PROP1+M1S allocation doesn't exist
Dependencies:
- Fair division
- EFX
- EF1
- PROP1
- Minimum fair share
- AnyPrice share
- Maximin share allocations
- Restricted agents, pairwise fairness, and groupwise fairness
- Additive goods and binary marginals
$\newcommand{\WMMS}{\mathrm{WMMS}}$ $\newcommand{\APS}{\mathrm{APS}}$ $\newcommand{\Ical}{\mathcal{I}}$ Consider a fair division instance $\Ical = ([n], [m], V, w)$, where $n=3$, $m=7$, and $w = (7/12, 5/24, 5/24)$. The agents have identical additive valuations, and all items are goods of value 1 each.
Let $A$ be an allocation where $|A_1| = 3$ and $|A_2| = |A_3| = 2$. Then all of the following hold:
- An allocation $X$ is EFX iff $X = A$.
- An allocation $X$ is M1S iff $X = A$.
- $\WMMS_1 = 3$ and $\WMMS_2 = \WMMS_3 = 15/14$.
- An allocation $X$ is groupwise-WMMS iff $X = A$.
- $\APS_1 = 4$ and $\APS_2 = \APS_3 = 1$.
- $X$ is PROP1 iff $X$ is APS.
- $X$ is groupwise APS iff $X$ is APS.
Consequences of the above observations:
- M1S and PROP1 are not simultaneously achievable for this instance.
- Groupwise WMMS and EFX don't imply PROP1.
- APS doesn't imply M1S.
Proof
Observations 5 and 6 follow from binary-additive.html.
When every item has value 1, an agent is EF1-satisfied by an allocation iff she is EFX-satisfied by the allocation. It is easy to see that allocation $A$ is EFX (and hence also M1S).
Suppose for some allocation $X$, we have $|X_2| ≤ 1$ and agent 2 is EF1-satisfied by $X$. If $|X_2| = 0$, then $|X_1| ≤ 1$ and $|X_3| ≤ 1$, which is a contradiction. If $|X_2| = 1$, then $|X_3| ≤ 2$, and so $|X_1| ≥ 4$. Then \[ \frac{|X_1|}{w_1} ≥ \frac{48}{7} > \frac{24}{5} = \frac{|X_2|}{w_2}. \] This is a contradiction. Hence, if agent 2 is EF1-satisfied by $X$, then $|X_2| ≥ 2$. So, if agent 2 is M1S-satisfied by $X$, then $|X_2| ≥ 2$. Similarly, if agent 3 is M1S-satisfied by $X$, then $|X_3| ≥ 2$.
Suppose for some allocation $X$, we have $|X_1| ≤ 2$ and agent 1 is EF1-satisfied by $X$. Then for some $j \in \{2, 3\}$, we have $|X_j| ≥ 3$. Hence, \[ \frac{|X_1|}{w_1} ≤ \frac{24}{7} < \frac{24}{5} ≤ \frac{|X_j|-1}{w_j}, \] which is a contradiction. Hence, if agent 1 is EF1-satisfied by $X$, then $|X_1| ≥ 3$. So, if agent 1 is M1S-satisfied by $X$, then $|X_1| ≥ 3$. Hence, if $X$ is M1S, then $X = A$.
For any allocation $X$, define \[ f(X) = \min_{j=1}^n \frac{|X_j|}{w_j}. \] Then $\WMMS_i = w_i\max_X f(X)$ for all $i \in [n]$. If $|X_1| ≤ 2$, then $f(X) ≤ |X_1|/w_1 ≤ 24/7$. If $|X_2| ≤ 1$ or $|X_3| ≤ 1$, $f(X) ≤ 24/5$. $f(A) = \min(3 \times 12/7, 2 \times 24/5) = 36/7$. Hence, $\max_X f(X) = 36/7$, so $\WMMS_1 = 3$ and $\WMMS_2 = 15/14$.
If an allocation $X$ is groupwise WMMS, then $X$ is WMMS, so $X = A$.
Consider a fair division instance $\Ical' = ([2], [5], V, w')$, where $w' = (14/19, 5/19)$, the agents have identical additive valuations, and all items are goods of value 1 each. To show that $A$ is groupwise WMMS for $\Ical$, it suffices to show that allocation $A'$ is WMMS for $\Ical'$, where $|A'_1| = 3$, and $|A'_2| = 2$.
Let $X$ be any allocation for instance $\Ical'$. If $|X_1| ≤ 2$, then $f(X) ≤ |X_1|/w'_1 ≤ 19/7$. If $|X_2| ≤ 1$, then $f(X) ≤ |X_2|/w'_2 ≤ 19/5$. $f(A') = \min(3 \times 19/14, 2 \times 19/5) = 57/14$. Hence, $\max_X f(X) = 57/14$, so $\WMMS_1(\Ical') = 3$ and $\WMMS_2(\Ical') = 15/14$. Hence, $A'$ is WMMS for $\Ical'$. Hence, $A$ is groupwise WMMS for $\Ical$.
Observation 7 can be inferred by repeatedly applying binary-additive.html to restricted instances.
Dependency for: None
Info:
- Depth: 8
- Number of transitive dependencies: 26
Transitive dependencies:
- /sets-and-relations/countable-set
- /analysis/sup-inf
- σ-algebra
- Set function
- Fair division
- Proportional allocation
- Restricted agents, pairwise fairness, and groupwise fairness
- Envy-freeness
- PROP1
- Minimum fair share
- EF1
- Maximin share of a set function
- Maximin share allocations
- Additive set function
- Subadditive and superadditive set functions
- PROP-unsat implies someone is superPROP
- EF1 implies PROP1
- Submodular function
- PROPx
- EFX
- Optimization: Dual and Lagrangian
- Dual of a linear program
- Linear programming: strong duality (incomplete)
- AnyPrice share
- Bound on k extreme values
- Additive goods and binary marginals