Maximal Condorcet Domains: Trends & Structures

• Maximal Condorcet domains are inclusion-maximal sets of linear orders ensuring that majority voting remains acyclic across odd voter profiles.
• Their construction relies on local triple 'never' conditions, geometric tiling, pseudoline arrangements, and recursive compositions to maintain acyclicity.
• Recent advances offer structural classifications, precise enumeration for small n, and improved asymptotic lower bounds that deepen our understanding of these domains.

Maximal Condorcet domains are inclusion-maximal sets of linear orders whose induced majority relation remains acyclic for every odd profile supported on the domain. In the standard finite-alternative setting, a Condorcet domain is a subset of the symmetric group $S_n$ such that majority voting never creates a directed cycle; maximality means that adjoining any further order destroys this property (Danilov et al., 2010). This notion is distinct from a maximum Condorcet domain, which is one of largest possible cardinality for a fixed number of alternatives (Danilov et al., 2010). The modern theory combines local triplewise “never” conditions, weak Bruhat order, rhombus tilings, pseudoline arrangements, binary trees, and large-scale computation, and recent work has sharpened both structural classification and extremal lower bounds (Akello-Egwell et al., 2023, Karpov et al., 12 Jan 2026).

1. Basic definitions and local structure

A linear order on a finite set of alternatives $[n]=\{1,\dots,n\}$ is identified with a permutation $u\in S_n$, written in one-line notation as a ranking $u(1)\succ u(2)\succ \cdots \succ u(n)$ (Slinko, 23 Jan 2025). A Condorcet domain is a set $D\subseteq S_n$ such that whenever every voter’s preference belongs to $D$, the pairwise majority relation is always acyclic (Slinko, 23 Jan 2025). A maximal Condorcet domain is then a Condorcet domain that cannot be enlarged by adding any further ranking without losing the Condorcet property (Slinko, 23 Jan 2025). Equivalent formulations appear throughout the literature: a domain is maximal iff for every $v\notin D$, the union $D\cup\{v\}$ contains a Condorcet triple (Liversidge, 2020).

The standard local description is via restrictions to triples. Sen’s theorem, as used repeatedly in the cited papers, says that maximal Condorcet domains satisfy, on every triple of alternatives, one of the three “never” conditions: never-top, never-bottom, or never-middle (Liversidge, 2020). In Fishburn-style notation, for a triple $\{a,b,c\}$, a condition $xN_{\{a,b,c\}k}$ means that alternative $[n]=\{1,\dots,n\}$0 never occupies position $[n]=\{1,\dots,n\}$1 in the restriction (Li, 26 May 2025). This local language is not merely descriptive: it underlies classification, geometric constructions, and search algorithms (Zhou et al., 25 Sep 2025).

A persistent source of confusion is the distinction between maximal and maximum. The former is inclusion-maximality; the latter is cardinality-maximality (Leedham-Green et al., 2023). The papers are explicit that a domain can be maximal under inclusion without having maximum size (Danilov et al., 2010). This distinction is central in both combinatorial theory and computation, since many large constructive families are maximal yet not globally largest (Slinko, 2020).

2. Classical subclasses and canonical constructions

A major subclass is formed by peak-pit domains, in which every triple restriction satisfies either a never-top or a never-bottom condition (Li, 26 May 2025). This class contains classical single-peaked and single-dipped domains and includes many of the largest known examples for small $[n]=\{1,\dots,n\}$2 (Karpov et al., 2023). Arrow’s single-peaked domains, which satisfy a never-bottom condition on every triple, are especially rigid: any maximal Arrow single-peaked domain on $[n]=\{1,\dots,n\}$3 elements contains exactly $[n]=\{1,\dots,n\}$4 linear orders, has exactly two terminal elements, and admits recursive contraction and extension operations (Liversidge, 2020).

Fishburn’s alternating scheme is one of the central historical constructions. For each triple $[n]=\{1,\dots,n\}$5, one imposes a hump condition if $[n]=\{1,\dots,n\}$6 is even and a hole condition if $[n]=\{1,\dots,n\}$7 is odd, producing a maximal Condorcet domain of size $[n]=\{1,\dots,n\}$8 (Danilov et al., 2010). This family long served as the benchmark for large domains, but later work showed that it is not always largest, even within peak-pit or tiling-type classes (Slinko, 2020). Still, it remains structurally important and reappears in several generalized constructions (Karpov et al., 2023).

A more recent family is given by generalized Fishburn domains $[n]=\{1,\dots,n\}$9, defined by a subset $u\in S_n$0 and the rule that for every triple $u\in S_n$1, one imposes $u\in S_n$2 if $u\in S_n$3 and $u\in S_n$4 if $u\in S_n$5 (Slinko, 2023). It is proved that every generalized Fishburn domain is a maximal Condorcet domain (Slinko, 2023). This family contains classical single-peaked domains, single-dipped domains, and Fishburn’s alternating domains as special cases (Slinko, 2023).

Another constructive mechanism is composition. Danilov and Karzanov’s operation

$u\in S_n$6

builds larger Condorcet domains from smaller ones (Slinko, 23 Jan 2025). Iterating from singleton domains yields

$u\in S_n$7

a maximal Condorcet domain of cardinality $u\in S_n$8 (Slinko, 23 Jan 2025). A different composition, the never-last composition $u\in S_n$9, glues domains on $u(1)\succ u(2)\succ \cdots \succ u(n)$0 and $u(1)\succ u(2)\succ \cdots \succ u(n)$1 and comes with sufficient conditions ensuring that the result is again maximal (Keehan et al., 2024). These constructions show that maximal domains need not be isolated objects; many arise recursively from smaller pieces.

3. Geometric and combinatorial models

The most developed geometric model is the theory of Condorcet domains of tiling type. For a rhombus tiling $u(1)\succ u(2)\succ \cdots \succ u(n)$2 of a zonogon $u(1)\succ u(2)\succ \cdots \succ u(n)$3, each directed snake from the bottom vertex to the top vertex yields a linear order, and the set $u(1)\succ u(2)\succ \cdots \succ u(n)$4 of all compatible orders is a complete Condorcet domain (Danilov et al., 2010). The main theorem of the tiling paper states exactly that $u(1)\succ u(2)\succ \cdots \succ u(n)$5 is a complete Condorcet domain, where “complete” means inclusion-wise maximal (Danilov et al., 2010). The same paper proves that tiling domains unify earlier constructions of Abello, Galambos–Reiner, Chameni-Nembua, and maximal hump-hole domains (Danilov et al., 2010).

The tiling viewpoint is closely connected to weak Bruhat order, inversion sets, ideals, and separated set-systems. For a snake $u(1)\succ u(2)\succ \cdots \succ u(n)$6, the set of tiles to its left corresponds to $u(1)\succ u(2)\succ \cdots \succ u(n)$7, and the left-right order of snakes reproduces Bruhat order (Danilov et al., 2010). This gives maximal Condorcet domains a distributive-lattice flavor and explains why they often contain maximal chains in the Bruhat lattice (Danilov et al., 2010).

A recent geometric identification due to “A hidden Condorcet domain in Loday’s realisation of the associahedron” shows that the intersection of the permutohedron and Loday’s associahedron is exactly the iterated $u(1)\succ u(2)\succ \cdots \succ u(n)$8-domain

$u(1)\succ u(2)\succ \cdots \succ u(n)$9

and has exactly $D\subseteq S_n$0 elements (Slinko, 23 Jan 2025). These common vertices form a maximal symmetric never-middle Condorcet domain, equivalently the permutations avoiding the patterns $D\subseteq S_n$1 and $D\subseteq S_n$2 (Slinko, 23 Jan 2025). The proof uses a simple root-weight inequality in Loday’s coordinates to force the new maximum $D\subseteq S_n$3 to appear only at one of the two extremes, thereby recovering the recursive $D\subseteq S_n$4-construction (Slinko, 23 Jan 2025).

Pseudoline arrangements provide another representation. Ordinary arrangements represent peak-pit maximal Condorcet domains with maximal width, while generalized arrangements of pseudolines extend this to maximal Arrow single-peaked domains, which are peak-pit but do not have maximal width (Slinko, 2024). The relevant criterion is tameness: a generalized arrangement yields a Condorcet domain iff it is tame, and every maximal Arrow single-peaked Condorcet domain is representable in this way (Slinko, 2024). This suggests that combinatorial representation theory extends beyond the classical maximal-width setting.

The most recent broad synthesis replaces permutation-specific geometry by Coxeter-theoretic structure. In the Coxeter-group generalization, closed Condorcet domains correspond to Condorcet root posets, and this correspondence gives a unified language for properties such as maximality, connectedness, peak-pit, and tiling type (Gao et al., 1 Jun 2026). The paper states that these results are novel even in type $D\subseteq S_n$5, the ordinary symmetric-group case (Gao et al., 1 Jun 2026). This suggests a conceptual shift from individual constructions toward a general root-poset formalism.

4. Enumeration, extremal size, and the maximal-versus-maximum divide

Complete enumeration is currently available only for small numbers of alternatives. “Condorcet Domains of Degree at most Seven” gives the first explicit enumeration of all maximal Condorcet domains on $D\subseteq S_n$6 alternatives (Akello-Egwell et al., 2023). The numbers of non-isomorphic maximal unitary Condorcet domains are

$D\subseteq S_n$7

for $D\subseteq S_n$8, respectively (Akello-Egwell et al., 2023). The same paper proves that for $D\subseteq S_n$9 every maximum Condorcet domain is isomorphic to a Fishburn alternating-scheme domain, and in particular the maximum size for $D$0 is $D$1 (Akello-Egwell et al., 2023).

For $D$2, the exact maximum is known: there exists a Condorcet domain of size $D$3, this value is optimal, and the extremal domain is unique up to isomorphism (Leedham-Green et al., 2023). The paper states that there are no maximal Condorcet domains of size $D$4, and that the largest Condorcet domain containing both the identity permutation and its reverse has size $D$5, namely the Fishburn domain (Leedham-Green et al., 2023). The same work emphasizes that the size-$D$6 extremal domain is a first example of a maximum Condorcet domain that is not a peak-pit domain of maximal width (Leedham-Green et al., 2023).

Beyond $D$7, the exact maximum remains unknown, and the literature turns to lower bounds and constructive records. A 2026 paper improves the best known lower bounds for every $D$8, and then uses a separate construction to obtain current best lower bounds for $D$9 (Karpov et al., 12 Jan 2026). The improved values reported there include

$v\notin D$0

for $v\notin D$1 (Karpov et al., 12 Jan 2026). The same paper improves the asymptotic lower bound on the maximum size of a Condorcet domain to $v\notin D$2 (Karpov et al., 12 Jan 2026).

A complementary 2023 heuristic-search paper found new large Condorcet domains of size $v\notin D$3 for $v\notin D$4 and $v\notin D$5 for $v\notin D$6, improving previous records $v\notin D$7 and $v\notin D$8 (Zhou et al., 2023). Those domains are peak-pit, but the authors note that their local restriction structure differs fundamentally from Fishburn domains (Zhou et al., 2023). This supports the broader theme that maximum-cardinality examples may not come from a single simple family.

The distinction between maximum size and local structural optimality is sharpened in “Local Diversity of Condorcet Domains.” That paper introduces $v\notin D$9-abundance and shows that Black’s single-peaked domain is asymptotically optimal for local diversity, even though Condorcet domains can be much larger in cardinality (Karpov et al., 2024). It also shows that for some $D\cup\{v\}$0, there are maximal Condorcet domains with better local diversity parameters than maximum-cardinality domains (Karpov et al., 2024). A plausible implication is that the extremal landscape is genuinely multi-objective: inclusion-maximality, cardinality, and local richness diverge.

5. Connectivity, peak-pit structure, and majority behavior

A central recent result is the equivalence, for maximal domains, of three notions that had previously been studied separately. “Equivalence of Connected and Peak-Pit Maximal Condorcet Domains” proves that for maximal Condorcet domains, the following are equivalent: being peak-pit, being connected in the permutahedron, and being directly connected (Li, 26 May 2025). The same paper also shows that “peak-pit maximal Condorcet domain” and “maximal peak-pit Condorcet domain” coincide (Li, 26 May 2025). This resolves a longstanding structural ambiguity and turns local triple restrictions into a full graph-theoretic characterization at maximality.

The proof architecture is highly local. On three alternatives there are exactly three maximal Condorcet domains, corresponding to never-top, never-middle, and never-bottom (Li, 26 May 2025). Among these, the never-middle one is disconnected in the permutahedron, while the never-top and never-bottom ones are peak-pit (Li, 26 May 2025). From this, connectedness rules out never-middle restrictions in maximal domains and hence forces peak-pittedness (Li, 26 May 2025). Conversely, an induction based on geodesics of adjacent transpositions shows that maximal peak-pit domains are directly connected (Li, 26 May 2025).

The subclass of tiling-type domains enjoys even stronger majority properties. A 2025 paper proves that if the support of a voting profile is a Condorcet domain of tiling type, then the majority relation is a prelinear order with only simple ties (Reiner et al., 23 Sep 2025). More precisely, majority is the intersection of two linear orders determined by thresholding a monotone tally function on a heap poset (Reiner et al., 23 Sep 2025). This is substantially stronger than mere acyclicity and gives explicit formulas for uniform-vote majorities in several classical maximal tiling-type examples, including Fishburn’s alternating scheme and lexicographically first reduced words for $D\cup\{v\}$1 (Reiner et al., 23 Sep 2025).

The broader Coxeter-theoretic treatment strengthens this perspective. It states that strictly positive voting profiles on connected Condorcet domains yield outcomes with only simple ties, and that tiling type is equivalent to the combination of being maximal and connected, together with maximal width (Gao et al., 1 Jun 2026). Since the paper also states that a maximal Condorcet domain is connected if and only if it is peak-pit, it subsumes and extends several of the recent type-$D\cup\{v\}$2 results (Gao et al., 1 Jun 2026).

6. New subclasses, algorithms, and open directions

One important recent structural class is that of bipartite peak-pit domains. These are peak-pit domains for which some subset $D\cup\{v\}$3 of the alternatives supports an Arrow single-peaked restriction, while the complement supports the dual of an Arrow single-peaked restriction (Karpov et al., 2023). The paper states that bipartite peak-pit domains include the largest Condorcet domains for each $D\cup\{v\}$4, and that midpoint bipartite peak-pit domains include all maximum Condorcet domains for $D\cup\{v\}$5, though not the $D\cup\{v\}$6 optimum (Karpov et al., 2023). Set-alternating schemes inside this class always produce copious, connected, maximal peak-pit domains and improve the asymptotic lower bound for maximum size to growth exceeding $D\cup\{v\}$7 (Karpov et al., 2023).

Algorithmic generation has also advanced. “An orderly algorithm for generation of Condorcet Domains” gives an efficient orderly method for generating all non-isomorphic maximal Condorcet domains by searching in the space of never-condition systems rather than raw sets of linear orders (Zhou et al., 25 Sep 2025). It observes that every maximal Condorcet domain is full, and gives a maximality criterion in terms of implied never conditions on triples (Zhou et al., 25 Sep 2025). The same framework adapts to important subclasses and extends enumerations for peak-pit-derived classes on $D\cup\{v\}$8, with supplementary data for $D\cup\{v\}$9 (Zhou et al., 25 Sep 2025).

The state of the art therefore combines exact classification only in small degrees with progressively stronger asymptotic and structural results in general. Exact maxima are known only for $\{a,b,c\}$0 (Karpov et al., 12 Jan 2026). Fishburn-type constructions remain important but are no longer believed to capture all extremal behavior (Slinko, 2020). Recent data suggest that large domains are often peak-pit, connected, copious or highly abundant, and close to coherent in their never-condition patterns, but these are not yet full characterizations of maximum domains (Karpov et al., 12 Jan 2026). A plausible implication is that the next advances will come from combining structural representation theory, such as root posets or generalized arrangements, with exhaustive or heuristic search.

The current picture is therefore unusually rich. Maximal Condorcet domains can be tiny or very large, recursively composable or geometrically rigid, locally defined by never-conditions yet globally visible in associahedra, tilings, and Coxeter root systems. Their theory now includes exact small-$\{a,b,c\}$1 classification, multiple infinite maximal families, equivalences between local and graph-theoretic properties, and record asymptotic lower bounds, but the general maximum-size problem and a full structural characterization of large maximal domains remain open (Akello-Egwell et al., 2023, Karpov et al., 12 Jan 2026).