On a Voter Model with Context-Dependent Opinion Adoption

The strongest positive case is narrow but real: the paper’s complete-graph results admit clean high-multiplicity mirrors in which the only continuous object is the population mass of each opinion. The most convincing anchor is Theorem 3; Theorem 1 gives a second, more direct mirror. Both theorems are proved in this paper, although Theorem 1 builds on the finite birth–death analysis of Glaz.

My lead problem is Weak-Bias Continuum Fixation, mirroring Theorem 3.

Consider a very large, well-mixed community. Every agent is structurally identical and has one of two opinions, \(0\) or \(1\). Thus the type set is \(T=\{0,1\}\), and a society is \(\mu=(1-p,p)\), where \(p\) is the mass holding opinion \(1\). Agents update asynchronously exactly as in the paper: an active agent samples a uniformly random neighbour and adopts that neighbour’s opinion with a probability depending on the ordered pair of opinions.

To obtain a non-degenerate population limit, use the standard weak-bias scaling
\[ \alpha_{10}=\frac12,\qquad \alpha_{01}^{(N)}=\frac12\left(1+\frac{\lambda}{N}\right), \]
where \(\lambda\in[-1/2,1/2]\). The \(N\)-agent approximant starts with \(k_N=\lfloor pN\rfloor\) agents holding opinion \(1\). The finite population has bias ratio \(r_N=1+\lambda/N\).

The continuous problem, which I would call \(\mathrm{WBCF}_\infty\), is:

Given rational \(p\), rational \(\lambda\in[-1/2,1/2]\), and an accuracy parameter \(\varepsilon>0\), output a rational \(z\) satisfying
\[ \left|z-\Phi(p,\lambda)\right|\le \varepsilon, \]
where
\[ \Phi(p,\lambda) = \lim_{N\to\infty} \Pr[\text{opinion }1\text{ fixates}] = \begin{cases} \dfrac{1-e^{-\lambda p}}{1-e^{-\lambda}},&\lambda\ne0,\[1.2ex] p,&\lambda=0. \end{cases} \]

Theorem 3 supplies the finite formula
\[ \phi_k=\frac{1-r^{-k}}{1-r^{-N}}, \]
and the displayed continuum problem follows by taking \(k/N\to p\) and \(r_N=1+\lambda/N\). A solution is therefore an approximation certificate for the limiting fixation probability. This is expected to be Class A: the answer has a closed form and can be approximated in polynomial time in the input and accuracy parameters.

This is a recognizable mirror of the authors’ question. It preserves their context-dependent adoption rule, their fixation objective, their complete-graph/regular-population setting, and their interpretation of \(r=\alpha_{01}/\alpha_{10}\) as a fitness-like bias. It does not introduce campaigning, control, or an optimization objective absent from the paper. The action is still stochastic opinion adoption; the computational task is evaluating its population-level outcome.

The regime is plausible for a platform, population-genetic system, or large online community with two standardized stances and approximately homogeneous communication. There are \(N\) agents but only two opinion types, so the high-multiplicity interpretation is especially strong. Rational \(p\) corresponds exactly to repeated finite voter types after denominator clearing.

The main weakness is that the nontrivial limit requires bias to shrink with \(N\). With a fixed \(r\neq1\), Theorem 3’s formula collapses in the limit: essentially any positive initial mass of the favoured opinion fixates with probability \(1\), while the disadvantaged opinion has probability \(0\). Thus \(\mathrm{WBCF}_\infty\) is an extension using a population-scaled parameter, not a literal fixed-parameter limit. I think the extension is defensible because the paper itself connects the model to population genetics, but it is the point the opposing case should attack.

A second, more direct mirror uses Theorem 1, “If \(\alpha_{01}=\alpha_{10}=\alpha\), then \(T_k(n)=\frac{1}{\alpha}n^2h(k/n)+O(n/\alpha)\).” Define \(\mathrm{CCT}_\infty\) as follows. The society again has two interchangeable opinion types with mass \(p\) for opinion \(1\), agents communicate on a clique, and updates are asynchronous and unbiased with adoption probability \(\alpha\). Let \(T_{k_N}(N)\) be the finite expected consensus time from \(k_N=\lfloor pN\rfloor\) agents holding opinion \(1\).

Given rational \(p\), rational \(\alpha>0\), and \(\varepsilon>0\), output a rational \(z\) satisfying
\[ \left|z-\Theta(p,\alpha)\right|\le\varepsilon, \]
where
\[ \Theta(p,\alpha) = \lim_{N\to\infty} \frac{\mathbb E[T_{k_N}(N)]}{N^2} = \frac{-p\ln p-(1-p)\ln(1-p)}{\alpha}. \]

The continuous object is \(p\), not time: the finite process still has discrete update rounds, and \(N^{-2}\) is simply the normalization identified by Theorem 1. This is a particularly clean Class A mirror. The theorem’s \(O(N/\alpha)\) error vanishes after normalization, while the finite rational-clone instances are recovered exactly whenever \(p\) is rational and \(N\) is a suitable multiple of its denominator.

This mirror is plausible for a large, repeatedly sampled population with only two behavioural types and homogeneous communication. It covers the paper’s expected-consensus-time question, and Proposition 1 reinforces it by showing that unbiased laziness merely scales the time by \(1/\alpha\). Further questions include the finite-\(N\) correction term, continuous limits for more opinions, and finite-type block-network analogues.

I would not claim that these mirrors cover the paper’s general-topology results, the synchronous biased case, or Theorem 5’s exact finite-population consensus time without additional work. Exact consensus is itself delicate in an atomless population: a deterministic mean-field trajectory may approach consensus without ever reaching it. Threshold consensus would be a sensible extension, but it should not be passed off as a direct mirror of the paper’s stopping-time definition.

There is also a formal source-gate caveat. The paper contains no named theorem asserting NP-hardness, membership in \(P\), W[1]-hardness, or FPT status. Its named results are exact formulas, bounds, and probabilistic identities. Therefore, under a strictly literal requirement that an anchor must be a classical complexity classification, this paper has no qualifying computational anchor. Under the broader reading that an explicit exact or approximation-evaluation theorem counts as an algorithmic result, Theorem 3 and Theorem 1 support the two Class A mirrors above. The positive claim should be made in precisely that limited form.