Nash Bargaining Solution

The solution must be Pareto optimal. This means no other feasible agreement exists that would make one player better off without making the other player worse off. The outcome lies on the Pareto frontier of the bargaining set, ensuring no value is 'left on the table' through inefficiency.

• Example: If agents can agree on a deal worth (5, 5) or (8, 2), the Nash Solution will not select (5, 5) if (8, 2) is also feasible, as moving to (8, 2) improves Player 1's outcome without harming Player 2 (who stays at 5). It selects the point on the frontier that maximizes the product of utilities.

If the bargaining problem is symmetric—meaning the set of feasible outcomes and the players' disagreement points are identical—then the solution must award equal payoffs to each player. The axiom enforces anonymous fairness; the solution cannot favor one player over the other based on arbitrary labels like 'Player A' vs. 'Player B'.

• Implication: This axiom handles cases where players are identical in bargaining power and context. If symmetry is broken (e.g., one agent has a better outside option), the solution adjusts accordingly, as captured by the Invariance to Affine Transformations axiom.

The solution is invariant if a player's utility function is scaled (multiplied by a positive constant) or translated (added a constant). This means the bargaining solution depends only on the ordinal preferences and the relative scale of utilities, not on arbitrary units of measurement.

• Technical Role: It allows utilities to be normalized. Typically, the disagreement point (d1, d2) is set to (0, 0), and the utilities are rescaled so that the feasible set is standardized. This is the mathematical step that enables the solution to be computed as the point maximizing (u1 - d1) * (u2 - d2).

If the solution for a bargaining set S is a point x, and we consider a smaller subset T of S that still contains x, then x must remain the solution for T. Removing 'irrelevant' alternative options that were not chosen should not alter the negotiated outcome.

• Critique & Context: This is the most debated axiom. It implies path independence of the negotiation logic. In multi-agent systems, it simplifies reasoning but can be restrictive if the shape of the feasible set conveys information about bargaining power. Alternatives like the Kalai-Smorodinsky solution replace IIA with a monotonicity axiom.

Nash proved that the unique solution satisfying all four axioms is the point (u1*, u2*) in the feasible set that maximizes the Nash Product: (u1 - d1) * (u2 - d2), where (d1, d2) is the disagreement point (the outcome if negotiation fails).

• Calculation: The solution is found by solving this constrained optimization problem. It geometrically corresponds to the point where the Pareto frontier is tangent to the highest possible rectangular hyperbola.
• Interpretation: Maximizing the product of gains represents a balance between efficiency (high total sum) and equity (avoiding highly unequal distributions).

The disagreement point (d1, d2), also called the threat point, is a critical parameter. It represents the utility each agent receives if negotiations break down. The Nash Solution explicitly incorporates these outside options, determining the baseline from which gains are calculated.

• Strategic Influence: Agents can engage in strategic maneuvering to improve their disagreement point before negotiation, thereby improving their final payoff. In multi-agent system design, this maps to an agent's Best Alternative to a Negotiated Agreement (BATNA).
• Normalization: By applying the Invariance axiom, we often set d1 = d2 = 0 to simplify the Nash Product to u1 * u2.

A state of resource allocation where no agent can be made better off without making at least one other agent worse off. It defines an efficiency frontier of possible agreements.

• Key Insight: The Nash Bargaining Solution is always Pareto optimal; it finds a point on this frontier.
• Example: In a bandwidth-sharing negotiation between two agents, a Pareto-optimal agreement uses all available bandwidth. Any reallocation that helps one agent would necessarily harm the other.

A foundational alternating-offers game-theoretic model for dividing a surplus between two agents. It incorporates time discounting, where delay in reaching an agreement reduces the value of the pie.

• Contrast with Nash: While Nash is axiomatic and cooperative, Rubinstein models non-cooperative, sequential strategic interaction.
• Subgame Perfect Equilibrium: The model yields a unique solution where the first agent's share is (1 - δ₂) / (1 - δ₁δ₂) of the surplus, with δ representing each agent's discount factor.

The inverse of game theory: designing the rules of a game (or negotiation protocol) so that the self-interested, strategic behavior of agents leads to a socially desirable outcome.

• Goal: Engineer protocols for properties like efficiency, revenue maximization, or truth-telling.
• Core Tool: The Revelation Principle, which states that for any mechanism, there exists an equivalent direct revelation mechanism where truth-telling is an equilibrium.

A mathematical representation of an agent's preferences, assigning a numerical value to each possible outcome. Agents are modeled as seeking to maximize their expected utility.

• Role in Nash Bargaining: The solution maximizes the product of the agents' utility gains (u₁ - d₁)(u₂ - d₂), where d is the disagreement point.
• Cardinal vs. Ordinal: Nash Bargaining requires cardinal, interpersonally comparable utility, not just ordinal rankings.

The private walk-away point in a negotiation: the minimum price a seller will accept or the maximum price a buyer will pay. It defines the agent's alternative if no agreement is reached.

• Link to Disagreement Point: In the Nash framework, the reservation price helps determine the disagreement payoff d_i. Any agreement must provide a utility higher than this baseline.
• Strategic Concealment: Agents often hide their true reservation price to gain a bargaining advantage.

A simple bilateral bargaining procedure where agents alternately make concessions from their previous offers. Concessions must be monotonic (cannot be retracted) until agreement or deadline.

• Structure: 1. Agents state initial positions. 2. They alternately propose new offers that are more favorable to the counterparty. 3. Agreement occurs when an offer is accepted.
• Practical Use: Provides a structured, predictable framework for automated agents to converge, unlike open-ended dialogue.