4.1. A Lower Bound
The next theorem shows that for planning horizons with more than two time periods, the lower bound on the compet-√
itive ratio of any algorithm is strictly greater than 2.
Theorem 10. The competitive ratio of any algorithm for DMPRP with a planning horizon T > 2 and with customer locations on the non-negative real line is greater than or equal to 1.44.
Proof. Consider the following instance with three time periods, T = 3. The customer set C1 contains a single customer at distance 1 from the depot (located at the origin) and the customer set C1|2 contains a single customer at distance a > 1 from the depot.
First, we consider algorithm A of the class of algorithms that at the beginning of the first time period decide to visit the customers in C1 and C1|2. Now assume that the customer set C2|3 contains a single customer at distance a from the depot and that C3 = ∅. Then, z2a, while the off-line optimum is z∗(I) = 2 + 2a. Thus, rA a.
Next, we consider algorithm A of the class of algorithms that at the beginning of the first time period decide to visit only the customers in C1. Now assume that the customer set
Cdepot. If algorithm2|3 contains a customer at distanceA at the beginning of the second timeab, with b > 1, from the
period decides to visit the customers in C1|2 and C2|3, then assume that C3 contains a customer at distance ab. Then, z = 2 + 2ab + 2ab, while the off-line optimum is
zbeginning of the second time period decides to visit only the∗(I) = 2a + 2ab. Thus, rA ≥ 1a++2abab. If algorithm A at the 2customers in+2a+2ab, while the off-line optimum isC1|2, then assume that C3 = ∅ z∗(I) = 2+2ab.
Thus,Consequently, for any algorithmrA ≥ 1+1+a+abab. A we have
rA . (3)
We want to establish the maximum lower bound (3). Suppose thata isfixedandfocusonthelasttwotermsoftheminimum. Since the first of these is an increasing function of b while the last is a decreasing function of b, the maximum value of the minimum of these two terms is attained at equality, i.e., when
1 ++2ab = 1 ++a + ab
, a ab 1 ab
which implies b . Therefore,
rA .
The first term is an increasing function of a and the second term is a decreasing function of a. Therefore, the maximum value of the lower bound (3) is attained at a 2.575 (with b 1.881) and its value is approximately 1.44. Thus, rA ≥
1.44. ■
4.2. Analysis of Algorithms
Next, we turn our attention to the performance of algorithms IMMEDIATE and DELAY.
Theorem 11. The competitive ratio of Algorithm IMMEDIATE for instances with customer locations on the nonnegative real line is 2 for any length of the planning horizon.
Proof. Assume T is even. (The case T is odd is handled similarly.) Let I be an instance for which the maximum ratio z(IMMEDIATE,z∗(I) I) is attained. It is easy to see that z(IMMEDIATE,I) = L1,1|2 + L2|3 + ... + LT−1|T + LT.
Next, consider an optimal solution. In an optimal solution all customers in C1|2 have to be served in time periods 1 and 2. Therefore, the cost incurred in an optimal solution over time periods 1 and 2 is at least L1|2. More generally, for t even, all customers in Ct−1|t have to be served in time periods t − 1 and t in an optimal solution. Therefore, the cost incurred in an optimal solution over time periods t − 1 and t is at least
Lt−1|t. Consequently, z∗(I) ≥ L1|2 + L3|4 + ··· + LT−1|T.
Analogously, for t odd (t ≥ 3), all customers in Ct−1|t have to be served in time periods t−1 and t in an optimal solution. Since an optimal solution also incurs a cost of at least L1 in the first time period and at least LT in the last time period, we find that
z∗(I) ≥ L1 + L2|3 + ··· + LT−2|T−1 + LT.
Combining these two observations we find
2z∗(I) ≥ L1,1|2 + L2|3 + ··· + LT−2|T−1 + LT−1|T + LT.
This shows that Algorithm IMMEDIATE never incurs a cost that is more than twice that of an optimal solution. ■
Уважаемый посетитель!
Чтобы распечатать файл, скачайте его (в формате Word).
Ссылка на скачивание - внизу страницы.