| paper | Optimization-Based Voting Rule Design: The Closer to Utopia the Better |
| authors | Piotr Faliszewski, Stanislaw Szufa, Nimrod Talmon |
| venue | ATAL 2018 |
| filed under | multiwinner · multiwinner |
| judged by | gpt-5.6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
The sole proposed anchor is an empirical finding, not a numbered theorem, lemma, proposition, or corollary asserting a computational result. The cited NP-hardness and polynomial-time claims belong to other papers and cannot supply the missing bit. A plausible continuous population extension therefore cannot rescue this paper under the anchor requirement.
fails bit a — no named computational result to mirror
The proposed single-μ objective replaces the paper's average over sampled elections; preserving the paper's question requires an outer distribution over \((q,μ)\).
fatal: False
The mirror covers the rule-design objective for OWA/Borda-based rules with continuous voter masses, but not the paper's unguaranteed empirical discoveries or its cited external complexity results.
Strictly speaking, this paper has no eligible anchor under the programme’s rules. It contains no numbered Theorem, Lemma, Proposition, or Corollary asserting a complexity result. Its numbered “Our Contribution” claims are empirical conclusions from local search and experiments, not formal computational theorems. The paper’s statements that Chamberlin–Courant is NP-hard and that Peters gives a polynomial-time algorithm are cited from elsewhere, not proved here. Thus there is no fully compliant theorem-anchored case to make.
The strongest honest positive case uses the nearest available anchor, “Our Contribution (2),” while admitting that it is not a theorem. The paper reports that, among OWA/Borda-based rules, optimizing closeness to the diversity utopia recovers Chamberlin–Courant, while intermediate utopias produce compromise rules resembling linear or harmonic OWA vectors. That claim is developed and experimentally supported in this paper.
A natural continuous mirror is:
\[ \textsc{Continuous-OWA-Utopia-Design}. \]
An instance contains a finite candidate set \(C\), candidate positions \(q(c)\in[0,1]\), a committee size \(k\), a finite set of ranking types \(T\), a rational society distribution \(\mu\) over \(T\), and a target utopic distribution \(U\) on \([0,1]\), such as the uniform distribution. Each type \(t\) is a complete ranking of \(C\), and \(\mu_t\) is the fraction of voters with that ranking. For a non-increasing OWA vector
\[ \lambda=(\lambda_1,\ldots,\lambda_k),\qquad 1=\lambda_1\ge\lambda_2\ge\cdots\ge\lambda_k\ge 0, \]
the score of a committee \(S\) is
\[ \operatorname{score}_{\lambda,\mu}(S) = \sum_{t\in T}\mu_t \sum_{j=1}^{k} \lambda_j\beta_m(i_j(t,S)), \]
where \(i_1(t,S)<\cdots<i_k(t,S)\) are the positions of the members of \(S\) in type \(t\)’s ranking and \(\beta_m\) is Borda scoring. The winning committee is the lexicographically first maximizer. If
\[ d_S=\frac1k\sum_{c\in S}\delta_{q(c)}, \]
the task is to find \(\lambda\) minimizing the Earth-mover distance \(W_1(U,d_S)\), where \(S\) is the committee selected under \(\lambda\).
This is recognizably the paper’s problem: the candidates remain finite, the rule class remains exactly OWA/Borda, the utopia and EMD objective remain intact, and only the voter population is replaced by its distribution. The paper’s voter ideal points induce ranking types; in the one-dimensional model, with fixed candidate positions, there are at most \(O(m)\) distinct ranking types. Thus a society of millions of voters can have only linearly many types, with \(\mu_t\) recording population shares. This is a strong high-multiplicity regime, not an arbitrary reinterpretation.
I would expect a mixed complexity picture. For fixed \(m\), \(k\), or bounded one-dimensional type structure, the problem is plausibly tractable by enumerating the finitely many regions of OWA-vector space in which the winning committee is unchanged. For unrestricted \(m\) and \(k\), hardness is more likely to transfer: a discrete profile maps directly to \(\mu_t=n_t/n\), and the special vector \((1,0,\ldots,0)\) gives Chamberlin–Courant. The paper’s cited NP-hardness therefore suggests that the continuum does not automatically dissolve the winner-determination combinatorics. A separate continuum-specific hardness result for optimizing \(\lambda\) would be an interesting follow-up.
The weakest point is decisive: the paper does not itself prove the named computational result needed by the brief. Moreover, its actual optimization is a finite-sample, randomized proof of concept rather than an exact high-multiplicity algorithm. So this is a credible continuous analogue of the paper’s central mathematical question, but not a properly anchored ChoCo case unless numbered contribution (2) is accepted as an algorithmic anchor.
The decisive negative point is formal: this paper has no eligible computational result to mirror. It contains no numbered Theorem, Lemma, Proposition, or Corollary establishing a complexity, exact, approximation, or parameterized result. “Our Contribution (2)” is an empirical report from randomized local search. It does not prove that the returned OWA vector is optimal, that Chamberlin–Courant is recovered in general, or even that the procedure terminates with a meaningful guarantee. The NP-hardness of Chamberlin–Courant and Peters’s polynomial-time result are cited results from other papers, not results established here.
The proposed \(\textsc{Continuous-OWA-Utopia-Design}\) is nevertheless a sensible new problem, but that is precisely its weakness as a mirror. The paper optimizes over a fixed finite sample of randomly generated elections and reports an average empirical distance. A single society distribution \(\mu\) instead produces one weighted election and one winning committee, hence one value \(W_1(U,d_S)\); that is a different objective. To recover the paper’s objective, one must add a distribution over candidate configurations, societies, or elections and optimize an expected distance. That may be worthwhile, but it formalizes a new stochastic design problem rather than a computational question asserted by this paper.
The high-multiplicity construction itself is not objectionable: in a one-dimensional model, many voters can indeed share one of \(O(m)\) ranking types. Nor would it be valid to reject the mirror because the answer might be easy or because the cited hardness might survive. The point is narrower and fatal: the paper never states the result whose continuous counterpart is being proposed. The same applies to its other claims, such as recovering \(k\)-Borda, SNTV, or Chamberlin–Courant experimentally.
Thus I cannot honestly claim that no worthwhile continuous population model could ever be built from this paper. The proposed model could support a new ChoCo project. But it cannot rescue this paper under the programme’s anchor requirement. The strongest negative verdict is therefore that the paper is not an admissible source of a continuous-computational mirror, even though the broader modelling idea remains plausible.
The adversarial triple: the proponent anchors on up to three named results; the opponent sees that case and must defeat every anchor; the judge decides which case convinced it. These are the pipeline’s own outputs, generated by tools/triple_run.py — no human edited them. The paper’s own text is not reproduced here beyond the quoted statement above.