1: \begin{abstract}

2: We investigate the phenomena of political bi-polarization in a population of interacting agents by means of a generalized version of the model introduced in \cite{Vazquez-2020} for the dynamics of voting intention.  Each agent has a propensity $p$ in $[0,1]$ to vote for one of two political candidates.  In an iteration step, two agents $i$ and $j$ with respective propensities $p_i$ and $p_j$ interact, and then $p_i$ either increases by an amount $h>0$ with a probability that is a nonlinear function of $p_i$ and $p_j$ or decreases by $h$ with the complementary probability.  We study the behavior of the system under variations of a parameter $q \ge 0$ that measures the nonlinearity of the propensity update rule.  We focus on the stability properties of the two distinct stationary states: mono-polarization in which all agents share the same extreme propensity ($0$ or $1$), and bi-polarization where the population is divided into two groups with opposite and extreme propensities.  We find that the bi-polarized state is stable for $q<q_c$, while the mono-polarized state is stable for $q>q_c$, where $q_c$ is a transition value that decreases as $h$ decreases.  We develop a rate equation approach whose stability analysis reveals that $q_c$ vanishes when $h$ becomes infinitesimally small.  This result is supported by the analysis of a transport equation derived in the continuum $h \to 0$ limit.  We also show by Monte Carlo simulations that the mean time $\tau$ to reach mono-polarization in a system of size $N$ scales as $\tau \sim N^{\alpha}$ at $q_c$ , where $\alpha(h)$ is a non-universal exponent. 

3: \end{abstract}

4: