Political Equilibrium Singularity

Equilibrium-Invariant Institutional Limits in Dynamic Political Games

Benny Dunavich

Abstract

Dynamic political games may admit multiple strategically distinct equilibria without generating multiple long-run institutional outcomes. This paper develops a general framework for analyzing that distinction. For a dynamic political game Γ, a pre-specified equilibrium correspondence 𝖤(Γ), an admissible domain of initial conditions 𝒟, and a pre-specified institutional classification map ι, define the Equilibrium Institutional Image

Ψ∞𝖤,ι(Γ;𝒟)=⋃σ∈𝖤(Γ)⋃s0∈𝒟ι[ℒΓ,σ(s0)],

where ℒΓ,σ(s0) is the essential long-run support generated by equilibrium σ from s0. A Political Equilibrium Singularity (PES) obtains when this image collapses to a single institutional equivalence class:

Ψ∞𝖤,ι(Γ;𝒟)={C⋆}.

The framework distinguishes PES from equilibrium uniqueness, local stability of representative states, historical persistence, stochastic equilibrium selection, and welfare evaluation. It defines a non-degenerate form in which distinct equilibrium outcome laws coexist with a common long-run institutional image, and a projected form in which complete institutional outcomes remain heterogeneous while a pre-specified institutional component is invariant. General results establish monotonicity in the admissible domain and equilibrium set, a sufficient uniform-contraction condition, a symmetry obstruction, and separation between distinct robust PES regions.

The paper does not claim that common limiting outcomes across multiple equilibria are a new phenomenon. Related structures appear in asynchronous coordination, implementation theory, dynamic political economy, institutional-change theory, and stochastic-selection models. Its proposed contribution is instead a general classification framework that treats equilibrium-invariant long-run institutional concentration as a domain-relative, equilibrium-concept-relative, and institutional-resolution-relative property of the equilibrium correspondence itself.

1. Introduction

Political equilibrium multiplicity and political institutional multiplicity are not the same object.

A dynamic political game may support several equilibria that differ in actions, coalition structures, policies, transition paths, or distributions of payoff, while all eventually generate the same institutional architecture. Conversely, a game with a unique equilibrium may sustain a recurrent process containing several institutional classes. The cardinality of the equilibrium set therefore does not determine the cardinality of the long-run institutional outcome set.

The distinction suggests a question that is different from ordinary equilibrium selection:

What is the complete long-run institutional image generated by the entire admitted equilibrium correspondence?

Let Γ denote a dynamic political game and 𝖤(Γ) a pre-specified equilibrium correspondence. Let 𝒟⊆S be an admissible domain of initial states. For each equilibrium σ∈𝖤(Γ) and initial state s0∈𝒟, let

ℒΓ,σ(s0)

denote the essential long-run support of the induced process. Let

ι:S→ℭ

map dynamic states into a pre-specified space ℭ of institutional equivalence classes.

The central object of the paper is

Ψ∞𝖤,ι(Γ;𝒟)=⋃σ∈𝖤(Γ)⋃s0∈𝒟ι[ℒΓ,σ(s0)].(1)

A Political Equilibrium Singularity at C⋆∈ℭ obtains when

PES𝖤,ι(Γ,𝒟;C⋆)⇔Ψ∞𝖤,ι(Γ;𝒟)={C⋆}.(2)

The term singularity refers only to singletonity of the long-run institutional image. It does not imply a technological singularity, a physical discontinuity, normative superiority, or unconditional historical inevitability.

Several distinctions follow immediately.

First, PES is not equilibrium uniqueness. Multiple equilibrium outcome laws can coexist with a singleton institutional image. The paper calls this Non-Degenerate PES (ND-PES).

Second, PES is not local stability. Local attraction of one representative state does not exclude other equilibria or admissible basins with different institutional limits. Conversely, PES need not imply local asymptotic stability of any particular representative state inside the target institutional class.

Third, PES is not historical persistence. A realized history observes one path, whereas Ψ∞ quantifies over the admitted equilibrium correspondence and initial domain.

Fourth, PES is not stochastic selection. A perturbation may select one recurrent outcome from a non-singleton unperturbed image. That selected outcome does not thereby become a PES target of the unperturbed game.

Finally, PES contains no welfare criterion. A uniquely reproduced institutional class may be efficient, Pareto dominated, adaptive, or rigid. Those properties require additional evaluative structure.

The framework is explicitly relative to three analytical choices: the equilibrium concept 𝖤, the admissible domain 𝒟, and the institutional resolution encoded by ι. These dependencies are not nuisances to be suppressed. They define the scope of a PES claim.

The paper develops four refinements. ND-PES separates multiplicity of equilibrium labels from multiplicity of realized outcome laws. Robust PES asks whether the singleton property survives in an open parameter neighborhood. Stochastically robust PES additionally asks whether vanishing perturbations concentrate invariant probability on the already-established PES class. Projected PES asks whether a pre-specified institutional component is singular even when complete institutional classes remain multiple.

The analytical contribution is deliberately limited. Existing work already demonstrates common limiting predictions across multiple equilibria, universal restrictions over equilibrium outcomes, endogenous political-institutional convergence, and selection under perturbations. The proposed contribution of PES is to place these questions at a common level of aggregation: the long-run institutional image of the complete admitted equilibrium correspondence.

The paper proceeds as follows. Section 2 positions the framework relative to neighboring literatures. Section 3 defines the dynamic environment and Equilibrium Institutional Image. Sections 4–6 develop PES, non-degeneracy, robustness, and projected singularity. Section 7 presents general results and obstructions. Section 8 gives minimal examples and counterexamples. Section 9 addresses identification and falsification. Section 10 concludes and states the research agenda. Appendices provide formal proofs, selected mechanism constructions, the notation registry, and the boundary with companion frameworks.

2. Relation to Existing Literatures

PES intersects several established literatures. The purpose of this section is not to argue that common limiting outcomes are new, but to isolate what changes when the analytical object becomes the institutional image of the entire equilibrium correspondence.

2.1 Dynamic coordination and all-equilibrium limiting predictions

An early direct antecedent is Lagunoff and Matsui (1997), who study repeated coordination under asynchronous choice. In their canonical alternating-move coordination environment, sufficiently patient players obtain a unique Perfect-equilibrium payoff despite the dynamic strategic environment. Their result identifies asynchronous inertia as a source of equilibrium-prediction restriction and provides an early connection between asynchronous coordination and the anti-folk logic relevant here.

Lagunoff and Matsui (2001) develop a related robustness result. For asynchronous play and a fixed sufficiently high discount factor, they identify an open neighborhood of a pure coordination game in which every Perfect equilibrium generates payoffs arbitrarily close to the Pareto-dominant payoff of the benchmark game. Common all-equilibrium prediction and robustness of that prediction therefore both have established antecedents.

Ambrus and Ishii (2015) provide a particularly close comparison to the motivating PES configuration. They study asynchronous infinite-horizon coordination games and derive conditions under which play is ultimately absorbed in the Pareto-dominant stage-game equilibrium for every Markov perfect equilibrium in their two-player environment, while also deriving broader common-limit results.

PES does not claim these phenomena as new. It instead introduces an institutional-image operator that makes the domain of initial conditions, institutional quotient, and complete equilibrium correspondence explicit:

𝖤(Γ)↦Ψ∞𝖤,ι(Γ;𝒟).

A common-limit result in a particular coordination model can therefore be interpreted as one possible instance of singletonity of that image.

2.2 Implementation theory

Implementation theory supplies an even more direct antecedent for universal quantification over equilibria. Mookherjee and Reichelstein (1990), for example, formulate Bayesian implementation in terms of mechanisms whose Bayesian equilibrium outcomes agree with a specified choice rule. Lee and Sabourian (2011) define repeated implementation so that the equilibrium set is nonempty and every equilibrium outcome path produces the prescribed social choice at every relevant history.

The logical resemblance is substantial:

all admitted equilibria→restricted outcome set.

The distinction lies in analytical direction. Implementation is a design problem: a target choice rule is given and a mechanism is sought whose equilibria implement it. PES begins with a given dynamic game and asks what long-run institutional classes its admitted equilibria generate.

PES also need not require identical equilibrium paths or complete outcomes. Distinct equilibrium outcome laws may differ substantially while sharing one long-run institutional class, or one projected institutional component.

Implementation theory therefore blocks any claim that universal equilibrium-to-target restrictions are themselves new. PES uses this logical structure for a different classification object.

2.3 Dynamic political economy and institutional change

Dynamic political economy provides the closest substantive setting. Duggan and Kalandrakis (2012) study infinite-horizon legislative policy making in which current policy determines the next period’s status quo. They prove existence of stationary Markov perfect equilibria, characterize equilibrium-induced transition probabilities, and derive conditions concerning invariant distributions in a canonical spatial setting.

Acemoglu, Egorov, and Sonin (2012) develop a framework in which current collective decisions alter future political power and characterize dynamically stable states as functions of initial conditions under their assumptions. Greif and Laitin (2004) develop a dynamic account of endogenous institutional stability and change based on self-reinforcement and quasi-parameters.

These literatures establish that endogenous institutional persistence, path dependence, political-power dynamics, and long-run distributions are already central objects of political economy.

PES does not propose a new universal mechanism of institutional persistence. It adds a classificatory question. Once a dynamic political model supplies equilibrium-induced long-run behavior, PES asks:

|⋃σ∈𝖤⋃s0∈𝒟ι[ℒσ(s0)]|=?

The focus is therefore whether institutional multiplicity survives equilibrium and initial-condition variation.

2.4 Stochastic stability

Stochastic-stability theory addresses a different logical problem. Kandori, Mailath, and Rob (1993) and Young (1993) study adaptive or evolutionary dynamics with small mutations or mistakes and characterize which conventions or equilibria retain long-run probability as perturbations vanish.

PES places the test earlier.

If the unperturbed equilibrium institutional image is

Ψ∞={CA,CB},

while perturbations select CA with asymptotic probability one, then CA is stochastically selected but the unperturbed game is not PES.

Thus:

stochastic selection asks which available long-run alternative dominates;

whereas

PES asks whether long-run institutional alternatives survive the admitted equilibrium structure at all.

The concepts are complementary rather than hierarchical.

2.5 Positioning

The appropriate novelty claim is narrow.

This paper does not claim that multiple equilibria with a common limit, all-equilibrium implementation, robust concentration of equilibrium predictions, endogenous institutional persistence, or stochastic equilibrium selection are new.

Its proposed contribution is the systematic use of

Ψ∞𝖤,ι(Γ;𝒟)

as a general institutional-image object, together with a classification architecture that is explicitly equilibrium-concept relative, domain relative, institutional-resolution relative, and distinct from perturbation selection and normative evaluation.

3. Dynamic Political Games and Institutional Images

3.1 Dynamic environment

Let

Γ=⟨N,(S,τ,𝒮),{Ai}i∈N,{ui}i∈N,P,{δi}i∈N⟩(3)

denote a dynamic political game.

Here N is the set of strategic actors; S is the state space; τ is a pre-specified topology on S; and 𝒮 is a σ-algebra containing the Borel σ-algebra generated by τ. Finite-state models use the discrete topology.

For each player i, Ai(s) is the feasible action set, ui is the relevant payoff function, P(⋅∣s,a) is the transition kernel, and δi, where required, is an intertemporal discount factor.

The representation is canonical rather than restrictive. Particular applications may use deterministic transitions, continuous state spaces, or richer information structures.

State-Sufficiency Requirement

The state representation must contain all history-dependent information relevant to feasible actions, payoffs, transitions, equilibrium behavior, and institutional classification. If omitted history affects any of these objects, the state must be augmented.

3.2 Equilibrium correspondence and outcome laws

Let

𝖤(Γ)(4)

be the complete set of equilibria admitted by a pre-specified solution concept.

For

σ∈𝖤(Γ),s0∈S,

let

ℙs0σ(5)

denote the induced law on realized state-action histories.

The equilibrium concept is part of the PES claim. It may not be altered retrospectively because a broader solution set generates an inconvenient institutional class.

3.3 Admissible initial domain

Let

𝒟⊆S(6)

be the domain of initial states over which the claim is evaluated.

The domain must be specified independently of the realized outcome. The term Global PES is reserved for cases in which 𝒟 contains every initial state deemed admissible by the model.

3.4 Institutional representation

Let

h:S→ℑ

map dynamic states to full institutional configurations.

Let

≅I

be a pre-specified institutional equivalence relation and define

ℭ=ℑ/≅I.(7)

Let

qI:ℑ→ℭ

be the quotient map and define

ι=qI∘h:S→ℭ.(8)

The measurable structure on ℭ is assumed sufficient for the probability statements used below, and ι is measurable.

Institutional-Classification Requirement

The maps and equivalence relation defining ι must be specified before the equilibrium image is evaluated. A rival institutional structure specified by the hypothesis may not be merged retrospectively with the proposed target.

3.5 Admissible reachability and contestability

Let 𝒜adm denote the set of admissible strategy profiles, not restricted to equilibrium strategies.

Define support-based admissible reachability by

ReachΓadm(𝒟)=⋃α∈𝒜adm⋃s0∈𝒟⋃t<∞suppτℙs0α(St∈⋅).(9)

In finite-state models this reduces to the set of states reachable with positive probability.

A substantive PES claim requires institutional contestability:

|ι[ReachΓadm(𝒟)]|≥2.(10)

This is a non-triviality guard, not part of the singleton definition itself.

3.6 Essential long-run support

For each

(σ,s0)∈𝖤(Γ)×𝒟,

let

ℒΓ,σ(s0)⊆S(11)

denote the essential long-run support of the induced process.

In a finite-state Markov setting it can be taken as the union of recurrent classes reachable with positive probability. In deterministic systems an appropriate ω-limit construction may be used. In more general environments the model must specify an equivalent support-based object.

The long-run operator excludes purely transient states and probability-zero path artifacts. It also satisfies Eventual-Class Consistency:

ℙs0σ(∃T∀t≥T:ι(St)=C)=1⇒ι[ℒΓ,σ(s0)]={C}.(12)

Condition (12) is automatic in the finite recurrent-class formulation and can be guaranteed in topological formulations through appropriate regularity of the long-run operator and institutional fibers.

3.7 Definition 1 — Equilibrium Institutional Image

Define

Ψ∞𝖤,ι(Γ;𝒟)=⋃σ∈𝖤(Γ)⋃s0∈𝒟ι[ℒΓ,σ(s0)].(13)

Thus

Ψ∞⊆ℭ.

The image contains institutional classes, not equilibria, paths, or raw dynamic states.

The use of a union is deliberate. Every equilibrium admitted by the selected solution concept and every initial state admitted by the domain has veto power over a singleton claim.

4. Political Equilibrium Singularity

4.1 Definition 2 — PES

For C⋆∈ℭ,

PES𝖤,ι(Γ,𝒟;C⋆)⇔Ψ∞𝖤,ι(Γ;𝒟)={C⋆}.(14)

Under the standing non-emptiness assumptions this is equivalent to

ι[ℒΓ,σ(s0)]={C⋆}

for every σ∈𝖤(Γ) and s0∈𝒟.

Exact PES is support-based. Any distinct positive-probability essential long-run class falsifies it, regardless of how small that probability is.

4.2 PES Diameter

If (ℭ,dℭ) is a metric space, define

DPES=diamdℭΨ∞.(15)

Lemma 1 — Diameter Characterization

If Ψ∞≠⌀ and dℭ is a genuine metric, then

DPES=0⇔|Ψ∞|=1.(16)

The diameter measures spread in the institutional image under the chosen metric. It is not automatically a measure of instability, welfare loss, or volatility.

4.3 Proposition 1 — Domain Monotonicity

If

𝒟1⊆𝒟2,

then

Ψ∞(Γ;𝒟1)⊆Ψ∞(Γ;𝒟2).(17)

Hence PES on a larger nonempty domain implies PES with the same target on every nonempty subdomain. The converse fails.

A Global PES claim is therefore stronger than a basin-specific claim.

4.4 Proposition 2 — Equilibrium-Set Monotonicity

For two solution sets satisfying

𝖤1(Γ)⊆𝖤2(Γ),

one has

Ψ∞𝖤1⊆Ψ∞𝖤2.(18)

Thus PES under a broader equilibrium set implies PES under a nonempty subset, but not conversely.

An equilibrium concept that admits more equilibria can preserve or destroy PES; it cannot restore a singleton by enlarging the set.

5. Non-Degenerate and Robust PES

5.1 Definition 3 — Outcome-Law Equivalence

For

σ,τ∈𝖤(Γ),

define

σ≡O𝒟τ

if

ℙsσ=ℙsτ∀s∈𝒟.(19)

When the domain is fixed, write σ≡Oτ.

This quotient eliminates equilibrium multiplicity that exists only through labels or off-path prescriptions.

5.2 Definition 4 — Non-Degenerate PES

ND-PES(C⋆)⇔PES(C⋆)∧|𝖤(Γ)/≡O|>1.(20)

ND-PES therefore means

distinct equilibrium outcome laws+one long-run institutional class.

Equilibrium selection may still matter for policy, timing, distribution, coalition formation, and transient dynamics. It simply does not change the long-run institutional class.

5.3 Parameter robustness

Let

𝚪={Γθ}θ∈Θ

be a parameterized family with common institutional comparison space ℭ.

Define

ℛ(C⋆)={θ:Ψ∞,θ={C⋆}},(21)

and

𝒰(C⋆)=IntΘℛ(C⋆).(22)

Definition 5 — ND-RPES

Γθ0 exhibits ND-RPES at C⋆ if there exists an open neighborhood U∋θ0 such that

ND-PESθ(C⋆)

for every θ∈U.

Thus both the singleton institutional image and nontrivial outcome-law multiplicity survive locally in parameter space.

5.4 Stochastic robustness

Consider perturbation families for which every relevant perturbed equilibrium

σε∈𝖤(Γθ,ε)

induces a time-homogeneous Markov process on the specified state representation, possibly after state augmentation, and admits at least one invariant probability measure.

Let

Pθ,εσε

denote the equilibrium-induced transition kernel.

Define the equilibrium-wide invariant-measure set

ℳθ,ε=⋃σε∈𝖤(Γθ,ε)Inv(Pθ,εσε).(23)

Assume

ℳθ,ε≠⌀.

Definition 6 — SR-ND-RPES

An ND-RPES at C⋆ on U is stochastically robust if

∀θ∈U:limε↓0infμ∈ℳθ,εμ(ιθ−1[{C⋆}])=1.(24)

The condition is pointwise in θ. Uniform stochastic robustness over the entire neighborhood would be stronger and is not required here.

The stochastic-robustness refinement is therefore restricted to model classes in which invariant-measure analysis of the equilibrium-induced perturbed process is well defined. This restriction does not narrow the underlying definition of PES.

The canonical hierarchy is

PES→ND-PES→ND-RPES→SR-ND-RPES.(25)

Stochastic concentration does not repair a failed PES condition. If the unperturbed image contains two institutional classes, selection of one under vanishing noise remains stochastic selection rather than PES.

6. Projected Institutional Singularity

Whole PES may be stronger than the institutional question of interest. A political order can remain heterogeneous in policy, personnel, coalition structure, or subordinate institutional form while preserving the same constitutive component.

Let

πK:ℭ→𝔎(26)

be a pre-specified projection into a kernel-class space.

Define

Ψ∞,K=πK[Ψ∞].(27)

6.1 Definition 7 — Projected PES

For k⋆∈𝔎,

PESK(k⋆)⇔Ψ∞,K={k⋆}.(28)

Let

ℱK(k⋆)=πK−1[{k⋆}]

be the target fiber.

Lemma 2 — Fiber Characterization

PESK(k⋆)⇔Ψ∞⊆ℱK(k⋆).(29)

Corollary 1 — Whole to Projected

If

PES(C⋆),

then

PESK(πK(C⋆))(30)

for every valid projection.

The converse fails.

6.2 Projection discipline

Projected PES is meaningful only if the projection is not selected to erase observed multiplicity.

Three restrictions therefore apply.

First, πK must be specified before the equilibrium image is evaluated.

Second, projected alternatives must be reachable:

|πK[ι[ReachΓadm(𝒟)]]|≥2.(31)

Third, the projection may not collapse a rival distinction that is itself part of the hypothesis being tested.

Projected PES therefore means neither total institutional uniformity nor a retrospectively chosen common denominator. Its substantive content is:

complete institutional plurality can coexist with singularity of a pre-specified frame.

Or, more compactly:

singularity of a frame does not imply singularity of everything inside the frame.

7. General Results and Obstructions

7.1 Theorem 1 — Uniform Institutional Contraction

Let

V:S→ℝ≥0

be measurable and suppose there exists c⋆>0 such that

V(s)≥c⋆𝟏{ι(s)≠C⋆}.(32)

Suppose also that there exist k∈ℕ, 0<ρ<1, and M<∞ such that for every admitted equilibrium, admissible initial state, and time t,

𝔼σ[V(St+k)∣ℱt]≤ρV(St).(33)

and, for 1≤r<k,

𝔼σ[V(St+r)∣ℱt]≤MV(St).(34)

Assume V(s0)<∞ for every s0∈𝒟.

Then

PES(C⋆).(35)

Indeed,

V(St)→0

almost surely under every admitted equilibrium and initial condition, and target separation implies eventual residence in C⋆. Eventual-Class Consistency then converts this pathwise conclusion into the long-run-support conclusion required by PES.

The theorem is sufficient only. PES need not admit a common scalar contraction representation.

7.2 Theorem 2 — Symmetry Obstruction

Let

g∈Aut(Γ,𝒟,ι)

include a bijection

gS:S→S

and corresponding transformations of players, actions, and strategies.

A valid PES automorphism preserves the complete PES specification.

Domain preservation

gS(𝒟)=𝒟.(36)

Game preservation

Feasible actions, payoffs, and transition probabilities are preserved under the appropriate relabeling.

Equilibrium preservation

σ∈𝖤(Γ)⇔gσ∈𝖤(Γ).(37)

Long-run equivariance

ℒΓ,gσ(gSs)=gS[ℒΓ,σ(s)].(38)

Institutional equivariance

There exists an induced bijection

gℭ:ℭ→ℭ

such that

ι(gSs)=gℭ(ι(s)).(39)

When no ambiguity exists, write gC for gℭ(C).

If

PES(C⋆),

then

gC⋆=C⋆∀g∈Aut(Γ,𝒟,ι).(40)

Hence

C⋆∈Fix[Aut(Γ,𝒟,ι)].(41)

A target exchanged with a distinct rival by a genuine automorphism cannot be uniquely selected as PES.

The appropriate substantive corollary is therefore:

Semantic-specific PES requires the relevant exchange symmetry to be broken somewhere in the complete PES specification.

The symmetry-breaking element may be a payoff, transition law, feasibility condition, admissible domain, institutional classification, or another component of the formal specification.

7.3 Proposition 3 — Regime Separation

Define

𝒩PES={θ∈Θ:θ∉𝒰(C)∀C∈ℭ}.(42)

Let

γ:[0,1]→Θ

be continuous with

γ(0)∈𝒰(CA),γ(1)∈𝒰(CB),CA≠CB.

Then there exists t⋆∈(0,1) such that

γ(t⋆)∈𝒩PES.(43)

A continuous transition between distinct robust targets therefore cannot remain entirely inside robust singleton-image regions.

The proposition does not imply that CA and CB coexist in the institutional image at the boundary.

7.4 Logical non-implications

The following properties are distinct at the level of the general framework: