🔢 mathematics

Genuine and Spurious Equilibria in the Graph Model for Conflict Resolution under Effective State Exclusion

This paper proposes a structural approach to modeling effectively excluded states in the Graph Model for Conflict Resolution by restricting the state space, thereby classifying equilibrium predictions as genuine, spurious, F-conditional, or excluded and characterizing how different stability concepts respond to such exclusions.

Original authors: Yukiko Kato

Published 2026-09-17

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Yukiko Kato

Original paper licensed under CC BY 4.0 (https://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

Conflict is rarely a simple game of two sides moving toward a single goal. In the real world, the board itself can change. A treaty might ban a specific military maneuver; a new law might make a certain business practice impossible; a physical barrier might render a location unreachable. When these hard constraints appear, the old map of the conflict becomes obsolete. For decades, experts who study strategic disputes have used a method called the Graph Model for Conflict Resolution to predict how such fights end. This method treats a conflict like a map of possible situations, or "states," connected by lines that show who can move where. By analyzing the preferences of the people involved—what they like and what they fear—the model identifies "equilibria," which are the stable outcomes where no one has a reason to change their position. It is a powerful tool for understanding everything from international diplomacy to environmental disputes.

However, a critical question has long lingered in the background of this work: what happens to these predictions when a piece of the map is suddenly removed? If a state is not just undesirable but effectively impossible to reach, does the old prediction still hold? For a long time, analysts tried to handle this by simply telling the model that the forbidden state was the worst possible outcome for everyone. But this approach creates a logical trap. In the logic of conflict, making a state "terrible" actually makes it a stronger threat. If a rival knows you will be devastated by a certain outcome, they might use that fear to force you into a corner. By ranking a state as the worst, you inadvertently strengthen its power to stabilize a bad situation, rather than removing it from the game entirely. The state remains a tool for leverage, even if it is supposed to be off-limits.

A researcher at the Institute of Science Tokyo, Yukiko Kato, has now developed a way to fix this. Instead of changing how much people dislike a state, Kato's work proposes a structural change: simply deleting the forbidden state and all the paths that lead to it from the model. This creates a new, restricted version of the conflict map. By comparing the predictions of the original map with the restricted one, the study reveals that not all stable outcomes are created equal. Some predictions are "genuine," meaning they remain stable whether the forbidden state exists or not. Others are "spurious," meaning they only looked stable because the model was counting on a threat that is now impossible. There are also "conditional" outcomes, which become stable only because the forbidden state is gone, and "excluded" outcomes, which were stable before but are now part of the forbidden zone.

The study proves that these categories are not just mathematical curiosities; they represent real strategic shifts. For instance, the research shows that some types of stability are fragile. If a stable outcome relied on a chain of reactions that passed through the forbidden state, that stability vanishes the moment the state is removed. In other cases, removing a state can actually create new stability. This happens when a state was previously unstable because a rival could escape a punishment by moving into the forbidden zone. If that escape route is sealed off, the punishment becomes effective, and the situation stabilizes. The paper demonstrates that these changes depend heavily on the specific rules of how the conflict is analyzed, with some rules being immune to these shifts and others being highly sensitive.

To test these ideas, the author applied the framework to a famous real-world case: the groundwater contamination dispute in Elmira, Ontario, in the late 1980s. In this conflict, a chemical company, the government, and local officials were locked in a struggle over how to clean up a toxic site. The company had the option to simply abandon the site and walk away, a move that served as a powerful threat in the original analysis. The study imagined a scenario where this abandonment option was legally impossible, perhaps due to strict financial guarantees or liability laws. When the model was run with this option removed, the results changed dramatically. One of the previously predicted stable outcomes, where the company stayed but the order was modified, turned out to be "spurious"—it only existed because the threat of walking away was in play. Another outcome, where the company accepted the original order without modification, became stable only because the escape route was blocked. This "conditional" stability meant that the agreement would hold only as long as the legal barrier remained in force.

The research also clarifies a common confusion between two different ways of judging a threat. One way is to ask if a threat is "credible"—would the person making the threat actually carry it out if it were in their interest? The other is to ask if a threat is "admissible"—is the path to that threat even open? The study shows these are independent. A threat can be perfectly credible (the person wants to do it) but structurally inadmissible (the path is blocked). Conversely, a threat can be admissible but not credible. This distinction matters because a model that only checks for credibility might miss the fact that a path has been physically or legally severed. Similarly, the study distinguishes between a state being "safe" (meaning the conflict cannot wander into the forbidden zone) and a prediction being "sound" (meaning the logic proving it is stable doesn't rely on the forbidden zone). A state can be safe from wandering into the forbidden zone, yet its stability proof can still be broken if it relied on a path that went through that zone.

Ultimately, this work provides a diagnostic tool for policymakers and analysts. It allows them to see exactly which predictions will survive a new rule, which will collapse, and which new agreements might become possible. It moves beyond simply recalculating the answer to a new map; it explains the mechanism of the change. The study confirms that removing a state is not just a matter of deleting a line on a graph; it fundamentally alters the strategic landscape by cutting off paths of leverage and escape. The research concludes that this structural exclusion can be implemented in the real world by severing the transitions that lead to the forbidden state, such as passing a law that removes a specific option. The framework does not solve the problem of finding the cheapest way to make these changes, but it provides the necessary vocabulary to understand what those changes will actually do to the outcome of a conflict. By separating the structural reality from the preferences of the players, the study offers a clearer, more honest view of how conflicts resolve when the rules of the game change.

Technical Summary: Genuine and Spurious Equilibria in the Graph Model for Conflict Resolution under Effective State Exclusion

Problem Statement
In the Graph Model for Conflict Resolution (GMCR), equilibrium predictions are conditional not only on decision makers' (DMs) preferences and behavioral assumptions but also on the modeled state space and transition structure. In many real-world conflicts, certain states are not merely undesirable but effectively unavailable due to regulatory regimes, treaty commitments, physical constraints, or policy counterfactuals. Standard GMCR modeling handles undesirability by ranking such states at the bottom of preference lists. However, the paper argues that this approach fails to capture effective exclusion. Under sanction-based stability analysis (e.g., General Metarationality or Symmetric Metarationality), ranking a state low strengthens its role as a sanction rather than removing it from the analysis. Consequently, there is a need for a framework that distinguishes between states that are structurally unavailable versus those that are simply dispreferred, and to determine how the exclusion of a set of states ( $F$ ) alters equilibrium predictions without conflating structural admissibility with preference-based credibility.

Methodology
The paper proposes a structural diagnostic framework based on the comparison of two graph models:

The Original Model ( $G$ ): The standard graph model containing the full set of states $S$ and transitions.
The Restricted Model ( $G \upharpoonright F$ ): A model obtained by removing the designated set of excluded states $F \subseteq S$ and all arcs incident to them. This restriction removes not only sanctions ending in $F$ but also any sanctioning sequences that traverse $F$ .

The core methodology involves a Comparative Equilibrium Diagnosis. By comparing the equilibrium sets $Eq_\sigma(G)$ and $Eq_\sigma(G \upharpoonright F)$ for a given stability concept $\sigma$ , every equilibrium prediction is classified into one of four categories:

Genuine: An equilibrium in $G$ that remains an equilibrium in $G \upharpoonright F$ .
Spurious: An equilibrium in $G$ (outside $F$ ) that ceases to be an equilibrium in $G \upharpoonright F$ .
F-conditional: A state that is not an equilibrium in $G$ but becomes one in $G \upharpoonright F$ .
Excluded: An equilibrium in $G$ that lies within $F$ .

The analysis relies on a Locality Principle (Lemma 3.4), which establishes that stability verification for a state depends only on the forward-reachable submodel. This principle allows the author to prove that if no arcs cross from admissible states into $F$ , the exclusion has no effect on stability verification.

Key Contributions
The paper makes four primary contributions:

General Diagnostic Framework: It defines a set-theoretic classification (Genuine, Spurious, F-conditional, Excluded) applicable to any stability concept, providing a semantics for how predictions change under contingent state exclusion.
Stability-Specific Theory: It characterizes the mechanisms driving these changes for four standard stability concepts:
- Nash Stability: Nash equilibria are never spurious; exclusion can only create new Nash equilibria (by removing improving moves) but never destroy existing ones.
- General Metarationality (GMR) and Sequential Stability (SEQ): These concepts can produce spurious equilibria. This occurs when the exclusion removes necessary sanctioning paths, including paths that merely traverse the excluded set $F$ (a phenomenon relevant for conflicts with three or more DMs).
- Symmetric Metarationality (SMR): SMR exhibits a sharp asymmetry. The "sanction side" (loss of sanctions) can produce spuriousness, while the "escape side" (loss of escape routes into $F$ ) can produce F-conditional stability.
Conceptual Separation: The paper formally distinguishes three properties often conflated in conflict analysis:
- Preferential Credibility: Whether a sanction is a preferred move for the sanctioning DM.
- Trajectory Safety: Whether rational or arbitrary trajectories can reach $F$ from a given state.
- Verification Soundness: Whether the stability certificate survives the structural exclusion of $F$ .
  The author proves that credibility and admissibility are independent (a sanction can be credible but structurally excluded), and that trajectory safety (even under rational play) does not guarantee verification soundness.
Structural Implementability: The paper demonstrates that the exclusion counterfactual is structurally implementable. By performing a "boundary intervention"—deleting all arcs from admissible states into $F$ —one can reproduce the predictions of the restricted model on the admissible state set without altering any preferences.

Results

Mechanisms of Change: The paper proves that Nash equilibria are immune to spuriousness. GMR and SEQ equilibria become spurious if and only if all sanctioning paths for a deviation are lost (either ending in $F$ or traversing it). SMR equilibria can become spurious via sanction loss or become F-conditional via the removal of escape routes that would otherwise allow a DM to evade a sanction.
Independence of Axes: Through the "Counter-escalation game" and the "Elmira groundwater conflict," the paper demonstrates that a sanction can be fully credible (an improvement for the sanctioner) yet structurally excluded, rendering a Sequentially Stable equilibrium spurious. Conversely, a state can be safe from reaching $F$ under improving trajectories (in the incentive-based safety region $\nu_{imp}$ ) yet be spurious because the sanctions supporting its stability run through $F$ .
Elmira Case Study: Applied to the Elmira groundwater conflict, the framework reveals that:
- State 4 (a classical equilibrium) is spurious under GMR/SMR because its stability relies entirely on the threat of abandonment (State 9), which is excluded.
- State 7 (jointly preferred by public actors) is F-conditional; it becomes stable only when the abandonment option is structurally removed.
- States 5 and 8 are genuine, remaining stable regardless of the exclusion.
- The analysis distinguishes between states that look identical under credibility refinements (e.g., St1 vs. St4) but differ fundamentally in their structural dependence on the excluded set.

Significance and Scope
The paper positions this work as a structural diagnostic tool rather than a design algorithm. Its significance lies in clarifying that equilibrium predictions are sensitive to the admissibility of states, not just their desirability. The framework allows analysts to determine whether a predicted resolution is robust to institutional changes (genuine), fragile and likely to be invalidated (spurious), or entirely dependent on a specific regime (F-conditional).

The author explicitly limits the scope of the paper:

The framework diagnoses the consequences of an exogenous exclusion set; it does not generate or justify the set itself.
It assumes a binary admissibility (states are either available or excluded) and does not model graded or probabilistic exclusion.
It establishes implementability (that the exclusion can be structurally enforced) but does not solve the optimization problem of finding the minimum-cost intervention to achieve a specific outcome.
The analysis holds preferences fixed to isolate the structural channel; it does not model how real-world regimes might alter valuations.

The paper concludes that while the boundary intervention is sufficient to implement the exclusion, the problems of minimal intervention and optimal structural design remain open combinatorial challenges for future research. The primary contribution is the vocabulary and theoretical foundation required to distinguish between preserved, invalidated, and induced predictions before attempting to engineer such regimes.

More like this