Invariance under Disapproving Unassigned Goods (IDU)

• IDU is an invariance axiom in fair division with binary valuations that keeps allocation outcomes unchanged when an agent disapproves a good they do not receive.
• It works alongside EF1, strategyproofness, neutrality, and minimal completeness to uniquely characterize rules like maximum Nash welfare and leximin.
• The axiom ensures Pareto-optimality and consistent tie-breaking, preventing manipulative reporting and promoting equitable, balanced allocations.

Searching arXiv for the cited paper and closely related work. First, I’ll look up the main paper by arXiv id. Searching for ([2607.10064](/papers/2607.10064)). Invariance under Disapproving Unassigned Goods (IDU) is an axiom for allocation rules in fair division with indivisible goods and binary valuations. In the setting studied in "Fair Division with Binary Valuations: Characterizations" (Brandl et al., 11 Jul 2026), IDU requires that if an agent lowers her approval of a good that she does not receive under the rule’s outcome, then the allocation must remain unchanged. Within this binary domain, IDU functions as a distinctive instance-consistency condition: together with envy-freeness up to one good (EF1), strategyproofness, neutrality, and minimal completeness, it characterizes the common rule given by maximum Nash welfare (MNW), leximin, and additive strictly-concave welfarism for any fixed number of agents (Brandl et al., 11 Jul 2026).

1. Formal setting and binary-allocation model

The domain consists of a set of agents $N = [n]$ and a set of indivisible goods $G = \{g_1,\ldots,g_m\}$. Each agent $i \in N$ has a binary additive valuation over bundles $S \subseteq G$. At the singleton level, each good is either approved or disapproved, so $v_i(\{g\}) \in \{0,1\}$ for all $g \in G$, and utility is additive:

$u_i(S) = v_i(S) = \sum_{g\in S} v_i(\{g\}).$

Utilities are nonnegative and normalized so that every approved good contributes utility $1$ and every disapproved good contributes $0$ (Brandl et al., 11 Jul 2026).

The approval profile is written as $P = (P_1,\ldots,P_n)$, where $G = \{g_1,\ldots,g_m\}$0 if and only if $G = \{g_1,\ldots,g_m\}$1. An instance is therefore $G = \{g_1,\ldots,g_m\}$2. An allocation $G = \{g_1,\ldots,g_m\}$3 partitions some subset of $G = \{g_1,\ldots,g_m\}$4 into disjoint bundles $G = \{g_1,\ldots,g_m\}$5; not all goods need be allocated. The notation $G = \{g_1,\ldots,g_m\}$6 denotes the set of all allocations on $G = \{g_1,\ldots,g_m\}$7.

Several axioms structure the analysis. An allocation is EF1 if for all $G = \{g_1,\ldots,g_m\}$8 with $G = \{g_1,\ldots,g_m\}$9, there exists $i \in N$0 such that

$i \in N$1

A rule is strategyproof if no agent can misreport and thereby strictly improve her utility. A rule is neutral if permuting labels of goods does not change the utility each agent receives. An allocation is minimally complete if every valued good is allocated and every unvalued good is unallocated; equivalently, every good $i \in N$2 with $i \in N$3 is unallocated, and every good with $i \in N$4 is allocated (Brandl et al., 11 Jul 2026).

The paper also uses non-redundancy and resource-monotonicity in a two-agent characterization. Non-redundancy requires that every allocated good be valued by its recipient. Resource-monotonicity requires that when one extra good is added, no agent’s utility decreases.

2. Definition, meaning, and examples of IDU

IDU is introduced as a condition on how a rule reacts to downward changes in approvals that concern goods not assigned to the deviating agent. Formally, for two instances $i \in N$5 and $i \in N$6 that are identical except that for some agent $i \in N$7 and good $i \in N$8 one has $i \in N$9 and $S \subseteq G$0, a rule $S \subseteq G$1 satisfies IDU if the following holds: if $S \subseteq G$2 and $S \subseteq G$3, then

$S \subseteq G$4

Equivalently, agent $S \subseteq G$5 may lower her approval from $S \subseteq G$6 to $S \subseteq G$7 on any good $S \subseteq G$8 she did not receive, and the allocation must remain invariant (Brandl et al., 11 Jul 2026).

The informal meaning given in the paper is that if an agent stops approving a good that is not allocated to her under the rule’s outcome, then this change is irrelevant to the rule’s choice and the allocation should not change. IDU is described as akin to independence of irrelevant alternatives, as having a monotonicity flavor, and as being related to non-bossiness. The monotonicity flavor comes from the fact that no agent can obtain a good by ceasing to value it. The relation to non-bossiness is that an agent should not be able to change others’ allocations without changing her own.

A simple example illustrates the axiom. With two agents and three goods, let agent 1 approve $S \subseteq G$9 and agent 2 approve $v_i(\{g\}) \in \{0,1\}$0. Suppose a rule allocates $v_i(\{g\}) \in \{0,1\}$1 to agent 2, $v_i(\{g\}) \in \{0,1\}$2 to agent 1, and leaves $v_i(\{g\}) \in \{0,1\}$3 unallocated. If agent 1 changes her report to disapprove $v_i(\{g\}) \in \{0,1\}$4, which she did not receive, IDU requires the same allocation to be returned. Likewise, if agent 2 disapproves $v_i(\{g\}) \in \{0,1\}$5, which is unassigned to her, the outcome must remain unchanged (Brandl et al., 11 Jul 2026).

The conceptual role of IDU is especially clear in the binary domain. It blocks manipulations in which an agent disapproves an unassigned good in order to alter the rule’s evaluation of the instance and trigger reallocations elsewhere. The paper explicitly notes that IDU is not implied by EF1, neutrality, or strategyproofness; it is independent and captures a different instance-consistency dimension.

3. Welfare objectives and the common rule in the binary domain

The paper studies several welfare objectives that coincide under binary additive valuations. The standard MNW objective is to maximize the product of utilities,

$v_i(\{g\}) \in \{0,1\}$6

or equivalently $v_i(\{g\}) \in \{0,1\}$7 when all utilities are positive. Because zero utilities may occur under indivisible goods, the paper adopts a tie-aware binary-domain definition. For an allocation $v_i(\{g\}) \in \{0,1\}$8, let

$v_i(\{g\}) \in \{0,1\}$9

Define $g \in G$0. Among allocations maximizing the number of agents with positive utility, MNW selects any allocation $g \in G$1 that maximizes

$g \in G$2

The leximin rule, by contrast, lexicographically maximizes the sorted utility profile $g \in G$3, where the utilities are in nondecreasing order (Brandl et al., 11 Jul 2026).

The additive strictly-concave welfarist rules maximize

$g \in G$4

where $g \in G$5 is increasing and strictly concave. The paper notes that under binary valuations, MNW, leximin, and all such strictly-concave additive welfarist rules coincide; it also notes that this class includes $g \in G$6-mean rules for $g \in G$7 as special cases. Benabbou et al. (2021) are credited with establishing this equivalence (Brandl et al., 11 Jul 2026).

The explanation given for this coincidence is discrete egalitarianism. Because utilities are counts of approved goods received, strictly concave objectives penalize inequality and reward balanced distributions. Leximin does so directly by maximizing the minimum utility and then the next minimum. MNW favors the same balancing tendency through the product objective. The paper further connects these objectives to a combinatorial exchange structure via critical paths. A plausible implication is that the binary domain is unusually rigid: once efficiency and fairness constraints are combined with invariance properties such as IDU, apparently different welfare criteria collapse to a single rule.

4. IDU, Pareto-optimality, and the any-number-of-agents characterization

A central lemma states that any rule satisfying both IDU and minimal completeness also satisfies Pareto-optimality (PO) (Brandl et al., 11 Jul 2026). The proof idea is specific to the binary domain. If some valued good $g \in G$8 were allocated to an agent who does not value it, then one can make all other agents disapprove $g \in G$9. By IDU, the good remains with the same recipient, but by minimal completeness the good would then have to be unallocated because it is unvalued. This contradiction implies that every valued good must be allocated to some agent who values it, which yields PO in the binary setting.

This interaction is structurally important because the main combinatorial characterization of MNW is stated for Pareto-optimal allocations. The paper defines a path as a sequence $u_i(S) = v_i(S) = \sum_{g\in S} v_i(\{g\}).$0 with $u_i(S) = v_i(S) = \sum_{g\in S} v_i(\{g\}).$1 for each $u_i(S) = v_i(S) = \sum_{g\in S} v_i(\{g\}).$2, and a path is critical if

$u_i(S) = v_i(S) = \sum_{g\in S} v_i(\{g\}).$3

Lemma 3.1 states that for any instance $u_i(S) = v_i(S) = \sum_{g\in S} v_i(\{g\}).$4 and Pareto-optimal allocation $u_i(S) = v_i(S) = \sum_{g\in S} v_i(\{g\}).$5, the allocation is MNW if and only if $u_i(S) = v_i(S) = \sum_{g\in S} v_i(\{g\}).$6 contains no critical path. The paper presents this as a characterization of the discrete “Nash-improving” exchanges that MNW forbids (Brandl et al., 11 Jul 2026).

The main theorem then gives the role of IDU in full generality. For any fixed $u_i(S) = v_i(S) = \sum_{g\in S} v_i(\{g\}).$7 and binary valuations, any allocation rule $u_i(S) = v_i(S) = \sum_{g\in S} v_i(\{g\}).$8 that satisfies EF1, strategyproofness, neutrality, minimal completeness, and IDU must maximize Nash welfare. The proof sketch proceeds by contradiction. If a rule satisfying these axioms fails to return MNW on some instance, then by Lemma 3.1 its allocation admits a shortest critical path. The proof then constructs modified instances by locally changing approvals along bundles in that path. IDU is used repeatedly to keep the outcome invariant when agents disapprove goods they do not receive, while neutrality controls the effects of relabeling goods. Strategyproofness and EF1 then force contradictions on the values along the path (Brandl et al., 11 Jul 2026).

The paper therefore identifies IDU as the decisive instance-consistency axiom in the many-agent binary domain. EF1 provides the local fairness inequalities, strategyproofness excludes profitable manipulations, neutrality removes label dependence, and minimal completeness ensures that all and only valued goods matter. IDU binds these ingredients together by stabilizing the outcome under carefully chosen changes in approval reports.

5. Two-agent replacement of IDU and tie-breaking refinements

For $u_i(S) = v_i(S) = \sum_{g\in S} v_i(\{g\}).$9, the paper gives an alternative characterization in which IDU is replaced by non-redundancy and resource-monotonicity. The theorem states that under binary valuations, any rule $1$0 that is EF1, non-redundant, strategyproof, resource-monotone, neutral, and minimally complete maximizes Nash welfare (Brandl et al., 11 Jul 2026).

The proof idea uses induction on the number of goods together with a compact description of two-agent binary instances by the characteristic tuple

$1$1

Neutrality implies that instances with the same $1$2 must yield the same utilities. Resource-monotonicity controls what happens when goods are added. Non-redundancy and minimal completeness ensure utilitarian efficiency in the binary domain, while strategyproofness excludes manipulations that would upset the leximin/MNW balance. The proof derives contradictions whenever a rule deviates from MNW.

The paper also develops tie-breaking refinements. One result states that if a rule is minimal complete, neutral, IDU, and MNW, then for any fixed $1$3 the rule must favor the same agent across tie-break-relevant instances. A second result states that if a rule is EF1, minimal complete, resource-monotone, strategyproof, neutral, and MNW, then for any fixed $1$4—the number of valued goods—tie-breaking must consistently favor the same agent across tie-break-relevant instances (Brandl et al., 11 Jul 2026).

These refinements show that the characterization is not only about which welfare objective is selected but also about how indifferences are resolved. In this domain, consistent tie-breaking is not an auxiliary design choice; it is necessary to preserve the package of neutrality, IDU, strategyproofness, and monotonicity properties.

6. Necessity of the axioms, algorithmic aspects, and related invariance principles

The paper proves that all axioms in both characterizations are necessary. For the any-number-of-agents result, Proposition 3.4 states that each of EF1, strategyproofness, neutrality, minimal completeness, and IDU is necessary: dropping any one permits a rule that satisfies the remaining four but does not maximize Nash welfare. For the two-agent result, Proposition 4.5 states that each of EF1, non-redundancy, strategyproofness, resource-monotonicity, neutrality, and minimal completeness is necessary: omitting any single axiom allows a rule that either fails to be MNW or violates the tie-breaking consistency required by Theorem 4.4 (Brandl et al., 11 Jul 2026).

The paper also reports that MNW, and equivalently leximin and strictly-concave welfarist rules, can be computed in polynomial time under binary valuations, attributing this to Darmann and Schauer (2015) and Barman et al. (2018). For practical implementation with tie-breaking, it describes an MNW procedure with consistent tie-breaking, denoted $1$5, consisting of three steps: discard all unvalued goods; allocate any good approved by exactly one agent to that agent; and on the remaining contested goods, balance agents’ counts to maximize the leximin/Nash objective using a consistent tie-breaker, such as lexicographic over utilities and then goods. The paper states that this yields allocations satisfying EF1, PO, non-redundancy, minimal completeness, strategyproofness, neutrality, and IDU in the binary domain (Brandl et al., 11 Jul 2026).

Two examples summarize the force of IDU. In Example 1.1, with $1$6 and $1$7, agent 1 approves $1$8 and agent 2 approves $1$9. MNW, leximin, and strictly-concave welfarism yield a balanced allocation with approvals respected. If agent 1 disapproves $0$0, which she does not receive, IDU requires the allocation to remain unchanged. A rule violating IDU could instead reallocate a contested good such as $0$1 from agent 2 to agent 1 even though the report changed only on an unassigned good; the characterization rules out precisely this kind of behavior (Brandl et al., 11 Jul 2026).

The paper situates IDU near familiar invariance ideas. Its spirit parallels independence of irrelevant alternatives in the sense that outcomes should not depend on valuations of options that an agent cannot obtain, and it resembles non-bossiness in that changes that do not affect one’s own allocation should not affect others. It also notes that many natural rules, including welfarist rules and picking sequences with consistent tie-breaking, satisfy IDU. This suggests that IDU is not an ad hoc restriction but a natural axiom for binary approvals. In the resulting characterization, it is the property that converts local fairness, strategic robustness, and completeness into a full identification of the common rule:

$0$2

For arbitrary numbers of agents, IDU is the key axiom that delivers this conclusion; for two agents, its role can be replaced by non-redundancy and resource-monotonicity without changing the identified rule (Brandl et al., 11 Jul 2026).