| paper | d4-conf-atal-2013-0014 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | yes |
P→P discrete result → its continuous mirror
Answer: An M-PCE is exactly computable in time \(\operatorname{poly}(m,\tau,L)\), with algebraic strategy coordinates represented by integer polynomials and rational isolating intervals. · discrete side: polynomial · adjudication: proved_with_minor_gaps
full proof & adversarial review →
Theorem 11
statement extracted from the paper’s text layer
Given two finite action sets and a finite support of member types \(t\), each assigned to side \(i(t)\in\{1,2\}\), with rational mass \(\mu_t\) and rational utility table \(r_t(a_1,a_2)\), let \(q_i=\sum_{t:i(t)=i}\mu_t>0\) and \(U_i^\mu(a_1,a_2)=q_i^{-1}\sum_{t:i(t)=i}\mu_t r_t(a_1,a_2)\). Compute mixed strategies \(x,y\) forming an M-PCE of the induced two-player game; equivalently, maximize \(\alpha\) subject to \(U_i^\mu(x,y)\ge \mathrm{BU}_i^\mu+\alpha\) for both \(i\), where \(\mathrm{BU}_i^\mu\) is the paper’s best-response benchmark in the induced game.
A continuous society of members divided between two federations, represented by a rational distribution \(\mu\) over complete utility-table types. Each federation is a strategic player whose payoff is the mean utility of its constituency; the federations choose mixed strategies over finite action sets, and the objective is to compute an M-PCE maximizing the common margin above the paper’s best-response benchmarks.
The member types affect only the associations’ payoff averages, so the theorem still computes equilibrium between two strategic actors rather than making member-level decisions part of the computation.
fatal: False
The mirror covers Theorems 10 and 11: PCE existence and construction, and M-PCE construction in general two-player games. It leaves the coco value, the axiomatic and algebraic characterizations, and the separate side-payment results aside.
The mirror I would defend is a game between two large associations—say, worker and employer federations—that each choose one of finitely many bargaining actions for their members. Members differ in their preferences over the possible pairs of actions. A type records the member’s side and their complete utility table \(r_t(a_1,a_2)\); all members of that type have the same table. The society is a distribution \(\mu\) over these types, with rational masses and support size \(\tau\). If side \(i\) has total mass \(q_i>0\), its association’s payoff is its constituency’s mean utility:
\[
U_i^\mu(a_1,a_2)=\frac{1}{q_i}\sum_{t:\,i(t)=i}\mu_t r_t(a_1,a_2).
\]
This gives a two-player game whose payoff tables are computed from the continuous population. The regime is many members—potentially millions—with relatively few complete utility types. Its high-multiplicity interpretation is exact: a finite constituency with \(n_t\) members of type \(t\) gives \(\mu_t=n_t/N\).
The associations’ decision variables are mixed strategies over their finite action sets, as in the paper. Those probabilities are inherited from its model; the continuous object in this mirror is the population distribution \(\mu\). The associations choose on behalf of their members, and members’ utilities determine the associations’ payoffs. That collective decision is the scenario that makes the two-player model plausible.
My lead anchor is Theorem 11, proved in this paper. It says that for any two-player normal-form game, an M-PCE can be computed in polynomial time. In the mirror, the problem is: given the finite action sets, rational type masses, and each type’s utility table, compute an M-PCE of the induced game. The solution is a pair of mixed strategies \((x,y)\) attaining the greatest possible common margin above the players’ best-response benchmarks:
\[
\alpha^*=\max_{x,y}\min_{i\in\{1,2\}}\bigl(U_i^\mu(x,y)-\mathrm{BU}_i^\mu\bigr).
\]
Here \(\mathrm{BU}_i^\mu\) is the best payoff player \(i\) can obtain when the other player best responds. The paper’s M-PCE definition guarantees such a profile exists. The expected classification is tractable: aggregate the rational type-weighted payoff tables, then use the paper’s polynomial-time algorithm. For the natural explicit input, this is polynomial in the payoff-table size, \(\tau\), and the bit length of the masses and utilities.
A second, independently useful anchor is Theorem 10, also proved here. It says that for any two-player normal-form game, one can decide in polynomial time whether a PCE exists and, if so, compute one. Its continuous mirror takes the same population input and asks whether there are mixed strategies \((x,y)\) satisfying
\[
U_i^\mu(x,y)\ge \mathrm{BU}_i^\mu\qquad\text{for both }i,
\]
returning such strategies when they exist. Unlike M-PCE, PCE need not exist. This problem is also expected to be tractable by aggregating the type-weighted payoffs and applying Theorem 10. A further question is how the regions of mass vectors \(\mu\) where a PCE exists change as constituency composition varies; for M-PCE, the analogous question is how \(\alpha^*\) varies with \(\mu\).
The authors should recognize both as their problems: after aggregation, the object being computed is exactly a PCE or M-PCE of a two-player normal-form game, with the paper’s benchmark and equilibrium definitions intact. The continuous input also provides a direct high-multiplicity formulation of the same computation. This case covers the paper’s PCE and M-PCE computation results, not its axiomatic characterizations or its separate side-payment construction. The paper supplies no hardness anchor here; both anchors are algorithmic.
The weakest point is that the continuum is the associations’ constituency, not a continuum of strategic players. The population affects the payoff tables, while the strategic decision remains with two collective actors. If a referee requires the members themselves to be the players, this mirror may not qualify. Its case rests on the associations being credible collective actors whose objectives are their constituencies’ mean utilities.
The strongest case against both anchors is that Theorems 10 and 11 concern a two-player game, while the proposed population only supplies its payoff tables. For type masses \(\mu_t\), the associations’ payoffs are
\[ U_i^\mu(a_1,a_2)=\frac{1}{q_i}\sum_{t:\,i(t)=i}\mu_t r_t(a_1,a_2). \]
Once those averages are computed, the type distribution disappears. PCE and M-PCE concern the associations’ mixed strategies and best-response benchmarks; neither result changes, for example, how population mass can be transferred, controlled, or campaigned over. The computation is therefore a two-player game computation with a population-based input encoding. That makes the proposed mirror a weak fit for a programme whose target is computational questions about a continuous society.
Making the members themselves the strategic players would give the population a more direct role, but it would change the problem. Theorems 10 and 11 rely on two players’ unilateral best responses. A continuum of agents needs a different strategic model and different cooperative-equilibrium definitions; an individual’s deviation may have no effect on aggregate behavior. That would be a new problem rather than a mirror of these theorems.
This case is limited. A large constituency represented by a federation is plausible, and common cardinal utility units can make the aggregation coherent. Theorem 10 gives a named computational problem for PCE existence and construction; Theorem 11 does the same for M-PCE. Authors could reasonably recognize computing either concept from constituency-derived payoffs as their problem. Under the programme’s stated standard, that recognition matters, and a simple reduction to their existing algorithm does not by itself defeat a well-posed mirror. So I cannot honestly claim the anchors have been defeated: the negative case turns on whether the programme requires population-level computation, and it remains vulnerable if it accepts a continuous population as the source of a collective game’s payoffs.
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.