Distributed Facility Location Problem

• Distributed facility location is a framework where facilities are optimally placed based on local group summaries instead of full agent data.
• It quantifies efficiency loss through distortion metrics and employs methods like median and quantile mechanisms to ensure strategyproof outcomes.
• The approach extends to multi-facility settings and diverse objectives, with practical applications in scalable distributed algorithms over continuous and discrete spaces.

A distributed facility location problem concerns the selection of one or more facility locations to optimally serve a population of agents, where the agents are partitioned into groups (districts) and only local information is available within each group. The global facility decision is then based on summaries or representative locations reported by each group, rather than full access to all agent locations. This distributed or hierarchical information structure leads to inevitable loss of efficiency, which is captured by the notion of distortion: the worst-case ratio between the cost incurred by the distributed mechanism and the optimal cost obtainable with full information. The problem is studied across various settings on the real line and more general metrics, with single or multiple facilities, and under requirements such as strategyproofness—where agents are incentivized to report truthfully. This article surveys the principal models, objectives, algorithmic mechanisms, distortion bounds, and current frontiers in the distributed facility location literature.

1. Formal Models and Frameworks

In the canonical distributed facility location problem, $n$ agents each have a true location $x_i \in \mathbb{R}$ and are partitioned into $k$ disjoint groups (districts) $D_1, \ldots, D_k$, with $N_d$ denoting the set of agents in group $d$. The problem is to select location(s) for facility construction so as to minimize a global cost function of agent-facility distances, but with the restriction that the global decision can only use limited information: typically, each group submits a representative location (or a set of representatives), computed solely from the positions of its own members.

There are two principal variants:

• Single-facility: One facility location is selected from a finite or infinite candidate set $Z \subseteq \mathbb{R}$ (discrete or continuous case).
• Multi-facility: $k$ facilities are to be located, with constraints often including disjointness and selection from among agent positions or from a finite candidate set.

Mechanisms operate in two (or more) phases:

1. Local aggregation: Each group computes representative(s) $r_d \in Z$ depending only on its local agent data.
2. Global selection: The global mechanism selects among the representatives to determine the final facility location(s).

The classical objective is to minimize the total distance from all agents to the facility (social cost), but several generalizations are studied:

• Sum cost: $\mathrm{SC}(y \mid x) = \sum_i |x_i - y|$
• Max cost: $\mathrm{MC}(y \mid x) = \max_i |x_i - y|$
• Sum-of-max: $$\mathrm{SoM}(y) = \sum_{d} \max_{i \in N_d} |x_i - y|$$
• Max-of-sum: $$\mathrm{MoS}(y) = \max_{d} \sum_{i \in N_d} |x_i - y|$$

Agents may act strategically, potentially misreporting their positions; thus, interest also centers on strategyproof (SP) mechanisms, which guarantee that truthful reporting is a dominant strategy for all agents (Filos-Ratsikas et al., 2020, Filos-Ratsikas et al., 2023, Deligkas et al., 21 Jan 2026, Sun, 2024).

2. Distortion: Quantifying Inefficiency

Distortion is the central analytical tool for distributed facility location. Given mechanism $M$ and an instance $I = (x,D,Z)$, the distortion is defined as

$$\mathrm{dist}(I | M) = \frac{\mathrm{SC}(M(I) | x)}{\min_{y \in Z} \mathrm{SC}(y | x)}$$

The worst-case distortion of $M$ is the supremum over all instances. This quantifies the unavoidable price of limiting global decisions to group-level summaries and, if relevant, further restricts to mechanisms that are SP. Matching lower and upper bounds on distortion for various objectives and settings give a precise map of what distributed frameworks can achieve (Filos-Ratsikas et al., 2020, Filos-Ratsikas et al., 2023).

3. Distortion Bounds and Mechanisms: Single-Facility Case

Discrete Alternatives ($Z$ finite)

• District-Minimum Median (DMM): Each group selects a group-specific median of agent locations (minimizing sum of local distances, breaking ties leftmost), and the global mechanism outputs the median among all group representatives. For $k$ symmetric districts, $\mathrm{dist}(\mathrm{DMM}) \leq 3$.
• District-Median Median (DDM): Each group selects its local agent median (or closest candidate) as representative; output is the median of these. DDM is strategyproof and achieves $\mathrm{dist}(\mathrm{DDM}) \leq 7$.

Hard lower bounds:

• Any distributed (not necessarily SP) mechanism: distortion $\geq 3-o(1)$.
• Any SP (including ordinal) mechanism: distortion $\geq 7-o(1)$.

These bounds are tight: DMM is optimal among general mechanisms, DDM among SP mechanisms (Filos-Ratsikas et al., 2020).

Continuous Alternatives ($Z = \mathbb{R}$)

• Continuous District Median (CDM): Analog of DDM in continuous space—local medians are reported, and the facility is placed at the global median of group medians. CDM is strategyproof and achieves distortion $\leq 3$.
• Hard bounds: any distributed mechanism (not necessarily SP) has distortion $\geq 2-o(1)$; any SP mechanism has distortion $\geq 3-o(1)$. Thus, for SP mechanisms, the upper and lower bounds match, but a gap remains for general mechanisms ($2 \leq \mathrm{opt} \leq 3$) (Filos-Ratsikas et al., 2020).

4. Generalizations: Multiple Facilities and Other Objectives

Mechanisms and bounds extend to $k > 1$ facilities, more elaborate objectives, and constraints such as facility placement at agent positions or within specified candidate sets:

• For two-facility location with sum-variant cost (agent's cost is the sum of distances to both facilities), the optimal approximation ratio for strategyproof mechanisms is exactly $1+\sqrt{2}$ (Deligkas et al., 21 Jan 2026).
• For two-facility max-variant cost (agent's cost is the distance to the farthest facility), the optimal ratio for SP mechanisms is $9/2$.
• General $k$-facility sum-variant: $\leq 3 + 2/k$ (SP), with corresponding lower bound $3-2/k$.
• Max-variant cost: $\leq 2(k+1)$ (SP), with lower bound $2k$ (Deligkas et al., 21 Jan 2026).

A parametric family of "Quantile Mechanisms" for the two-facility constrained-distributed max-variant problem yields constant-factor distortion (e.g., distortion $3$ for max-of-max, $2+\sqrt{5}$ for max-of-average) under appropriate choices of local and global quantiles (Xu et al., 11 Aug 2025).

For group fairness or alternate objectives ("sum-of-max," "max-of-sum," etc.), tight distortion bounds are also established. For example, for sum-of-max in single-facility, the unrestricted distortion is $1$, but strategyproofness forces a jump to $1+\sqrt{2}$ (Filos-Ratsikas et al., 2023, Sun, 2024).

The following table summarizes several key distortion results (all deterministic, distributed):

Objective	Unrestricted	Strategyproof
Social Cost (Sum)	2	3
Max Cost (Max)	2	2
Sum-of-Max (SoM)	1	$1+\sqrt 2$
Max-of-Sum (MoS)	2	$1+\sqrt 2$

(Filos-Ratsikas et al., 2023, Filos-Ratsikas et al., 2020, Sun, 2024)

5. Strategyproofness: Mechanism Design Constraints

Strategyproof distributed mechanisms must ensure that no agent can reduce their cost by misreporting, even though the local aggregation process and group-level summarization provide leverage for manipulation. The main structural insight is that any SP mechanism with bounded distortion must be SP at both the local (within-group) and the global aggregation levels (Filos-Ratsikas et al., 2020, Filos-Ratsikas et al., 2023).

For sum objectives (e.g., minimizing total distance), reporting the local median is SP at the group level, and global selection via medians or quantiles preserves SP. For more general objectives or constraints, the mechanism must balance individual and collective incentives carefully, often via order-statistic–based rules tuned to the objective (e.g., the $(1-1/\sqrt{2})k$-leftmost-of-rightmost for sum-of-max cost) (Filos-Ratsikas et al., 2023).

Mechanisms for multi-facility settings generalize these order-statistic paradigms and demonstrate that "strategyproofness gap" (difference in optimal distortion achievable with and without strategyproofness) is often nontrivial (Deligkas et al., 21 Jan 2026, Xu et al., 11 Aug 2025).

6. Extensions: Discrete Candidates, Obnoxious Location, and Open Problems

Recent research has expanded the distributed facility location paradigm along several axes:

• Discrete candidate locations: In settings with a finite set of possible facility sites, distortion remains bounded (e.g., SP distortion $3$–$5$ for several objectives), but impossibility results arise for obnoxious objectives (where agents prefer the facility to be far away)—no SP mechanism with bounded distortion exists when only discrete sites are allowed (Sun, 2024).
• Obnoxious facility location: Impossible to design SP mechanisms with bounded distortion for sum-of-min or min-of-sum objectives when $Z$ is finite, or for group-SP mechanisms in $Z=[0,1]$ (Sun, 2024).
• Heterogeneous or constrained two-facility: A two-parameter quantile mechanism operating with group-wise local quantile representatives and global selection based on quantile orderings achieves tight upper and lower constant distortion bounds for various pooled and fair cost objectives (Xu et al., 11 Aug 2025).
• Asymmetric groups: If group sizes differ, distortion upper bounds scale with the ratio of largest to smallest group—continuous extensions remain an open problem (Filos-Ratsikas et al., 2020).
• Open directions: Gaps remain for multi-facility settings (e.g., between $3-2/k$ and $3+2/k$ for sum-variant cost), and for randomized or group-strategyproof mechanisms. Other priorities include extending to higher-dimensional metrics, more general networks, and alternative fairness or group-utility objectives (Filos-Ratsikas et al., 2020, Filos-Ratsikas et al., 2023, Sun, 2024).

7. Distributed Facility Location in Metric and Graph Models

Beyond the real line, distributed algorithms for metric facility location (with opening and connection costs, metric constraints, and even outlier models) focus on parallel/communication models such as CONGEST, MPC, and $k$-machine:

• Deterministic (1.861+$\epsilon$)-approximation in $O(n^{3/4}\log_{1+\epsilon}^2 n)$ rounds for metric uncapacitated facility location in the CONGEST model, via a distributed primal-dual simulation and iterative randomized selection (Briest et al., 2011).
• $O(1)$-approximation in expected $O(\log \log n)$ rounds on clique networks and bipartite networks, exploiting reduction to 2-ruling set computation and efficient probabilistic message dissemination, leveraging new lower bounds and ruling set constructions (Berns et al., 2013, Hegeman et al., 2013).
• Large-scale algorithms for facility location with outliers (robust facility location and with penalties) achieve $O(1)$-approximation in $O(\log \log \log n)$ or $\tilde{O}(n/k)$ rounds under explicit/implicit metric representations (Inamdar et al., 2018).

These distributed algorithms highlight that scalable, near-optimal facility location is achievable in high-throughput communication-constrained environments; many of the theoretical results on distortion in hierarchical or strategic settings inform the design and analysis of such protocols.

---

Key References:

• "Approximate mechanism design for distributed facility location" (Filos-Ratsikas et al., 2020)
• "Settling the Distortion of Distributed Facility Location" (Filos-Ratsikas et al., 2023)
• "Distributed Agent-Constrained Truthful Facility Location" (Deligkas et al., 21 Jan 2026)
• "Distributed Facility Location Games with Candidate Locations" (Sun, 2024)
• "Constrained Distributed Heterogeneous Two-Facility Location Problems with Max-Variant Cost" (Xu et al., 11 Aug 2025)
• "A Distributed Approximation Algorithm for the Metric Uncapacitated Facility Location Problem in the Congest Model" (Briest et al., 2011)
• "Super-Fast Distributed Algorithms for Metric Facility Location" (Berns et al., 2013)
• "Large-Scale Distributed Algorithms for Facility Location with Outliers" (Inamdar et al., 2018)
• "A Super-Fast Distributed Algorithm for Bipartite Metric Facility Location" (Hegeman et al., 2013)