When limited resources change what makes a network stable

A collaboration can be worth maintaining without being worth expanding. And even when everyone wants more collaboration, someone may have no resources left to contribute. These distinctions are central to a theoretical study by Julien Fixary, Rémi Castera and Rida Laraki, posted as an arXiv preprint on November 4, 2026. Its results are mathematical, not empirical evidence about how actual networks form, and the supplied source does not establish peer-reviewed status.

The paper extends weighted network models, in which relationships have adjustable intensities rather than simply existing or not existing. It allows interactions among groups of any size and introduces two restrictions: each group has a maximum interaction intensity, and each participant has a budget shared across all their groups.

Stability, one interaction at a time

The authors call their basic solution concept coalitionwise stability. Despite the name, it does not test every coordinated rearrangement that a group could make. It examines changes to one group’s interaction intensity while holding all other intensities fixed.

A feasible network is stable under this definition when two conditions hold. First, no participant can strictly improve their payoff by reducing an interaction involving them. Second, no feasible increase in a group’s intensity strictly benefits every member. An indifferent member is therefore enough to prevent an increase from counting as a destabilizing deviation.

The authors prove that such a network exists when payoffs are continuous and componentwise quasiconcave. The latter condition concerns changing one interaction while fixing the others: an intermediate intensity cannot give a payoff below both endpoint payoffs. It does not require payoffs to be jointly concave across the entire network.

Existence means at least one allocation satisfies these conditions. It does not establish uniqueness, social efficiency or a process by which participants will reach that allocation.

A stylized research example illustrates the distinction. Three researchers can undertake two bilateral projects and one three-person project, with researcher 1 involved in all three. Everyone benefits from increasing their own projects. Without binding budgets, stability requires every project to reach its capacity. With unit budgets and unit capacities, any nonnegative allocation whose three project intensities sum to one is stable. Researcher 1’s exhausted budget blocks further expansion, leaving a continuum of stable allocations.

Why small disturbances become difficult

The paper then asks whether stability depends on fragile circumstances, such as exact indifference. Drawing inspiration from trembling-hand perfection in game theory, the authors develop stronger stability requirements.

Their perturbative approach slightly restricts the feasible environment. Every interaction must remain above zero and below its capacity, while every participant must leave a positive amount of budget unused. A perturbed equilibrium is a limit of networks stable in these restricted environments as the restrictions vanish.

The authors prove existence under continuity and componentwise quasiconcavity, with the additional standing assumption that the original feasible set has a nonempty interior. Under continuity, these limits are also coalitionwise stable.

The restriction can matter. In one example, a collaboration is sustainable only because a participant is indifferent while another project remains inactive. Requiring even a tiny positive intensity in that other project makes the participant prefer to reduce the collaboration. The perturbative refinement eliminates those positive-intensity outcomes.

A carefully qualified robustness concept

A different approach perturbs surrounding interactions and asks whether a fixed reference intensity remains stable. Resource constraints complicate this test: adding a small amount to one project can make another project’s reference intensity exceed someone’s budget.

The authors show that two natural extensions of trembling-hand stability can fail to exist even with linear payoffs. Keeping the reference intensity feasible is too demanding in one example. Dropping that requirement still fails in another, because perturbations change the available opportunities for expansion.

Their final definition instead holds the allowable deviation range at the range associated with the limiting network, while evaluating payoffs against nearby interior feasible networks. This is a specific mathematical robustness test—not a requirement that every evaluated combination remain feasible after a disturbance.

For continuous payoffs that are strictly monotonic in each relevant interaction, the authors prove equivalence between this trembling-hand definition and perturbed equilibrium. They also establish trembling-hand existence under a weaker fixed-sign monotonicity assumption: payoffs may have flat regions, but each participant’s direction of preference for an interaction must remain unchanged as other interactions vary. This weaker result does not establish the same equivalence.

The applications cover stylized collaboration, communication and public-good sharing models. They illustrate where the assumptions apply, rather than validating predictions with observations. The central contribution is therefore conditional: resource constraints need not destroy existence of stable networks, but stronger claims about robustness depend on both the payoff assumptions and precisely how disturbances are defined.