Borda Aggregation Dynamics of Preference Orderings on Networks

Abstract: We introduce and analyze a discrete-time network process in which each node holds a (weak) preference ordering over a finite set of alternatives and updates by local Borda aggregation. At each step, a node forms a weighted average (row-stochastic random-walk normalization) of its neighbors' Borda score vectors and projects the aggregated score back to a weak order. Updates are bounded: in each round, a node advances by at most one step along a shortest path in the fixed graph of preference orderings, following the direction prescribed by its neighbors' Borda-aggregated preferences. Our emphasis is dynamical: we develop sufficient conditions, stated directly in terms of graph topology, weights, and the bounded step rule, for (i) self-sustained oscillations in the absence of persistent sources, and (ii) forced oscillations under contrarian persistent camps. We also record robustness (structural stability) away from score-tie hyperplanes and contrast synchronous (Variant S) and asynchronous (Variant A) updating.

• The paper introduces a dynamical systems framework that employs bounded-step Borda aggregation to evolve weak ordinal preference orderings.
• It rigorously characterizes the emergence of fixed points, periodic orbits, and oscillatory behaviors under both synchronous and asynchronous updating.
• The analysis reveals how network topology, spectral properties, and discrete preference geometry jointly govern consensus, polarization, and cycling dynamics.

Borda Aggregation Dynamics of Preference Orderings on Networks

Introduction and Motivation

This paper establishes a rigorous dynamical systems framework for the evolution of (possibly weak) ordinal preference orderings on directed, weighted social networks, where local aggregation is performed via Borda scores. Each agent holds a weak order over a finite set of alternatives. Temporal evolution proceeds through synchronous or asynchronous updates, in which agents aggregate neighbors' Borda scores using a random-walk-normalized influence matrix and project these back to the discrete space of weak orders via a bounded-step update rule. The resulting state space is the product $\Omega(m)^{V}$, where $\Omega(m)$ denotes the lattice of all weak orders on $m$ alternatives.

Unlike classical opinion dynamics—where agent states are scalar or vectorial and evolution is linear—the aggregation and update mechanisms here are both nonlinear and combinatorially structured. The use of the move graph $H$ (usually, the cover graph of the weak-order lattice) to define elementary transitions between weak orders, paired with the restriction that agents can only move one edge per update, endogenizes behavioral friction and cognitive/institutional boundedness.

The work positions itself in contrast to static social-choice aggregation, emphasizing preference dynamics as an emergent, network-mediated process whose equilibria and cycles are dictated by both the topology of the influence graph $G$ and the combinatorial geometry of preference states.

Mathematical Model

Given a finite, directed influence graph $G$ with row-stochastic weight matrix $W$, each agent's local state is a weak order from $\Omega(m)$. A central mechanism is the Borda score mapping, which transforms a weak order into a vector in $\mathbb{R}^m$, always using the standardized convention that indifferences are handled via average Borda scoring.

At every synchronous update (Variant S), non-persistent nodes perform:

1. Local Borda aggregation: Each node forms the convex combination of its in-neighbors' Borda score vectors according to $W$.
2. Projection: The resulting vector is projected back to $\Omega(m)$0 by mapping strictly decreasing scores to strict orderings and ties to indifference classes.
3. Bounded-step update: Instead of instantaneously jumping to the target ordering, nodes transition one edge along a shortest path in move graph $\Omega(m)$1 toward their Borda-aggregated target, encoding incremental adjustment.

The model supports persistent (stubborn) nodes, whose preference ordering remains fixed. Asynchronous updating (Variant A) allows for only one node to update per time step, under either deterministic or random schedules.

Fixed Points and Generic Finite-State Dynamics

Due to the finite state space and deterministic dynamics (with fixed tie-breaking), every trajectory is eventually periodic; that is, after some transient, it enters a periodic orbit of some period $\Omega(m)$2. Fixed points correspond to static preference profiles (consensus or pluralist equilibria). Sufficient conditions for consensus are naturally realized when $\Omega(m)$3 is strongly connected and the initial configuration is homogeneous, but the model admits non-consensus, multistable fixed points as well due to the possible occurrence of ties and the bounded step rule.

The paper rigorously shows that, under the model assumptions, the existence, multiplicity, and network dependence of fixed points and periodic orbits can be fully characterized by the interplay of the influence topology, weights, and the geometry of $\Omega(m)$4. Importantly, the system's asymptotic behavior (convergence, cycling, multistability) is nontrivially controlled by these elements.

Mechanisms for Oscillatory Dynamics

Self-Oscillations without Persistence

The existence of directed cycles in $\Omega(m)$5 and cycles in the move graph $\Omega(m)$6 induces endogenous oscillatory behavior, even in the absence of persistent nodes. The authors formalize this by constructing traveling wave solutions: on a directed cycle of $\Omega(m)$7 nodes (each influenced only by its predecessor), initializing the system on a $\Omega(m)$8-cycle in $\Omega(m)$9 yields a $m$0-periodic orbit. This is a direct consequence of the combinatorial structure of the update map and is formalized in Theorem 4.1. The mechanism depends crucially on the bounded-step constraint: agents are prohibited from jumping directly to their local Borda-optimal ranking, resulting in persistent non-convergent trajectories.

Forced Oscillations via Contrarian Persistent Camps

The presence of persistent nodes, partitioned into antipodal camps (i.e., pinned to orderings $m$1 and $m$2 under the weak-order antipode involution), catalyzes forced oscillations in reachable free nodes. When the network's free-node subgraph exhibits bipartite or periodic (imprimitive) structure in the random-walk kernel, the model systematically constructs even-period oscillations through what the authors term parity or spectral mechanisms.

If the corresponding $m$3 kernel has period two (e.g., for undirected bipartite graphs or directed structures with alternating influence), then aggregation of Borda scores under contrarian camps results in alternating target signals, which—combined with the bounded-step rule—enforces deterministic period-2 (or, by lifting, $m$4) cycles in node states. The underlying mechanism is connected to the $m$5 eigenmode of bipartite random walks, and is formalized in Theorem 5.2. These results are not mere artifacts of synchronous updating but arise directly from the network’s spectral and topological features.

The minimality and reachability of forcing gadgets (such as a two-node oscillator under appropriate weighted connections to antipodal persistent nodes) are explicitly analyzed.

Robustness and Structural Stability

A critical contribution is the establishment of structural stability of the system away from Borda score tie-hyperplanes. When aggregated scores are separated by a finite margin from tie boundaries, periodic symbolic dynamics (sequences of targets and moves) persist under arbitrarily small perturbations to the influence weights. This result implies that qualitative features of cyclic or convergent behaviors are insensitive to small network perturbations, contingent on the preference profiles maintaining distance from scoring degeneracies. The boundary layer of tie-hyperplanes delineates regions of topological change in the discrete dynamics.

Synchronous vs Asynchronous Updating

The behavior of the model under asynchronous (single-node) updates (Variant A) diverges sharply from the synchronous case. The spectral parity mechanisms responsible for even-period cycles are typically destroyed, as two-step decoupling across bipartite partitions no longer holds. Instead, under mild additional conditions—such as the existence of strict Lyapunov functions—almost-sure convergence to fixed points prevails. However, the paper notes that metastable cycling on recurrent classes (especially on tie boundaries) may still arise in specific configurations.

Restriction to Structured Preference Domains

The framework extends to specialized preference domains such as single-peaked orderings. When updates and targets are restricted to remain single-peaked with respect to a fixed axis, the dynamics are invariant within that domain. The connectivity of the move graph is reduced, and some mechanisms for oscillatory behavior are eliminated, but the essential structure of bounded, network-mediated ordinal adjustment persists.

Implications and Theoretical Significance

The model synthesizes classical social influence processes, ordinal preference geometry, and bounded rationality. It yields explicit characterizations of when and how consensus arises, when persistent polarization is supported dynamically, and when cycling (both endogenous and forced) is guaranteed. The findings highlight that neither consensus nor persistent oscillation is a generic consequence of networked aggregation, but rather emerges from structural properties of both $m$6 and the move graph $m$7.

Potential applications include deliberative policymaking, committee procedures, and online social platforms. The model's generality supports extensions to adaptive or co-evolving networks, noise, alternative bounded update schemes, and empirically validated network structures.

Conclusion

This work provides a mathematically rigorous and general framework for understanding the dynamical evolution of ordinal preferences on social networks via bounded Borda aggregation. It demonstrates that a rich spectrum of behaviors—including robust periodic orbits, metastability, and consensus—arises from the interplay between network topology, Borda-based aggregation, and the discrete geometry of orderings. The model's tractability and clarity make it a strong candidate foundation for future research in networked social choice, opinion dynamics, and collective decision theory.