Source
Hansen and Ghrist (2020; SIAM J. Appl. Math. 81(5), 2021), LIT-1646, §§2–4 (pp. 3–10 of arXiv 2005.12798v1); read in NOTE-1222.
What was actually shown
The structure. Each agent v has an opinion space F(v), and each edge e has a discourse space F(e) whose basis topics “are not necessarily the same as any of the basis topics” of either agent. The linear maps F_{v⊴e}: F(v) → F(e) are expressions of opinion. The agents “have expressed consensus when F_{u⊴e}(x_u) = F_{v⊴e}(x_v) … this does not imply that u and v have the same opinions: it means that their expressions of personally held opinions have the appearance of agreement” (§3, p. 7). Expression can be selective, combining, scaled, or sign-reversed (“one can model agents who lie”), and it can differ by edge (p. 8).
The theorem. With L_F = δᵀδ, solutions of dx/dt = −αL_F x converge, as t → ∞, to the orthogonal projection of x(0) onto H⁰(G; F) = ker δ = ker L_F (Theorem 4.1, p. 9). L_F is symmetric positive semidefinite, so the flow is exp(−tαL_F), which tends to the projection onto its kernel. Convergence is exponential. This could have come out otherwise only if L_F had negative eigenvalues, which is what happens in §12’s signed variant (Proposition 12.1).
The two consequences the paper draws. “If this sheaf has no nontrivial global sections, the only stable opinions will be everywhere zero: an uninteresting solution” (p. 9). And on a sheaf with sign-reversed maps, stable polarized distributions are global sections. In Example 4.2 (p. 9, Fig. 4) two agents who dislike a politician “tell a polite lie to their neighbors on the left but are frank with each other”, and “any initial opinion distribution will converge to one of these polarized distributions”. On the constant sheaf the same flow gives ordinary consensus at the average.
What this does not say
- That an agreement in expression reveals the private opinions. A global section fixes only the images F_{v⊴e}x_v. Agents “do not know the veracity of their neighbors’ expressed opinions” (p. 9), and the model gives them no way to find out.
- That people form opinions this way. The paper has no data, and the authors say evolving opinions “is perhaps not how real people engage in discourse” (p. 8).
- That the expression maps are given by anything in particular. In §§3–7 they are fixed background. How they change is a separate dynamic, in THEORY-1088.
- That the limit is reached from any start to any section. It is the projection of the starting opinions, so the starting point decides which section is reached. Steering to a chosen section is a control question, which §6 answers with relative cohomology.
- Anything nonlinear. The bounded-confidence and antagonistic variants (§§11–12) have more equilibria, or are unstable.