all remaining customers. Let i denote the decision to immediately serve all available customers, and let d denote the decision to delay all customers that can be postponed. Let the sequence id denote an algorithm that applies IMMEDIATE on the first day and DELAY on the second day. Similarly we define the other decision sequences ii, di, and dd.

Theorem 13. The off-line optimal solution z∗(I) for instances I with customers on the non-negative real line and a planning horizon of three time periods is given either by z(id,I) or z(di,I). Moreover, z(di,I) < z(id,I) if and only if L1 + L3 < min(L1|2,L2|3).

Proof. The decision sequences ii and dd are dominated by id, since z(ii,I) = L1,1|2 + L2|3 + L3 L1,1|2 + L2|3,3 = z(id,I) and

z(dd,I) = L1 + L1|2 + L2|3,3 L1,1|2 + L2|3,3 = z(id,I).

To prove the second part of the theorem, observe that z(id,I) = L1,1|2 + L2|3,3 = max(L1,L1|2) + max(L2|3,L3) and z(di,I) = L1 + L1|2,2|3 + L3 = L1 + max(L1|2,L2|3) + L3.

We compare the two non-dominated decision sequences by analyzing a number of cases:

•  If L1 L1|2 and L3 L2|3, then z(id,I) ≤ z(di,I) because z(id,I) = L1 + L3 L1 + max(L1|2,L2|3)+ L3 = z(di,I).

•  If L1 L1|2 and L3 < L2|3, then z(id,I) ≤ z(di,I) because z(id,I) = L1 + L2|3 ≤ L1 + max(L1|2,L2|3)+ L3 = z(di,I).

•  If L1 < L1|2 and L3 L2|3, then z(id,I) ≤ z(di,I) because z(id,I) = L1|2 + L3 L1 + max(L1|2,L2|3)+ L3 = z(di,I).

•  If L1 < L1|2 and L3 < L2|3, then z(id,I) ≤ z(di,I) if and only if min(L1|2,L2|3) ≤ L1 + L3 because z(id,I) = L1|2 + L2|3 ≤ L1 + max(L1|2,L2|3)+ L3 = z(di,I) ⇔ min(L1|2,L2|3) ≤ L1 + L3.

We conclude that z∗(I) = z(di,I) < z(id,I) if and only if

L1 + L3 < min(L1|2,L2|3).                                                            ■

Theorem    14.      The    competitive    ratio    of    Algorithm

SMART(p) for instances with customers on the non-negative real line and a planning horizon of three time periods is p+p p , with a best ratio of  for p = 2.

Proof. The performance of SMART(p) can be analyzed by studying its three possible decision sequences. There are only three decision sequences that need to be analyzed, because the decision sequence ii will never be applied by SMART.

Case dd. SMART(p) applies dd under the following conditions:

L1L|21,,21||23 >> pLpL11|2 that imply |  |                  1p2L1.

Since p > 1, the inequality L1 < L1|2 < L2|3 holds. For the case z(id,I) ≤ z(di,I), the ratio is

while for the case z(di,I) < z(id,I), the ratio is

rSMART

< L1 + L1|2 + L2|3 < L1 + L1|2 + pL1|2

                                    L1 + L2|3                          L1 + pL1|2

L1

                         <         |2 + pL1|2 < 1 + p.

                                   pL1|2                     p

Therefore,

+ p

rSMART(I).

p

Case di. SMART(p) applies di under the following conditions:

L1,1 2 > pL1L1

that imply

             L1|2,2|3 ≤ pL1|2                                         |                  | .

If z∗(I) = z(di,I), then rSMART(I) = 1. Thus, we consider the case for which the off-line optimum is z∗(I) = z(id,I) and the ratio is

3,3 rSMART(I)   .

If L2|3 < L1|2, then rSMART(I) ≤ L1 + L1+|2 + L2|3,3 ≤ L1 + L1|2 < 1 + p.

                               L1|2       L2|3,3                    L1|2                   p

Therefore,

L1 ++2pL1|2 L1 ++2p2L1 ≤ 1 ++2p2 .

     L1|2      pL1|2         pL1       p2L1           p        p2