EF doesn't imply PROP for supermodular valuations

Dependencies:

  1. Fair division
  2. Envy-freeness
  3. AnyPrice share
  4. Maximin share allocations
  5. PROP1
  6. PROPm
  7. Submodular function
  8. Bound on k extreme values

Consider a fair division instance with 2 agents and 4 goods. The agents have identical valuations and equal entitlements. Let $a, b \in \mathbb{R}_{\ge 0}$ such that $3a < b$. The valuation function $v$ is given by \[ v(S) \defeq \begin{cases} |S|a & \textrm{ if } |S| ≤ 3 \\ 3a + b & \textrm{ if } |S| = 4 \end{cases}. \] Then $v$ is supermodular, no PROP1 allocation exists, and if $a > 0$, then no PROPm allocation exists. Let $A$ be an allocation where each agent gets 2 goods. Then $A$ is EF, APS, and MMS.

Proof

For any $g \in [4]$ and $S \subseteq [4] \setminus \{g\}$, we have \[ v(g \mid S) = \begin{cases} a & \textrm{ if } |S| ≤ 2 \\ b & \textrm{ if } |S| = 3 \end{cases}. \] Hence, $v$ is supermodular.

The proportional share is $(3a + b)/2$. In any allocation, some agent gets at most 2 goods, and even if she is given an additional good, her valuation is $3a < (3a+b)/2$. Hence, no allocation is PROP1, and no allocation is PROPm if $a > 0$. (When $a = 0$, every allocation is PROPm.)

$v(A_1) = v(A_2) = 2a$, so $A$ is EF. It is easy to check that the MMS is $2a$.

If we set the price of each good to 1, then at most 2 goods are affordable. Hence, APS is at most $2a$. Moreover, for any price vector, the cheapest 2 goods are affordable, and their total valuation is $2a$. Hence, APS is at least $2a$.

Dependency for: None

Info:

Transitive dependencies:

  1. /sets-and-relations/countable-set
  2. /analysis/sup-inf
  3. σ-algebra
  4. Set function
  5. Fair division
  6. Proportional allocation
  7. Envy-freeness
  8. PROP1
  9. Maximin share of a set function
  10. Maximin share allocations
  11. Submodular function
  12. PROPx
  13. PROPm
  14. Optimization: Dual and Lagrangian
  15. Dual of a linear program
  16. Linear programming: strong duality (incomplete)
  17. AnyPrice share
  18. Bound on k extreme values