| paper | d4-conf-aaai-2010-0019 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | yes |
other→NP-hard discrete result → its continuous mirror
Answer: \(\mathrm{PSPACE}\text{-complete}\), hence \(\mathrm{NP}\text{-hard}\). · discrete side: other · adjudication: proved
full proof & adversarial review →
Theorem 3
statement extracted from the paper’s text layer
Given a finite set \(A\), supported propositional IIA types \(I_1,\ldots,I_\tau\) with rational masses \(\mu_i\) summing to \(1\), and sets \(X,Y\subseteq A\), decide whether \(\sum_{i:\,X\succeq_{I_i}Y}\mu_i\ge\frac12\), where \(X\succeq_{I_i}Y\) means an improving sequence under \(I_i\) leads from \(Y\) to \(X\).
A society consists of agents grouped by complete propositional IIA rule set \(I_i\), with rational type masses \(\mu_i\). For a pair of feasible argumentation extensions \(X,Y\subseteq A\), each type contributes its paper-defined improving-sequence dominance judgment; the decision asks whether the mass supporting \(X\succeq Y\) reaches one half.
Majority aggregation is not studied in the paper, and the translations alone do not establish why majority is the right collective rule; the stakeholder and extension-selection setting makes it a plausible question, but leaves that modeling choice open.
fatal: False
The mirror covers propositional IIA dominance and the computational bridge to CP-net dominance in Theorems 3 and 4. It leaves the lifted Hoare and Smyth orderings, predicate IIA complexity, and the broader argumentation proposal untouched.
My strongest case is Mass-IIA-Dominance, a continuous majority version of the paper’s dominance problem for incremental-improvement preferences.
Picture a large membership organization comparing packages of accepted arguments or policy claims. Let \(A\) be the shared set of arguments or claims; each member’s preferences over subsets of \(A\) are represented by a propositional IIA rule set \(I\). Members who share the same complete rule set are one type. If there are \(N\) members but only \(\tau\) distinct rule sets, the society is given by the types \(I_1,\ldots,I_\tau\) and their rational masses \(\mu_1,\ldots,\mu_\tau\), with \(\sum_i\mu_i=1\). The regime of interest is \(N\gg\tau\): many people, relatively few distinct preference models. This is plausible where members use shared preference templates or belong to recurring stakeholder groups; it is not a claim that every committee or population has high multiplicity.
The precise question is: given \(A\), those supported IIA types and masses, and two sets \(X,Y\subseteq A\), does at least half the society’s mass count \(X\) as at least as good as \(Y\) under its IIA model? Formally, decide whether
\[
\sum_{i:\,X\succeq_{I_i}Y}\mu_i \ge \frac12,
\]
where \(X\succeq_{I_i}Y\) means that the paper’s IIA semantics admits an improving sequence from \(Y\) to \(X\). The output is yes or no. This asks whether \(X\) has majority improving-sequence dominance over \(Y\); among a shared set of feasible argumentation extensions, it gives a pairwise relation for selecting preferred extensions. The masses are part of the population input, not a relaxation of the sets being compared.
The lead computational anchor is Theorem 3, proved in this paper. It translates any CP-net over \(A\) into an IIA rule set \(I_N\): positive CP-statements become gain rules and negative ones become loss rules, with conditions translated to refer to membership in the current set. For every pair \(E_1,E_2\subseteq A\), the theorem states that \(E_1\succeq_N E_2\) exactly when \(E_1\succeq_{I_N}E_2\). This is the direction that carries computational hardness from CP-net dominance into IIA dominance.
Theorem 4, also proved here, supplies the converse encoding. For any propositional IIA rule set \(I\), the authors construct a CP-net \(N_I\) with auxiliary variables that record gains, losses, and trades. For all \(E_1,E_2\subseteq A\), \(E_1\succeq_I E_2\) exactly when some state reachable from the encoding of \(E_2\) under \(N_I\) projects to \(E_1\). The paper says both reductions are polynomial-time constructible. Neither theorem is itself labeled a complexity-class result; the paragraph after Theorem 4 draws the consequence, using known CP-net results cited from Goldsmith et al. (2008), that IIA dominance is PSPACE-complete.
The resulting mass problem is also PSPACE-complete. With one type of mass \(1\), the majority query is exactly the paper’s IIA dominance query, so its hardness carries even at \(\tau=1\). For membership, check the types in sequence using polynomial space and sum their binary-rational masses exactly. This is a Class B mirror: the underlying hardness survives because it is already present within one preference type. The continuous formulation offers a meaningful complexity question, but I would not claim that continuization makes it tractable.
The mirror covers the paper’s IIA dominance result and its complexity bridge to CP-nets. It does not purport to continuize the Hoare and Smyth set orderings or the paper’s full argumentation proposal. A natural next question is whether the majority-dominance relation over a feasible extension set can have cycles, and what algorithms can find its maximal extensions. The predicate version of IIA is another follow-up, since the paper leaves its complexity questions open.
The weak point is that the paper studies a single preference model, while this mirror adds a society and a majority rule. That aggregation layer is not a result of the paper. The case for acceptance is that every member type uses exactly the paper’s own IIA semantics, and the query aggregates those same dominance judgments by population mass. A high-multiplicity society of such decision-makers is a plausible collective-choice setting, but authors who regard the single-agent model as essential could see this as an adjacent extension rather than a strict mirror.
Theorem 3 and Theorem 4 are genuine computational bridges: their polynomial translations let complexity results move between CP-nets and propositional IIA. But they concern one preference model at a time. The PSPACE-completeness statement that follows is inherited from known CP-net results; neither theorem studies a society, an aggregate preference, or a population parameter.
The proponent’s majority query is well-defined, and its type definition is sound: a complete IIA model can be a type, with rational mass assigned to each model. The difficulty is that this adds the decisive operation. The paper asks whether one model admits an improving sequence from \(Y\) to \(X\); the mirror asks whether a majority of models do. Majority aggregation is a reasonable social-choice problem, but it is not a population version of the computation in either theorem. The same wrapper could be placed around almost any binary judgment procedure.
A stronger application would use recurring stakeholder preference templates to rank feasible argumentation extensions by majority dominance. That is plausible, and the paper’s extension-selection motivation gives it some connection to the authors’ interests. Still, it requires a collective decision rule the paper neither represents nor analyzes. Without that added rule, a population of models gives only separate dominance queries; with it, the central question becomes how to aggregate their judgments. The IIA translations do not supply a reason to study that aggregation in particular.
So the strongest negative case is that this is an adjacent collective-choice model rather than a continuous mirror of the paper’s computational results. It is not a decisive case: the application is coherent, the population really is continuous in the programme’s sense, and the authors might recognize majority selection among extensions as their problem. I do not see an honest argument that rules out every worthwhile mirror here.
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.