[Paper Review] An FPTAS for Bargaining Networks with Unequal Bargaining Powers

Yashodhan Kanoria|arXiv (Cornell University)|Aug 1, 2010

Game Theory and Voting Systems19 references3 citations

TL;DR

This paper presents the first Fully Polynomial-Time Approximation Scheme (FPTAS) for computing unequal division (UD) solutions in bargaining networks with asymmetric bargaining powers. By combining belief propagation for stable matching and a novel iterative edge rebalancing process, the method ensures convergence to an ϵ-approximate UD solution in poly(|V|, 1/ϵ) time, resolving a long-standing open problem in computational tractability for asymmetric bargaining models.

ABSTRACT

Bargaining networks model social or economic situations in which agents seek to form the most lucrative partnership with another agent from among several alternatives. There has been a flurry of recent research studying Nash bargaining solutions (also called 'balanced outcomes') in bargaining networks, so that we now know when such solutions exist, and also that they can be computed efficiently, even by market agents behaving in a natural manner. In this work we study a generalization of Nash bargaining, that models the possibility of unequal 'bargaining powers'. This generalization was introduced in [KB+10], where it was shown that the corresponding 'unequal division' (UD) solutions exist if and only if Nash bargaining solutions exist, and also that a certain local dynamics converges to UD solutions when they exist. However, the bound on convergence time obtained for that dynamics was exponential in network size for the unequal division case. This bound is tight, in the sense that there exists instances on which the dynamics of [KB+10] converges only after exponential time. Other approaches, such as the one of Kleinberg and Tardos, do not generalize to the unsymmetrical case. Thus, the question of computational tractability of UD solutions has remained open. In this paper, we provide an FPTAS for the computation of UD solutions, when such solutions exist. On a graph G=(V,E) with weights (i.e. pairwise profit opportunities) uniformly bounded above by 1, our FPTAS finds an \eps-UD solution in time poly(|V|,1/\eps). We also provide a fast local algorithm for finding \eps-UD solution, providing further justification that a market can find such a solution.

Motivation & Objective

• To resolve the open problem of computational tractability for unequal division (UD) solutions in bargaining networks with asymmetric bargaining powers.
• To design a fast, local algorithm that converges to approximate UD solutions in polynomial time, overcoming exponential convergence bounds of prior dynamics.
• To establish that stability of allocations must be preserved during rebalancing, as instability leads to exponential convergence times.
• To generalize the concept of balanced outcomes to asymmetric settings by formalizing and computing UD solutions.
• To provide theoretical justification for market mechanisms that can naturally converge to UD outcomes in asymmetric networks.

Proposed method

• The method begins with a maximum weight matching and stable allocation using belief propagation, ensuring initial stability.
• It introduces a two-phase algorithm: first, compute a stable allocation; second, apply iterative edge rebalancing to achieve ϵ-UD solutions.
• The edge rebalancing process uses a non-expansive operator Tκ = κT + (1−κ)I, where T performs asymmetric surplus splitting based on bargaining powers.
• Convergence is guaranteed via a termination condition based on the sup-norm difference between consecutive iterates, ensuring ‖Tγᵗ − γᵗ‖∞ ≤ ϵ.
• The algorithm maintains stability throughout by operating only within the space of stable allocations, preventing exponential delays due to instability.
• Theoretical bounds are derived using concentration inequalities and non-expansivity, showing that convergence time is polynomial in |V| and 1/ϵ.

Experimental results

Research questions

• RQ1 Can unequal division (UD) solutions in bargaining networks with asymmetric bargaining powers be computed efficiently, despite prior algorithms failing to generalize from symmetric cases?
• RQ2 Why do existing local dynamics for symmetric bargaining networks fail to converge in polynomial time in the asymmetric case, and can this be remedied?
• RQ3 Is there a way to maintain stability during iterative rebalancing to avoid exponential convergence times?
• RQ4 Can a local, decentralized algorithm converge to an ϵ-approximate UD solution in time polynomial in the network size and 1/ϵ?
• RQ5 Does preserving stability during rebalancing fundamentally enable efficient computation of UD solutions?

Key findings

• The paper presents the first FPTAS for computing unequal division (UD) solutions in bargaining networks with asymmetric bargaining powers, achieving an ϵ-approximate solution in time poly(|V|, 1/ϵ).
• The proposed local algorithm converges to an ϵ-UD solution in polynomial time, with convergence time bounded by O(1/(κ(1−κ)ϵ²)) iterations, where κ is a parameter controlling the rebalancing step size.
• The construction demonstrates that instability in allocations—such as the presence of a 'bad' edge with γi + γj ≤ wij − 1—can lead to exponential convergence times, even when the initial state satisfies ϵ-correct division.
• By preserving stability throughout the rebalancing process, the algorithm avoids exponential delays, proving that stability is a critical component of efficient computation.
• The algorithm’s convergence to an exact UD solution is guaranteed in the limit (as t → ∞), recovering the existence of UD solutions as a corollary.
• The method provides a fast, local mechanism that justifies how markets with asymmetric bargaining power could naturally converge to UD outcomes, even in the absence of global coordination.