Professional Documents
Culture Documents
H(T ), where R∗j is any request cover of the optimal R∗ . complete the DP table, the running time of Step 4 is O(T m2l ).
Since there are nl police teams used for NHS patrol, the total
Let Pi = Rj∗ − (R1G ∪ R2G ∪ . . . RiG ) (i = 1, 2, . . . , h) and running time of Algorithm ?? is O(T nl m2l ).
P0 = Rj∗ , meaning the requests that is covered by Rj∗ but not
by R1G ∪ R2G ∪ . . . RiG . Let ui = |Pi |, we have ui−1 ≥ ui and
u0 ≥ u1 ≥ u2 ≥ . . . ≥ ug = 0. Based on the greedy rule, 4 Proof of Theorem 2
we have that the number of request in set Qi = RG i is larger Proof. Let P opt = hpopt opt opt
1 , p2 , . . . , pnl i be the optimal
the number of request in Pi−1 = Rj∗ −(R1G ∪R2G ∪. . . Ri−1 G
), (OPT) patrol plan for each police team ak (1 ≤ k ≤ nl ) and
where ui−1 ≤ |Qi |. PVRopt is its PVR value. Let P G = hpG G G
1 , p2 , . . . , pnl i be the
∗
We will buy additional pages to include the appendix in the final patrol plan for each ak derived from Algorithm ??, PVRG k is
version if the paper is accepted. its PVR value at the kth iteration and PVRG nl is its final PVR
value. Let Akl = aξ . For PVRG
S
1≤ξ≤k k , we have
PVRG G
k − PVRk−1 = max (PVRG G
ξ − PVRk−1 ) (2)
aξ ∈Al \Ak−1
l
PVRopt − PVRG
k−1
≥ (3)
nl
Eq.(2) holds for that at the kth iteration, police team ak
has the largest team size among the remaining police team-
s Al \ Ak−1
l . Ineq.(3) holds for that at the kth iteration, ak
uses DP to construct the optimal patrol plan pk . Combining
with the following two facts that 1) the total police teams in
Al \ Ak−1
l is bounded by nl , each of the team size is decreas-
ing, and 2) the PVR function satisfies the law of diminish-
ing marginal returns, we have the total optimal marginal PVR
PVRopt − PVRG k−1 contributed by the remaining police team-
k−1
s Al \ Al is at most nl · (PVRG G
k − PVRk−1 ). Combining
with the Lemma 2 and Theorem 3 proposed in [Khuller et
al., 1999] for budgeted maximum coverage problem, we can
achieve this (1 − 1/e)-approximation factor.
5 Proof of Proposition 3
Proof. For each police team ak ∈ Ah used for HS patrol, he
has at most T sub-plans (i.e., |pk | ≤ T (Step 4). For each
sub-plan pwk (1 ≤ w ≤ |pk |), checking whether it is feasible
for grafting needs to consider ml NHSs (Step 5). If pw k is
feasible, computing its optimal NHS grafting plan pw,Gk has
the computational complexity of order O(T ml ). Thus, for
each ak , computing all of his feasible NHS grafting plans
needs O(T ml + T 2 ml ) = O(T 2 ml ) computations. Since
there are nh police teams for HS patrol, the total computation
complexity of Algorithm ?? is O(T 2 nh ml ).
References
[Khuller et al., 1999] Samir Khuller, Anna Moss, and
Joseph (Seffi) Naor. The budgeted maximum coverage
problem. Information Processing Letters, 70(1):39–45,
1999.
Load #l Random Uniform ASC PASC
2 0.14±0.007 0.11±0.002 0.194±0.008 0.23±0.010
4 0.18±0.009 0.13±0.004 0.193±0.004 0.30±0.010
6 0.22±0.011 0.14±0.005 0.193±0.003 0.35±0.010
8 0.26±0.011 0.14±0.006 0.192±0.003 0.40±0.010
10 0.30±0.011 0.15±0.005 0.192±0.004 0.43±0.013
12 0.33±0.011 0.16±0.007 0.191±0.002 0.46±0.013
14 0.36±0.017 0.16±0.009 0.190±0.003 0.50±0.016
16 0.39±0.017 0.17±0.007 0.189±0.002 0.53±0.014
18 0.42±0.016 0.17±0.008 0.187±0.001 0.56±0.014
10 0.45±0.018 1.170±0.007 0.186±0.001 0.58±0.017