Abstract:We study the two-center problem on cactus graphs in facility locations, which aims to place two facilities on the graph network to serve customers in order to minimize the maximum transportation cost. In our problem, the location of each customer is uncertain and may appear at $O(m)$ points on the network with probabilities. More specifically, given are a cactus graph $G$ and a set $\calP$ of $n$ (weighted) uncertain points where every uncertain point has $O(m)$ possible locations on $G$ each associated with a probability and is of a non-negative weight. The problem aims to compute two centers (points) on $G$ so that the maximum (weighted) expected distance of the $n$ uncertain points to their own expected closest center is minimized. No previous algorithms are known for this problem. In this paper, we present the first algorithm for this problem and it solves the problem in $O(|G|+ m^{2}n^{2}\log mn)$ time.
What problem does this paper attempt to address?
The paper attempts to solve the two - center problem for uncertain points on the cactus graph. Specifically, the problem aims to place two facilities (centers) in a given cactus - graph network to minimize the maximum transportation cost. The challenge here is that the location of each customer is uncertain and may appear at one of \(O(m)\) points on the network, and each location has a corresponding probability.
### Problem Description
Given:
- An undirected cactus graph \(G=(V, E)\)
- A set of \(n\) uncertain points \(P = \{P_1, P_2,\ldots, P_n\}\), where each uncertain point \(P_i\) may appear at \(m\) locations \(p_{i1}, p_{i2},\ldots, p_{im}\) on the graph \(G\), each location has a probability \(f_{ij}\) and a non - negative weight \(w_i\)
Objective:
Find two points (centers) \(q_1^*\) and \(q_2^*\) on the graph \(G\) such that the maximum of the weighted expected distances from all uncertain points to their nearest centers is minimized. That is:
\[
\min_{q_1, q_2\in G}\max_{P_i\in P} w_i\cdot\min\left(E[d(P_i, q_1)], E[d(P_i, q_2)]\right)
\]
where \(E[d(P_i, q)]=\sum_{j = 1}^{m}f_{ij}\cdot d(p_{ij}, q)\) represents the expected distance from the uncertain point \(P_i\) to the point \(q\).
### Research Background and Contribution
Previously, there has been some research on similar problems on tree networks, but on cactus graphs, especially when each uncertain point has multiple possible locations (i.e., \(m>1\)), there are no known algorithms. This paper proposes the first algorithm to solve this problem, with a time complexity of \(O(|G|+m^2n^2\log mn)\).
### Solution Overview
1. **Vertex - Constrained Version**: Transform the problem into a vertex - constrained version, that is, restrict each location to the vertices of the graph.
2. **Decision Problem**: Given a value \(\lambda\), determine whether two centers can be placed on the graph such that the objective value does not exceed \(\lambda\).
3. **Tree Representation**: Convert the cactus graph into its tree representation, where each node uniquely represents a cycle, a hinge, or a non - cycle vertex.
4. **Binary Search**: Perform at most two "binary searches" on the tree representation to find the edge or cycle that must contain the peripheral center.
5. **Feasibility Check**: Determine whether two centers can be placed on the found peripheral - center edge to cover all uncertain points.
Through these steps, the optimal objective value \(\lambda^*\) can be finally determined and the corresponding two center locations can be found.