Obnoxious Facility Location Problem

• Obnoxious Facility Location Problem is a facility location optimization where facilities are deliberately sited far from users to mitigate undesirable externalities.
• Mechanisms employing threshold-majority rules and randomized corner selections ensure strategyproof allocation while approximating optimal social welfare under spatial constraints.
• Results demonstrate approximation ratios from 2 to 8, highlighting trade-offs in deterministic versus randomized approaches and spurring further research in mechanism design.

An obnoxious facility location problem is a class of facility location optimization in which facilities are deliberately sited to maximize their remoteness from users, as agents view the presence of the facility as an undesirable externality—such as landfill sites, noisy infrastructure, or hazardous installations. Formally, agent utility is an increasing function of distance to the facility: each agent prefers to be as far from the obnoxious facility as possible. Recent research has rigorously developed incentive-compatible, strategyproof, and approximately welfare-optimal location mechanisms, extending the theory across multiple facility settings, with constraints and agent heterogeneity (Han et al., 11 Aug 2025).

1. Formal Model: Agent Utility, Preferences, and Social Welfare

Let $N = \{1, \dots, n\}$ denote the agents, each with a private location $x_i \in [0,1]$ (the classical agent space). In the two-obnoxious-facility formulation, agents have a public optional preference $p_i = (p_{i1}, p_{i2}) \in \{0,1\}^2$ (require $p_{i1} + p_{i2} \ge 1$), encoding which of the two facilities affect them. The designer selects facility locations $y_1, y_2 \in [0,1]$, subject to a minimum separation constraint $|y_1 - y_2| \ge d$ with $d \in [0, 1]$.

Agent $i$'s utility is defined as

$$u_i((x_i, p_i), (y_1, y_2)) = \sum_{j=1}^2 p_{ij} |x_i - y_j|,$$

with expectation taken if the mechanism is randomized. Social utility is the aggregate:

$$SU(f \mid x, p) = \sum_{i=1}^n u_i((x_i, p_i), f(x, p))$$

where $f$ is either a deterministic or randomized facility location rule.

The approximation ratio of a mechanism $f$ is

$$\max_{(x,p)} \frac{SU(\mathrm{OPT} \mid x,p)}{\mathbb{E}[SU(f \mid x,p)]},$$

where $\mathrm{OPT}$ is the optimal placement of $F_1, F_2$ respecting $|y_1 - y_2| \ge d$ (Han et al., 11 Aug 2025).

2. Characterization of Truthful Mechanisms and Strategyproofness

Mechanisms are evaluated on strategyproofness (SP), ensuring agents cannot benefit from misreporting $x_i$, and on group strategyproofness (GSP), which prevents coalitions from all benefiting via a joint misreport:

• Deterministic mechanisms: Placement depends on publicly computed counts (e.g., majorities of affected agents), so no agent can influence facility locations to strictly improve her own utility. For the $d=0$ case, each facility is located at the endpoint furthest from the majority of affected agents; the threshold structure ensures SP (Han et al., 11 Aug 2025).
• Randomized mechanisms: If the outcome distribution is independent of reports, the mechanism is trivially GSP.

In both types, all manipulation attempts can only decrease an agent's distance sum under the prescribed rules, directly implying SP or GSP.

3. Mechanism Design and Approximation Results

The main results for two-obnoxious-facility location with optional preferences and minimum separation constraints on $[0,1]$ can be summarized as follows (Han et al., 11 Aug 2025):

Setting	Deterministic SP Approx Ratio	Randomized GSP Approx Ratio	Lower Bound Det	Lower Bound Rand
$d = 0$	$\le 4$	$\le 2$	$\ge 2$	$\ge 14/13$
$d > 0$	$\le 8$	$\le 2$	$\ge 2$	$\ge 14/13$

• For $d = 0$, a deterministic threshold-majority mechanism yields a ratio $\le 4$; a mechanism randomizing uniformly over all corners of $[0,1]^2$ achieves $\le 2$.
• For $d > 0$, a thresholded case analysis yields a deterministic mechanism with ratio $\le 8$; a heads/tails randomization between $(0,1)$ and $(1,0)$ achieves $\le 2$.
• Lower bounds of $2$ (deterministic) and $14/13$ (randomized) are achieved by two-agent worst-case instances.

Mechanism Structure Overview:

• Deterministic, $d=0$:
  • For each facility $F_j$, count affected agents to the left/right of $1/2$.
  • Place $F_j$ at endpoint opposite the majority. Guarantees no profitable deviation.
• Randomized, $d=0$:
  • Each possible corner $(0,0)$, $(0,1)$, $(1,0)$, $(1,1)$ is selected with equal probability (fully independent of reports).
• Deterministic, $d>0$:
  • Partition agents into those affected by both facilities ($B$) and those affected by only one ($A$).
  • Case analysis: serve the “double-affected majority” or place at opposite ends based on group sizes.
• Randomized, $d>0$:
  • Flip a coin to decide between endpoints $(0,1)$ and $(1,0)$.

All deterministic and randomized mechanisms leverage public partitioning and majority-style rules to maintain SP/GSP.

4. Robustness, Lower Bounds, and Hardness

Lower bounds are established via carefully crafted two-agent instances:

• Deterministic lower bound: For $n=2$, both agents caring about one facility, any deterministic SP mechanism must place in the interval between them, resulting in an approximation ratio of at least 2 after more refined “swap” argumentation.
• Randomized lower bound: For $n=2$ and separated agents, “swap-report” arguments force any randomized SP mechanism to have ratio at least $14/13$.

These lower bounds are tight up to constants for the cases analyzed; narrowing the interval between attainable upper and lower bounds remains open (Han et al., 11 Aug 2025).

5. Impact of Optional Preferences and Constraint Generalizations

Allowing agents to have optional preferences (the facility does or does not affect them) forces the mechanism to track distinct agent majorities for each facility, and special provisions to mix agents affected by one or both. The mechanism's approximation ratio and proof techniques build on refined partitions:

• $N_1 \setminus N_2$ (agents affected only by $F_1$),
• $N_2 \setminus N_1$ (agents affected only by $F_2$),
• $N_1 \cap N_2$ (agents affected by both).

The presence of a minimum separation $d$ significantly complicates the case analysis and degrades worst-case approximation guarantees (from 4 to 8 deterministically), as trade-offs between group welfare and SP constraints become more acute, especially when imposing spatial constraints (Han et al., 11 Aug 2025).

6. Open Questions and Research Directions

Key open problems and prospects include:

• Closing the gap between the best known randomized approximation ratio (2) and deterministic for $d=0, d>0$.
• Tightening the known lower bounds beyond $2$ and $14/13$.
• Exploring other objectives such as egalitarian welfare or quadratic distance aggregation.
• Extending models to settings with more facilities, continuous or networked domains, or richer agent preference structures (e.g., fractional optionality).
• Algorithms for more complex spatial or graph-based domains, and connections to distributed mechanism design (Han et al., 11 Aug 2025).

Advances in this domain contribute to the general field of approximate mechanism design without money and inform the optimal and truthful siting of facilities with negative externalities under complex agent and spatial constraints.