2
$\begingroup$

Consider the following variant of the assignment problem. There are $I$ men and $J$ women. Each man can stay single or be assigned to a woman, and vice versa. Each potential match has value $u_{ij}$. The value of being single is $u_{i0}$ and $u_{0j}$ for men and women, respectively. Let $\mu_{ij}$ be an indicator equals one if $i$ is matched with $j$. $\mu_{i0}$ and $\mu_{0j}$ are the euqivalent indicator for singels. Each man can be matched with one woman only or stay single, and likewise for women.

$$ \mu_{i0} + \sum_{i=1}^{I} \mu_{ij} = 1 \quad \forall i$$

$$ \mu_{0j} + \sum_{j=1}^{J} \mu_{ij} = 1 \quad \forall j$$

The problem is to find a feasible assignment that maximizes the total value

$$ Max_{\mu}(\sum_{i=1}^{I}\sum_{j=1}^{J}u_{ij} \mu_{ij} + \sum_{i=1}^{I} u_{i0} \mu_{i0} + \sum_{j=1}^{J} u_{0j} \mu_{0j}) $$

My question is whether someone knows of a reference for an article that develops the auction algorithm for this case. I didn’t find it in Bertsekas (1998) or in a search on Google Scholar, but maybe I'm using the wrong terminology.

Edit: I posted a potential algorithm, but I would still appreciate references, so I won't need to reinvent the wheel.

$\endgroup$
0

2 Answers 2

1
$\begingroup$

I'm not aware of a reference and I doubt there is one, as this can be trivially solved using standard algorithms for the assignment problem. For each man $m$, add another artificial woman $a_m$ who can only be paired with $m$; the pair $(m,a_m)$ corresponds to $m$ remaining single. Then solve the resulting unbalanced assignment problem.

$\endgroup$
4
  • $\begingroup$ Very elegant, thanks! $\endgroup$
    – Muly
    Commented Aug 10 at 18:22
  • $\begingroup$ On a second thought: some of these artificial men and women might not be assigned eventually (if at least one real match is happening), which should not happen in a standard assignment problem. Perhaps a solution is to add also potential links between all artificial men and women (but this would add too many non empty cells to the matrix) $\endgroup$
    – Muly
    Commented Aug 10 at 20:05
  • $\begingroup$ @Muly, See revised answer, which I believe addresses your concern. Does it seem right to you? (There are no artificial mean.) $\endgroup$
    – D.W.
    Commented Aug 10 at 21:25
  • 1
    $\begingroup$ I got it now; thanks a lot! I now see that it does appear on Bertsekas (1998) as an exercise (7.11) with the solution you suggested. $\endgroup$
    – Muly
    Commented Aug 11 at 5:57
0
$\begingroup$

I'm thinking now about the following extension of the auction algorithm (still need to check if it works)

Normalize the surplus of singles to 0. Let $u_{ij}$ be the (normalized) joint surplus and $m_{ij}$ the meetings ($i$ and $j$ can be matched only if $m_{ij}=1$).

  1. Start with an empty assignment $S$, a vector of initial prices $w_i = 0$, and some $\epsilon >0$

  2. Iterate on the two following phases

a) Bidding Phase

Let $J(S)$ be a nonempty subset of women $j$ that are unassigned under the assignment $S$. For each woman $j \in J(S)$

  • Find a ``best'' man $i_j$ having maximum value, i.e.,

$$ i_j = \text{arg} \max_{i \in m(j) } u_{ij} - w_i$$
and the corresponding value

$$ v_j = \max_{i \in m(j) } u_{ij} - w_i$$ and find the best value offered by men other than $ i_j $ $$q_j = \max_{i \in m(j) , i \ne i_j } u_{ij} - w_i$$

  • If $v_j > 0$, woman $j$ makes a ``bid'' for man $i_j$.

  • The bid is given by: $$b_{ij} = w_{i_j} + v_j - q_j + \epsilon$$
    b) Assignment Phase

For each man $i$, let $B(i)$ be the set of women from which $i$ received a bid.

  • If $B(i)$ is non-empty, increase $w_i$ to the highest bid:

$$ w_i = \max_{j \in B(i)} b_{ij}$$

  • Assign $i$ to the woman in $B(i)$ attaining the maximum above
  1. Terminate when no new assignment is made
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.