| paper | d4-conf-wine-2020-0027 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
Theorem 4.3 and Theorem 5.3 establish price-of-anarchy welfare bounds, but assert neither computation nor a complexity guarantee. The LPs serve as proof tools, while the proposed population of auction instances averages separate two-buyer games rather than distributing mass across agents in one game. Neither anchor meets the computational threshold.
fails bit a — no named computational result to mirror
No theorem is mirrored. The proposed distribution over auctions averages separate instances and does not continuize the buyers in either result.
There is no qualifying computational-complexity anchor in this paper. No numbered theorem, lemma, corollary, or proposition says that an underlying problem is in P, NP-hard, FPT, or similar. The paper uses LPs to prove efficiency bounds; it does not study the complexity of computing an auction outcome or an efficiency bound for a given instance.
The nearest named results are proved here, but they are analytic. Theorem 4.3 says that for every \(T\), every equilibrium path in a two-buyer auction with non-decreasing concave valuations has efficiency at least \(1-1/e\), and that this bound is asymptotically tight as \(T\) grows. Theorem 5.3 says that with general non-decreasing valuations, the price of anarchy is exactly \(1/T\). These are not computational-complexity results, so they cannot serve as the required anchors.
The closest population lift I can state precisely is a population of independent auction markets. Its instance is a fixed \(T\), a finite support of complete valuation-pair types \(q=(V_1^q,V_2^q)\), and rational masses \(\mu_q\) summing to \(1\). Choose an equilibrium path \(P_q\) for each type. The objective is to minimize aggregate efficiency,
\[
\frac{\sum_q \mu_q\,\mathrm{sw}(P_q)}
{\sum_q \mu_q\,\mathrm{opt}(q)}.
\]
The question is to find the minimizing path choices and their ratio. Theorem 5.3 implies a lower bound of \(1/T\), and a distribution concentrated on a tight profile attains it. But this continuizes a population of *markets*, while each market still has exactly two buyers; the masses only average completed auctions. It is not a continuous population of the agents in the paper’s problem, and I do not think the authors or a referee would recognize it as their continuous analogue.
Making the buyers themselves into large cohorts would change the strategic model, so it would not preserve the paper’s question. Thus I can offer this precise nearby construction, but no honest continuous problem per qualifying anchor. The paper does contain named results, but none of the required kind.
The strongest case against a mirror is that this paper proves no computational-complexity result. Theorem 4.3 gives a price-of-anarchy bound for a fixed two-buyer game with concave valuations; Theorem 5.3 gives the bound for general valuations. Both are statements about welfare along equilibria, not about the complexity of computing an outcome, bound, or algorithm. The LPs are proof certificates over one pair of buyers’ marginal values. They contain no population of types, high-multiplicity input, or algorithmic problem whose complexity the paper studies. A mirror would have to supply that missing computational question before it could mirror either theorem.
The most natural construction is a distribution over complete auction instances \(q=(V_1^q,V_2^q)\), with rational masses \(\mu_q\), and an objective such as
\[
\frac{\sum_q \mu_q\,\mathrm{sw}(P_q)}
{\sum_q \mu_q\,\mathrm{opt}(q)}.
\]
That is well-defined, but its population consists of markets, each still containing exactly two buyers. The masses average separate games; they do not continuize the agents inside the game. Treating many repeated auctions as a population of markets could be useful for another project, but it does not give these results a continuous-agent mirror.
A stronger alternative would put a continuum of buyers into one auction. That changes the model in a more fundamental way. With indivisible items and individual allocations, finitely many items go to finitely many agents, so welfare per unit of population can collapse. If instead buyers receive fractional shares, or negligible agents act as price takers, the two-buyer forward-utility differences and strategic bids at the heart of the paper disappear. Keeping two mass-bearing coalitions preserves the strategic game, but the coalitions remain the two players; their mass adds no high-multiplicity structure. This dilemma applies to both anchors: Theorem 4.3’s concavity assumption concerns each buyer’s marginal values, and Theorem 5.3’s tight example depends on the distinct marginal-value sequences of the two buyers. The limit either loses the paper’s strategic object or retains only two strategic actors.
The paper’s \(T\to\infty\) statement in Theorem 4.3 varies the number of items, not the population. So it does not provide a hidden population regime to continue. A many-buyer auction or a continuum of auction markets might still be worth studying on its own, but that would be a new computational question rather than a continuous mirror of these theorems. I cannot establish that no such research could ever be worthwhile; the negative case is strongest for the paper as written, because neither named result supplies a computational anchor to continuize.
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.