Helpful context:

Nash’s (1950) axiomatic bargaining solution is elegant: it characterizes a unique outcome by four axioms - Pareto optimality, symmetry, scale invariance, and independence of irrelevant alternatives. But the IIA axiom has always been the contentious one. Kalai and Smorodinsky (1975) replaced it with a monotonicity condition and recovered a different, equally principled solution. The two solutions coincide on some problems and diverge significantly on others - and understanding why reveals something deep about what it means for a bargaining outcome to be “fair.”

The disagreement turns on a question of responsiveness. IIA says that expanding the feasible set in ways that do not affect the current solution should not change that solution - a consistency requirement. Individual monotonicity says that expanding what is achievable for one player should not make that player worse off - a responsiveness requirement. Neither implies the other, and both are defensible. What is not defensible is treating the choice between them as purely technical: it reflects a substantive commitment about whether a solution should track the global shape of the feasible set or only its local geometry near the solution point.

Both solutions agree on symmetric problems and diverge when the feasible set is asymmetric. The divergence is not a flaw in either solution - it is precisely where the axioms do different work, and studying the divergence is the clearest way to understand what each axiom buys.

The Bargaining Problem

A bargaining problem is a pair $(S, d)$ where $S \subseteq \mathbb{R}^2$ is the feasible utility set - compact, convex, and comprehensive (if $x \in S$ and $y \leq x$ componentwise then $y \in S$) - and $d \in S$ is the disagreement point, the payoff each player receives if no agreement is reached. The set $S$ encodes all utility pairs that can be achieved by some agreement; $d$ encodes the outside options.

A bargaining solution is a function $F$ mapping each bargaining problem $(S, d)$ to a point $F(S, d) \in S$. Without loss of generality, normalize $d = (0, 0)$; feasibility requires $S$ to contain points strictly dominating $d$ in both coordinates (otherwise bargaining is trivial for one party).

The Pareto frontier of $S$ (relative to $d$) is the set of points in $S$ that are not strictly dominated by any other point in $S$. Any sensible solution must select from the Pareto frontier - otherwise there is an outcome that both players prefer, and no rational mechanism would stop there.

Nash Bargaining Solution

Nash (1950) imposed four axioms on a bargaining solution $F$:

  1. Pareto optimality: $F(S,d)$ lies on the Pareto frontier of $S$.
  2. Symmetry: if $S$ is symmetric about the line $x_1 = x_2$ and $d_1 = d_2$, then $F_1(S,d) = F_2(S,d)$.
  3. Scale invariance: if $S'$ is obtained from $S$ by independent positive affine rescalings of the two utility axes, $F$ transforms accordingly (the solution is not affected by the units used to measure utility).
  4. Independence of irrelevant alternatives (IIA): if $T \subseteq S$ and $F(S,d) \in T$, then $F(T,d) = F(S,d)$.

Theorem (Nash 1950). The unique bargaining solution satisfying these four axioms is: $$F^N(S, d) = \arg\max_{x \in S, x \geq d} (x_1 - d_1)(x_2 - d_2)$$

The Nash solution maximizes the product of gains from disagreement. Geometrically: among all rectangular hyperbolae $(x_1 - d_1)(x_2 - d_2) = c$ with $c > 0$, find the largest $c$ such that the hyperbola still intersects $S$. The solution is the intersection point. Equivalently, it is the point on the Pareto frontier that maximizes the area of the rectangle from $d$ to $x$.

The Problem with IIA

IIA demands that the solution be insensitive to the parts of $S$ that are not selected. If we remove some feasible options - as long as we keep the proposed solution - the solution should not change. This is internally consistent but has a troubling implication: the solution depends only on the local shape of the Pareto frontier near the solution point, not on the global structure of $S$.

Kalai and Smorodinsky’s objection is concrete. Consider two bargaining problems $(S, d)$ and $(S', d)$ with $S \subseteq S'$ - so $S'$ gives player 1 strictly more options, including options with higher utility for player 1 along the Pareto frontier. Intuitively, player 1 is in a stronger position in $(S', d)$: there are more favorable agreements available. Yet the Nash solution can give player 1 a smaller share in $(S', d)$ than in $(S, d)$. This violates a natural monotonicity intuition.

More precisely: fix player 2’s maximum achievable utility at the same level in both problems. If player 1’s maximum achievable utility increases, player 1’s solution payoff should not decrease. The Nash solution does not guarantee this.

Individual Monotonicity

Define the utopia point $a(S, d) = (a_1, a_2)$ where: $$a_i = \max\{x_i \mid x \in S, x \geq d\}$$

This is the most player $i$ could ever receive from any feasible agreement - player $i$’s maximum aspiration. The utopia point is generally not feasible: it is typically impossible for both players to simultaneously achieve their individual maxima.

Individual monotonicity: if $T \subseteq S$ and $a_i(T, d) = a_i(S, d)$ (player $i$’s maximum is the same in both problems), then $F_i(S, d) \geq F_i(T, d)$. An expansion of the feasible set that does not reduce player $i$’s maximum should not reduce player $i$’s solution payoff.

This axiom replaces IIA in Kalai and Smorodinsky’s characterization. Together with Pareto optimality, symmetry, and scale invariance, it uniquely determines the Kalai-Smorodinsky solution.

The Kalai-Smorodinsky Solution

Theorem (Kalai-Smorodinsky 1975). The unique bargaining solution satisfying Pareto optimality, symmetry, scale invariance, and individual monotonicity is the Kalai-Smorodinsky solution $F^{KS}$, defined as the Pareto optimal point on the line segment from $d$ to $a(S, d)$:

$$F^{KS}(S, d) = \max\left\{t \in [0,1] : d + t\bigl(a(S,d) - d\bigr) \in S\right\} \cdot \bigl(a(S,d) - d\bigr) + d$$

Equivalently, $F^{KS}(S, d)$ is the unique Pareto optimal point $x \in S$ satisfying: $$\frac{x_1 - d_1}{a_1(S,d) - d_1} = \frac{x_2 - d_2}{a_2(S,d) - d_2}$$

The solution gives each player the same proportion of their maximum possible gain. If player 1 can gain at most 4 and player 2 can gain at most 2 (from the disagreement point), the KS solution splits these maxima proportionally: both players receive the same fraction, say 60%, of what they could individually achieve - so player 1 gets $0.6 \times 4 = 2.4$ and player 2 gets $0.6 \times 2 = 1.2$.

Geometric construction: draw the line from the disagreement point $d$ to the utopia point $a$; the KS solution is where this line crosses the Pareto frontier of $S$.

Example. Let $S = \{(x_1, x_2) : x_1 + 4x_2 \leq 4, x_1, x_2 \geq 0\}$ and $d = (0,0)$.

  • Utopia point: $a_1 = 4$ (set $x_2 = 0$), $a_2 = 1$ (set $x_1 = 0$), so $a = (4, 1)$.
  • KS line: $x_2 = \frac{x_1}{4}$. Intersect with Pareto frontier $x_1 + 4x_2 = 4$: $x_1 + x_1 = 4$, giving $x_1 = 2$, $x_2 = \frac{1}{2}$. So $F^{KS} = (2, \frac{1}{2})$.
  • Nash solution: maximize $x_1 x_2$ on $x_1 + 4x_2 = 4$, i.e., $x_1 x_2 = (4 - 4x_2)x_2 = 4x_2 - 4x_2^2$. First-order condition: $4 - 8x_2 = 0$, so $x_2 = \frac{1}{2}$, $x_1 = 2$. Here $F^N = F^{KS} = (2, \frac{1}{2})$.

For linear Pareto frontiers, the two solutions agree. Divergence occurs on curved frontiers. Consider the feasible set whose Pareto frontier is the arc of the ellipse $\frac{x_1^2}{4} + x_2^2 = 1$ in the first quadrant. Here $a = (2, 1)$ and $d = (0, 0)$. The KS line is $x_2 = x_1/2$; substituting into the ellipse: $\frac{x_1^2}{4} + \frac{x_1^2}{4} = 1$, so $x_1 = \sqrt{2}$, $x_2 = \frac{\sqrt{2}}{2}$. For Nash: maximize $x_1 x_2$ subject to $\frac{x_1^2}{4} + x_2^2 = 1$. Using Lagrange multipliers, $x_2 = \lambda \frac{x_1}{2}$ and $x_1 = 2\lambda x_2$, giving $x_2 = \lambda \frac{x_1}{2}$ and $\lambda = \frac{x_1}{2x_2}$, so $x_2^2 = \frac{x_1^2}{4}$, i.e., $x_2 = \frac{x_1}{2}$. Here they agree again because the ellipse is symmetric in rescaled coordinates. To see genuine divergence, one needs a problem with asymmetric curvature of the frontier; on such problems the Nash product’s maximizer will not generally lie on the $d$-to-$a$ line.

Nash vs Kalai-Smorodinsky: The Axiomatic Tradeoff

The two solutions embody different views of fairness.

The Nash solution maximizes the product of gains. It is sensitive to the local curvature of the Pareto frontier at the solution point - it finds where the marginal rate of substitution between the two players' utilities equals their ratio of gains from disagreement. It ignores the global shape of $S$.

The KS solution equalizes proportional gains relative to each player’s maximum. It tracks the global shape of $S$ through the utopia point. If player 1’s achievable maximum increases, $F^{KS}$ responds by giving player 1 more; the Nash solution may not.

Neither solution satisfies the other’s distinctive axiom:

  • Nash satisfies IIA but violates individual monotonicity.
  • KS satisfies individual monotonicity but violates IIA.

There is no solution satisfying all five axioms (Pareto optimality, symmetry, scale invariance, IIA, and individual monotonicity) on all bargaining problems - the axiom systems are genuinely incompatible. The choice between them is a substantive normative choice, not a technical one.

In practice: when agents have very different maximal aspirations, KS is often considered more natural, since it directly accounts for what each player could individually achieve. When the concern is coherence under shrinkage of the feasible set, Nash’s IIA is the cleaner criterion.

Summary

Property Nash Kalai-Smorodinsky
Pareto optimality Yes Yes
Symmetry Yes Yes
Scale invariance Yes Yes
IIA Yes No
Individual monotonicity No Yes
Formula $\arg\max_{x \geq d} \prod_i (x_i - d_i)$ Pareto point on line from $d$ to $a(S,d)$
Sensitivity to global shape of $S$ No (local only) Yes (via utopia point)
Proportional gains Not guaranteed $\frac{x_i - d_i}{a_i - d_i}$ equal for all $i$

Read next: