| paper | d4-conf-atal-2014-0345 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | yes |
NP-hard→P discrete result → its continuous mirror
Answer: Polynomial-time solvable: \(\mathrm P\). · discrete side: hardness · adjudication: proved
full proof & adversarial review →
Theorem 1
statement extracted from the paper’s text layer
Given a finite set of complete valuation types \(T\), rational masses \(\mu_t\) summing to \(1\), a fixed rational plot length \(\ell\), piecewise-constant rational valuation densities \(f_t\), and a piecewise-constant rational per-capita capacity profile \(b\) on \([0,1]\), determine the exact complexity of maximizing welfare over mass allocations to length-\(\ell\) intervals. Each unit of allocated mass must receive one whole interval; the mass of type \(t\) assigned to intervals is at most \(\mu_t\), and total mass assigned to intervals containing location \(z\) is at most \(b(z)\) almost everywhere. The objective is \(\sum_{t\in T}\int_{\mathcal I_\ell}(\int_I f_t(z)\,dz)\,dx_t(I)\), where \(\mathcal I_\ell=\{[a,a+\ell]:0\le a\le1-\ell\}\) and \(x_t\) is the allocation measure for type \(t\).
A continuum of advertisers with type distribution \(\mu\) competes for parallel copies of a broadcast schedule, with per-capita capacity \(b(z)\). Each individual gets at most one integral, fixed-length interval; type mass may split across interval choices. Choose measures \(x_t\) to maximize \(\sum_t\int_{\mathcal I_\ell}v_t(I)\,dx_t(I)\), subject to \(\int dx_t\le\mu_t\) and \(\sum_t\int\mathbf{1}[z\in I]dx_t(I)\le b(z)\) almost everywhere.
This mirrors Theorem 1's fixed-length welfare optimization. It leaves Theorems 2–4's approximation mechanisms, truthfulness guarantees, and discrete-resource results outside the mirror.
The strongest case I believe is a narrow mirror of Theorem 1, the paper’s fixed plot size welfare hardness result. It states that maximizing utilitarian welfare is NP-hard even when every allocated piece must have the same fixed length \(\ell\); the paper also says this holds in the discrete setting when each piece contains the same number of items. The theorem is explicitly “Following from [2]” (Aumann, Dombb, and Hassidim, *Computing Socially-Efficient Cake Divisions*), so this paper cites the hardness rather than proving it.
I would ask whether that hardness survives when the *population* is continuous. Call the problem Mass Fixed-Plot Welfare. An instance has a finite set \(T\) of complete valuation types, rational masses \(\mu_t\) summing to \(1\), a fixed rational plot length \(\ell\), and a per-capita resource capacity profile \(b\) over \([0,1]\). Each type \(t\) has a non-atomic valuation measure \(v_t\), given by a nonnegative rational piecewise-constant density. The profile \(b(z)\) means the supply of parallel copies of the schedule at location \(z\), per unit of population. Thus a population of \(N\) agents has \(N b(z)\) units of capacity there.
An allocation is a collection of nonnegative measures \(x_t\) over the length-\(\ell\) intervals \(I\subseteq[0,1]\). Here \(x_t(A)\) is the mass of type \(t\) receiving one whole interval from \(A\). Require \(x_t(\mathcal I_\ell)\leq\mu_t\) for every type, and \(\sum_t x_t(\{I:z\in I\})\leq b(z)\) almost everywhere. The question is to maximize \(\sum_t\int_{\mathcal I_\ell}v_t(I)\,dx_t(I)\), and return an optimal allocation measure. Every individual still gets at most one integral, fixed-length interval; only the type-level mass can split across choices.
The regime I have in mind is a large market for repeated or parallel broadcast inventory. Many advertisers have the same complete valuation curve because they use the same campaign template, target the same audience, and face the same terms. There may be millions of advertisers but only a modest number \(\tau\) of such types, with inventory scaling in proportion to the market. This keeps the paper’s central decision intact—who gets which contiguous time slot to maximize total value—while making high multiplicity meaningful. It also uses the programme’s permitted supply co-modification.
I expect this version to be Class A for fixed plot size and the stated finite input format. It has the shape of a linear optimization problem over interval choices, with type-mass limits and local capacity constraints. A plausible route is dual prices for capacity plus a pricing problem over intervals; with fixed \(\ell\) and piecewise-constant rational input, that pricing step resembles maximizing a moving-window value. The work to prove is a polynomial-time exact method in \(\tau\), the number of breakpoints, and the input bit length. Further questions include how to round the mass allocation to a finite market, whether truthful approximation mechanisms survive when types report their valuations, and how the answer changes for arbitrary interval lengths or richer valuation functions.
This mirrors only Theorem 1’s fixed-size utilitarian welfare optimization, not Theorems 2–4’s mechanisms, truthfulness guarantees, or approximation ratios. I think the authors would recognize it as a large-market version of their fixed-size allocation problem: it preserves valuation by location, the one-interval restriction, and the welfare objective.
My weakest point is the resource scaling. In the paper’s single-cake setting, dividing capacity by a population of \(N\) makes per-capita supply vanish as \(N\) grows. Keeping a nondegenerate problem requires the parallel or repeated inventory in my scenario, and the authors or a referee could regard that as too substantial a change. The type restriction may also remove the identity-level combinatorics behind the cited hardness. That is why this is a credible question about whether their hardness dissolves under high multiplicity, not a claim that Theorem 1’s NP-hardness transfers.
I don’t see an honest way to defeat the strongest anchor. Theorem 1 is a named computational result: optimal welfare is NP-hard even with fixed-size plots. The fact that the paper attributes it to [2] makes it inherited, but it still gives a clear problem to mirror.
The best objection is that the proposed model escapes the scarcity of a single cake by scaling inventory with population. With one fixed cake and \(N\) bidders, per-capita supply tends to zero. The repair replaces that setting with repeated or parallel inventory, then asks a fractional packing problem over types and intervals. That could make the result less specific to this paper’s single-cake geometry.
But the objection does not hold against the better mirror. The programme explicitly permits supply to scale when continuizing the population forces that change. Repeated broadcast inventory is a plausible setting; advertisers with the same complete valuation can form a type; and splitting a type’s mass across allocations still gives each individual one whole interval. The cake’s existing continuity is in the outcome, so it does not mean the population-continuous computational question is already answered. If the resulting packing problem is tractable, that is a potentially valuable answer, not a reason the question was unworthy.
So the negative case is weak. I can identify a substantial modeling change in the proposed repair, but not a sound basis for rejecting every recognizable population-continuous version of Theorem 1.
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.