When Votes Change and Committees Should (Not)

The best mirror is a high-multiplicity version of the paper’s own multistage SNTV model, with committees remaining discrete and only the population being continuized.

For a fixed candidate set \(C\) and stages \(1,\ldots,\tau\), a type is a complete temporal voting trajectory
\[ \theta=(\theta_1,\ldots,\theta_\tau)\in(C\cup\{\emptyset\})^\tau . \]
Here \(\theta_t\) is the candidate chosen by that type at stage \(t\), with \(\emptyset\) representing abstention. A society is a rational distribution \(\mu\) over the finitely many trajectory types in the instance. If \(Q\subseteq C\) is a committee, its supported population at stage \(t\) is
\[ \operatorname{score}^{\mu}_t(Q) =\sum_{\theta:\theta_t\in Q}\mu_\theta . \]

The decision variable is still a sequence of ordinary committees \(Q_1,\ldots,Q_\tau\), with \(|Q_t|\le k\). The objective is to maximize the minimum supported mass,
\[ \max_{Q_1,\ldots,Q_\tau}\min_t \operatorname{score}^{\mu}_t(Q_t), \]
subject either to \(|Q_t\triangle Q_{t+1}|\le\ell\) in the conservative model or \(|Q_t\triangle Q_{t+1}|\ge\ell\) in the revolutionary model. The decision version asks whether this value is at least a rational threshold \(\rho\).

This is a genuine population continuation rather than fractional committee selection. A finite election embeds exactly: given \(n\) agents, group agents with the same temporal trajectory, set \(\mu_\theta=n_\theta/n\), and replace \(x\) by \(\rho=x/n\). Conversely, rational masses can be realized by a sufficiently large finite population.

A convincing regime is a large recurring community planning a sequence of exhibitions, buffets, or events. There may be tens of thousands of participants but only a few dozen recurring preference trajectories: for example, cohorts of visitors, departments, or customer segments that choose the same preferred bundle or sculpture on each day. Thus the population size \(N\) is much larger than the number \(r\) of nonzero trajectory types, while \(m\) and \(\tau\) remain operationally meaningful. The full trajectory is important: it preserves that the same agents appear across all stages, rather than replacing them by unrelated stage-wise marginals.

My lead anchor is Theorem 1(i), proved in this paper, with technical details marked \(\star\): MSNTV is NP-hard even for two agents, \(\ell=0\), \(x=1\), and \(k=|C|/2\). Its continuous counterpart is:

Continuous Conservative Multistage SNTV\(_\infty\). Given \(C\), \(\tau\), \(k\), \(\ell\), a rational distribution \(\mu\) over temporal SNTV types, and \(\rho\in[0,1]\), decide whether there are committees \(Q_1,\ldots,Q_\tau\) such that \(|Q_t|\le k\), \(\operatorname{score}^{\mu}_t(Q_t)\ge\rho\) for every \(t\), and \(|Q_t\triangle Q_{t+1}|\le\ell\) for every consecutive pair.

The expected classification of the unrestricted problem is Class B: hardness transfers directly through the rational finite-election embedding. This is not a continuum-specific hardness claim; the difficult combinatorics live in the candidate choices and temporal consistency, not in the number of named voters. The more interesting ChoCo question is whether the hardness survives in the genuinely high-multiplicity restriction \(r\ll N\), or whether the type-compressed version becomes tractable.

The paper’s Theorem 1(ii), also proved here and obtained using Lemma 1, gives a separate revolutionary anchor: RMSNTV is NP-hard even for \(\ell=2k\), \(x=1\), and \(k=|C|/2\). Its mirror is:

Continuous Revolutionary Multistage SNTV\(_\infty\). Given the same data, decide whether there are committees \(Q_1,\ldots,Q_\tau\) with \(|Q_t|\le k\), \(\operatorname{score}^{\mu}_t(Q_t)\ge\rho\), and \(|Q_t\triangle Q_{t+1}|\ge\ell\) at every transition.

This too is expected to be Class B in the unrestricted setting. It is not merely a cosmetic variant: in the high-multiplicity application, it asks for a deliberately rotating sequence of bundles or exhibition items rather than continuity between stages. The natural further question is whether large-support mass distributions make the lower-bound transition constraints easier, and whether a model with both lower and upper bounds has a distinct continuous complexity profile.

A third useful anchor is Theorem 2(ii), proved in the paper by reduction from Dominating Set: MSNTV is W[2]-hard parameterized by \(k\), even when \(x=1\) and \(\ell=0\). The corresponding parameterized problem is exactly Continuous Conservative Multistage SNTV\(_\infty\), parameterized by committee size \(k\). The same parameter-preserving finite embedding gives W[2]-hardness for the unrestricted rational-mass version. This asks whether population aggregation can remove the apparent difficulty of choosing a small temporally consistent committee. The expected answer is again Class B unless the type-compressed restriction is imposed as an additional parameter.

There is also a plausible tractable side. Theorem 4 gives an \(O(\tau m^{2k+1}n)\)-time algorithm for both MSNTV and RMSNTV. Its state-graph argument only needs stage-wise score computations, so in the explicit type-mass model the \(n\)-scan should be replaceable by a weighted scan over \(r\) types, yielding an exact running time of roughly
\[ \tau m^{2k+1}r\cdot\operatorname{poly}(L), \]
where \(L\) is the mass encoding length. Thus the continuous mirror should retain XP parameterized by \(k\), and the paper’s Corollary 1 should plausibly extend to FPT parameterized by \(m\). This is a meaningful computational benefit: repeated participants are processed as one weighted cohort, not copied individually.

The mirror covers the paper’s MSNTV and RMSNTV decision problems, their NP-hardness results in Theorem 1, the W[2]-hardness result in Theorem 2(ii), and plausibly the algorithms in Theorem 4 and Corollary 1. It does not claim to continuize the online variants, the kernelization theorems, or more expressive preference rules without separate work.

The weakest point is that the paper’s strongest hardness theorem uses only two agents. That proves exact hardness of the general mass model, but says nothing by itself about the intended regime \(N\gg r\). If every participant has an idiosyncratic temporal trajectory, then \(r=N\) and the mirror is merely a weighted restatement. The positive case therefore depends on treating the high-multiplicity restriction as the real research object: recurring temporal vote types, rational masses encoded compactly, and algorithms or hardness results measured in \(r,m,\tau\), and \(L\), not in an expanded population. That is a genuine continuous-population question the paper’s authors would recognize, while preserving their committees, scores, time dimension, and conservative/revolutionary distinction exactly.