--- title: Egalitarian-Equivalence in Fair Division url: https://www.emergentmind.com/topics/egalitarian-equivalence type: topic --- # Egalitarian-Equivalence in Fair Division Searching arXiv for recent papers on egalitarian-equivalence and closely related cooperative-game/fair-division work. Egalitarian-equivalence is a fairness concept that, in one standard formalization, requires the existence of a common reference bundle such that every agent is indifferent between her assigned bundle and that reference bundle. In the homogeneous indivisible-object model with money, this requirement is written as: for each valuation profile \(v\), there exists \(z_0\in \{0,1\}\times \mathbb{R}\) such that \(u(f_i(v);v_i)=u(z_0;v_i)\) for every agent \(i\) [2507.09152]. In adjacent literatures, the same normative orientation appears through equal-division rules, egalitarian Shapley families, equal-surplus extensions, and equality-sensitive welfare criteria. The concept therefore has both a narrow technical meaning in fair division and a broader role as a benchmark for comparing egalitarian and marginalist principles [2507.09152, 1608.01540, 2605.20113]. ## 1. Reference-bundle formulation In the homogeneous indivisible-object allocation problem with money, each agent receives a bundle \(z_i=(x_i,t_i)\in \{0,1\}\times \mathbb{R}\), where \(x_i=1\) means receiving an object, \(x_i=0\) means receiving no object, and \(t_i\) is the transfer. Utilities are quasi-linear: \[ u((x_i,t_i);v_i)=v_i x_i-t_i. \] A mechanism \(f\) is egalitarian-equivalent if, for each valuation profile \(v\in V\), there exists a reference bundle \(z_0\in \{0,1\}\times \mathbb{R}\) such that \[ u(f_i(v);v_i)=u(z_0;v_i)\quad \text{for each } i\in N. \] Thus every agent must be indifferent between her realized bundle and a single common benchmark [2507.09152]. This formulation becomes especially restrictive under individual rationality and no subsidy. A key lemma states that if the reference bundle has no object, then it must be exactly \(\mathbf{0}=(0,0)\). Formally, if \(x_0=0\), then \(z_0=\mathbf{0}\) [2507.09152]. The significance of this restriction is that egalitarian-equivalence is not merely a loose equal-treatment condition: in this environment it imposes a strong common-indifference structure on feasible allocations. Several other papers in the supplied corpus use “egalitarian” language without employing this exact reference-bundle definition. That distinction is substantive. Some works study equal surplus, equal influence, equal minted shares, or the worst-off agent’s welfare; these are related fairness criteria, but not the same theorem-level notion as egalitarian-equivalence in the fair-division sense [1406.7642, 1507.06827, 2402.16145]. ## 2. Egalitarian-equivalent division under additive utilities For divisible items with additive utilities, the paper on dividing goods or bads under additive utilities studies the Egalitarian Equivalent rule directly. The manna consists of one unit of each item, each agent \(i\) has additive utility vector \(u_i\), and a feasible allocation \(z=(z_i)_{i\in N}\) satisfies \[ \sum_{i\in N} z_i = e^A. \] Utilities are \(U_i=u_i\cdot z_i\), with each \(u_i\) defined only up to positive rescaling [1608.01540]. The rule is normalized by imposing \[ u_i\cdot e^A=1 \qquad \forall i. \] For goods, the Egalitarian rule \(F^{eg}\) selects the utility profile that is maximal under leximin ordering: \[ U^{eg}\in \arg\max_{\Psi(\mathcal Q)} \succeq_{lx}. \] The paper identifies this as the classical egalitarian equivalent/leximin rule in the additive domain [1608.01540]. For bads, the rule selects the unique efficient normalized utility profile \(U^{eg}\) such that \[ U_i^{eg}=U_j^{eg}\qquad \forall i,j. \] In this case, egalitarian-equivalence takes the form of equal disutility at the efficient frontier [1608.01540]. The comparison with Competitive Equilibrium with Equal Incomes is central. For goods, an allocation is competitive if there exists \(p\in \mathbb R_+^A\) with \(\sum_{a\in A}p_a=n\) such that \[ z_i \in \arg\max_{y_i\in \mathbb R_+^A}\{u_i\cdot y_i \mid p\cdot y_i\le 1\}\qquad \forall i. \] For bads, the mirrored condition is \[ z_i \in \arg\min_{y_i\in \mathbb R_+^A}\{u_i\cdot y_i \mid p\cdot y_i\ge 1\}\qquad \forall i. \] Both rules satisfy the Fair Share Guarantee, but only \(F^{eg}\) satisfies the Strict Fair Share Guarantee [1608.01540]. The comparison is sharply asymmetric across goods and bads. For goods, the competitive rule is Resource Monotonic, whereas for three or more agents the egalitarian rule is not. The paper also introduces Independent of Lost Bids and shows that changing a lost bid does not affect the competitive outcome, while the Egalitarian rule admits simple profitable local manipulation [1608.01540]. For bads, the ordering reverses: the Egalitarian rule is single-valued, continuous in utilities, and computationally simple, while the competitive rule can be multivalued, can have as many as \(2^{\min\{n,m\}-1}\) utility profiles if \(n\neq m\) and \(2^{n-1}-1\) if \(n=m\), and admits no continuous selection [1608.01540]. This contrast is one of the clearest demonstrations that egalitarian-equivalent division is not uniformly dominated by market-based notions. ## 3. TU-games: equal division, Shapley, and null player neutrality In transferable-utility cooperative games, egalitarian-equivalence appears through the relation between the Shapley value and equal division. For a game \(v\in V^N\), the Shapley solution is \[ \varphi^{Sh}_i(v) = \sum_{S \subseteq N : i \in S} \frac{(s-1)!(n-s)!}{n!}\left( v(S) - v(S \backslash i) \right), \] while the equal division solution is \[ \varphi^{ED}_i(v) = \frac{v(N)}{n}. \] The standard \(\alpha\)-egalitarian Shapley family is \[ \varphi^\alpha(v) = \alpha \varphi^{ED}(v) + (1-\alpha)\varphi^{Sh}(v), \qquad \alpha\in[0,1]. \] This family interpolates between Shapley at \(\alpha=0\) and equal division at \(\alpha=1\) [2605.20113]. The central new axiom is null player neutrality. For each \(v,w,u\in V^N\) and each \(i\in N\), if \(w(N)=u(N)\) and \(i\) is null in both \(w\) and \(u\), then \[ \varphi_i(v+w) = \varphi_i(v+u). \] This weakens coalitional strategic equivalence. Coalitional strategic equivalence requires that if \(i\) is null in \(w\), then \(\varphi_i(v+w)=\varphi_i(v)\); null player neutrality only requires invariance across null-player augmentations with the same grand-coalition worth [2605.20113]. The characterization theorem is exact. A solution \(\varphi\) satisfies efficiency, linearity, symmetry, and null player neutrality if and only if there exists \(\alpha\in\mathbb{R}\) such that \[ \varphi \equiv \alpha \varphi^{ED} + (1-\alpha)\varphi^{Sh}. \] Equivalently, \[ \varphi_i(v)=\alpha\frac{v(N)}{n} + (1-\alpha)\varphi_i^{Sh}(v). \] The noteworthy point is that the characterization extends the classical family beyond convex combinations: negative \(\alpha\) and \(\alpha>1\) are allowed [2605.20113]. The dual nullifying-player analogue is even sharper. If null player neutrality is replaced by nullifying player neutrality, then efficiency, symmetry, and nullifying player neutrality characterize the equal division solution uniquely: \[ \varphi \equiv \varphi^{ED}. \] The paper also stresses that this result is stronger than an earlier theorem of Brink (2007), because it drops linearity and replaces the nullifying player property with nullifying player neutrality [2605.20113]. ## 4. Geometric generalization: the egalitarian Shapley axis A complementary development places the egalitarian Shapley family inside a geometric structure on the space \(L\) of linear value maps. The relevant subspace \(L^{ESL}\) consists of efficient, symmetric, linear value maps. The paper proves a canonical linear isomorphism \[ \Phi_{\mathrm{strat}}: L^{ESL} \longrightarrow \mathbb{R}^{n-1}, \qquad \Psi \longmapsto \big(\varepsilon_1(\Psi),\ldots,\varepsilon_{n-1}(\Psi)\big), \] under which every \(\Psi\in L^{ESL}\) decomposes uniquely by coalition size: \[ \Psi^{(a)} = (1-\varepsilon_a(\Psi))\,Sh^{(a)} + \varepsilon_a(\Psi)\,ED^{(a)} \qquad (1 \le a \le n-1). \] The classical egalitarian Shapley family is exactly the diagonal: \[ \varepsilon_1=\cdots=\varepsilon_{n-1}=\varepsilon. \] The paper therefore identifies the line \[ \{ \Psi^\varepsilon \mid \varepsilon \in \mathbb{R}\} \] as the egalitarian Shapley axis, or marginalism–egalitarianism axis [2605.22847]. This geometry yields a precise generalization of egalitarian-equivalence. If all coalition-size strata share the same parameter \(\varepsilon\), the value lies on the classical axis. If the parameters differ by coalition size, the result is a stratified egalitarian Shapley value. The stratum-wise identity \[ (\Psi - Sh)^{(a)} = \varepsilon_a(\Psi)\,(ED - Sh)^{(a)} \] formalizes the way in which each coalition size can carry its own egalitarian coefficient [2605.22847]. The projection of any \(\Psi\in L^{ESL}\) onto the diagonal has parameter \[ \varepsilon^*(\Psi) = \frac{\langle ED - Sh,\; \Psi - Sh\rangle_L} {\langle ED - Sh,\; ED - Sh\rangle_L} = \sum_{a=1}^{n-1} w_a\,\varepsilon_a(\Psi), \] with weights \[ w_a := \frac{\|(ED - Sh)^{(a)}\|_L^2}{D_n} = \frac{\binom{n}{a}\left(\frac{1}{a}-\frac{1}{n}\right)}{D_n}, \qquad D_n := \sum_{a=1}^n \binom{n}{a}\left(\frac{1}{a}-\frac{1}{n}\right). \] The goodness-of-fit statistic is \[ R^2(\Psi) = \frac{E_w[\varepsilon_a(\Psi)]^2}{E_w[\varepsilon_a(\Psi)^2]} = 1 - \frac{\operatorname{Var}_w(\varepsilon_a(\Psi))}{E_w[\varepsilon_a(\Psi)^2]}. \] This is presented as a literal regression-statistics analogue of the coefficient of determination [2605.22847]. At \(n=4\), the paper classifies several standard values relative to this axis. The Banzhaf value has \(R^2=2/203\approx 0.99\%\), the equal-surplus-division value has \(R^2=11/29\approx 37.9\%\), and the solidarity value has \(R^2=4563/4582\approx 99.6\%\). Asymptotically, \(R^2(ESD)\to 1\), \(R^2(So)\to 1\), and \(R^2(Bz)\to 1/2\) [2605.22847]. This suggests a precise sense in which some classical alternatives are almost entirely aligned with the egalitarian Shapley axis, while others remain structurally different. ## 5. Strategy-proofness, non-obvious manipulability, and mechanism design The indivisible-object model with money provides a direct test of how egalitarian-equivalence interacts with incentive constraints. The environment has \(n\) agents, \(m\) homogeneous indivisible objects with \(n>m\), at most one object per agent, and quasi-linear preferences. The first characterization theorem states that a mechanism satisfies egalitarian-equivalence, strategy-proofness, individual rationality, and no subsidy if and only if it is an uncompromising \(\tau\)-Vickrey mechanism combined with the no-trade [2507.09152]. The Vickrey component assigns objects to agents with valuations above the \((m+1)\)-th highest valuation and charges winners \(v^{m+1}\): \[ x_i(v)= \begin{cases} 1 & \text{if } v_i>v^{m+1},\\ 0 & \text{if } v_i