💻 computer science

The Condorcet Dimension of Metric Spaces

This paper establishes that in two-dimensional metric spaces with Manhattan or infinity norms, the Condorcet dimension of proximity-based elections is bounded by 4, while also demonstrating that any set of voter preferences can be embedded into a sufficiently high-dimensional metric space for any $p$ -norm.

Original authors: Alexandra Lassota, Adrian Vetta, Bernhard von Stengel

Published 2026-08-07

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Alexandra Lassota, Adrian Vetta, Bernhard von Stengel

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are at a massive party where everyone has to pick a favorite song from a playlist. If there's one song that beats every other song in a head-to-head vote, that's the clear winner. But what if the music tastes are so mixed that no single song wins against all the others? Maybe Song A beats Song B, Song B beats Song C, but Song C beats Song A. It's a perfect loop of disagreement, and the party stalls. This is the heart of a famous puzzle in social science called the "Condorcet paradox."

To fix this, scientists ask a different question: instead of looking for one perfect winner, can we find a small "dream team" of songs? If this team is chosen, no single song outside the team can beat the whole group. This is called a "Condorcet winning set." The big mystery is: how big does this team have to be? In the worst-case scenario, could the team need to be half the size of the playlist? Or is there a magic limit where a tiny group is always enough to satisfy the crowd? This question matters because it helps us understand if democracy can ever find a stable, fair solution, or if we are doomed to endless cycles of disagreement.

Now, let's zoom in on a specific type of party: one where everyone's preferences are based on how "close" a candidate feels to them. Imagine a map where voters and candidates are dots. The closer a candidate is to a voter, the more they like them. This is the "spatial model" of voting. The researchers in this paper, Alexandra Lassota, Adrian Vetta, and Bernhard von Stengel, wanted to know: if everyone is living on a flat, two-dimensional map (like a piece of paper), how big does our "dream team" of candidates need to be to beat everyone else?

They discovered that if the map uses two specific ways of measuring distance—the "Manhattan norm" (like walking city blocks, where you can't cut diagonally) or the "infinity norm" (where you care most about the single biggest difference between you and a candidate)—the answer is surprisingly small. They proved mathematically that a team of just four candidates is always enough to form a winning set. No matter how many candidates there are or how the voters are scattered, you never need more than four to beat the rest of the field.

However, the paper also makes it clear that this isn't a magic trick that works everywhere. They show that in some two-dimensional scenarios, you definitely need at least two candidates; a single winner isn't guaranteed. They also point out that while they proved the limit is four for these specific maps, they don't know if the limit is actually three for all possible elections (even those on maps with more dimensions). They suspect it might be three, but they haven't proven it yet.

The researchers also tackled a different problem: if you have a messy list of preferences that doesn't look like it fits on a map, can you force it onto one? They showed that you can always squeeze any set of voter preferences into a high-dimensional space (a map with many axes) to make it work, and they gave a recipe to do this quickly on a computer. But the main takeaway for our two-dimensional world is the "Rule of Four": in a flat world where people vote based on proximity, a small squad of four is the ultimate safety net against chaos.

Technical Summary: The Condorcet Dimension of Metric Spaces

Problem Definition
The paper investigates the Condorcet dimension of an election, defined as the minimum cardinality of a Condorcet winning set. A Condorcet winning set $S \subseteq C$ (where $C$ is the set of candidates) is a subset such that for every candidate $j \notin S$ , more than half of the voters prefer at least one candidate in $S$ over $j$ . While Elkind, Lang, and Saffidine [16] established that a Condorcet winning set of size at most $\lceil \log m \rceil$ always exists (where $m$ is the number of candidates), the exact upper bound for the Condorcet dimension in general elections remains an open question. The authors note that while instances with Condorcet dimension 3 are known, no instances with dimension greater than 3 have been found, leading to a conjecture that the dimension is at most 3 in general.

The paper focuses specifically on the spatial model of voting, where voters and candidates are located in a metric space, and preferences are determined by proximity (closer candidates are preferred). The study restricts attention to 2-dimensional metric spaces ( $D=2$ ) under specific norms, contrasting with the general case where embedding dimensions may need to be high ( $D = \min[m-1, n]$ ).

Methodology
The authors employ a combination of graph-theoretic arguments and geometric analysis within specific metric spaces:

Graph-Theoretic Framework: The problem is modeled using a majority digraph (a tournament where an arc $(j, i)$ exists if a majority of voters prefer $i$ over $j$ ). A Condorcet winning set corresponds to a dominating set in this digraph. The authors leverage the property that a dominating set in a tournament of size $m$ has size at most $\lceil \log m \rceil$ .
Geometric Construction (Manhattan Norm): For the 2-dimensional Manhattan norm ( $L_1$ $L_{1}$ ), the authors partition the plane into four quadrants based on the median coordinates of the voters (shifted to the origin). They define a candidate set $S = \{c_1, c_2, c_3, c_4\}$ $S = {c_{1}, c_{2}, c_{3}, c_{4}}$ , where each $c_i$ $c_{i}$ is the candidate in quadrant $Q_i$ $Q_{i}$ closest to the origin.
- They prove via case analysis that for any candidate $c$ in a quadrant, the candidate $c_i$ closest to the origin in that same quadrant is preferred by at least half the voters (specifically those in the adjacent or opposite quadrants).
- This establishes that $S$ is a dominating set in the majority digraph.
Geometric Transformation (Infinity Norm): For the 2-dimensional Infinity norm ( $L_\infty$ ), the authors utilize a geometric rotation argument. They demonstrate that rotating the coordinate axes by 45 degrees transforms the $L_\infty$ distance metric into a scaled $L_1$ (Manhattan) distance metric. Consequently, the preference structures are preserved, allowing the $L_1$ result to apply directly to the $L_\infty$ case.
Lower Bound Construction: To establish that the dimension is not always 1, the authors construct a specific 2-dimensional instance with 3 voters and 3 candidates under any $p$ -norm ( $p \ge 1$ ) that results in a Condorcet cycle, proving the dimension is at least 2.
Embedding Constructions: The paper addresses the inverse problem: embedding arbitrary voter preferences into a metric space. It provides polynomial-time constructions for embedding any set of strict preferences into a $p$ -norm metric space of dimension $D = \min[m, n]$ . This extends previous results known for Euclidean ( $L_2$ ) and Manhattan ( $L_1$ ) norms to general $p$ -norms.

Key Contributions and Results

Upper Bound for $D=2$ : The primary result is that in a 2-dimensional metric space under either the Manhattan norm or the Infinity norm, the Condorcet dimension is at most 4. This is a significant improvement over the general logarithmic bound for these specific spatial settings.
Lower Bound for $D=2$ : The authors prove that for any $p$ -norm in 2 dimensions, there exist election instances where the Condorcet dimension is at least 2 (i.e., a Condorcet winner does not necessarily exist).
General Embedding Theorems:
- Theorem 7: Any election with $m$ candidates and $n$ voters can be embedded in a metric space of dimension $D=m$ for any $p$ -norm ( $p \ge 0$ ) in polynomial time.
- Theorem 8: Any such election can be embedded in a metric space of dimension $D=n$ for any $p$ -norm ( $p > 1$ ) in polynomial time.
- These constructions generalize the findings of Bogomolnaia and Laslier [7] and Chen et al. [11] to arbitrary $p$ -norms.

Significance and Claims
The paper claims to provide the first constant upper bound on the Condorcet dimension for specific, low-dimensional spatial models of voting. By restricting the embedding dimension to $D=2$ and utilizing natural norms (Manhattan and Infinity), the authors demonstrate that the complexity of finding a winning set is bounded by a small constant (4), rather than growing logarithmically with the number of candidates.

The authors explicitly state that their results do not resolve the general conjecture that the Condorcet dimension is at most 3 for all elections (where $D$ is unbounded). They acknowledge that recent independent work by Charikar et al. [10] and Nguyen et al. [28] has established a constant upper bound of 5 (improved from 6 and 16) for general elections using probabilistic sampling and minimax theorems. The significance of this paper lies in its geometric approach to the $D=2$ case, which offers a tighter bound (4) for spatial models and clarifies the relationship between metric dimension and the existence of small Condorcet winning sets.

The paper concludes by identifying the gap between the proven lower bound of 2 and the upper bound of 4 for $D=2$ as an open problem, as well as the general question of whether the dimension is universally bounded by 3.

Technical Summary: The Condorcet Dimension of Metric Spaces

More like this