Phase Transitions Without Instability: A Universal Mechanism from Non-Normal Dynamics

Abstract

We identify a new universality class of phase transitions that arises in non-normal systems, challenging the classical view that transitions require eigenvalue instabilities. In traditional bifurcation theory, critical phenomena emerge when spectral stability is lost; here, we show that transitions can occur even when all equilibria are spectrally stable. The key mechanism is the transient amplification induced by non-orthogonal eigenvectors: noise-driven dynamics are enhanced not by lowering energy barriers, but by increasing the effective shear of the flow, which renormalizes fluctuations and acts as an emergent temperature. Once the non-normality index $κ$ exceeds a critical threshold $κ_c$, stable equilibria lose practical relevance, enabling escapes and abrupt transitions despite preserved spectral stability. This pseudo-criticality generalizes Kramers' escape beyond potential barriers, providing a fundamentally new route to critical phenomena. Its implications are broad: in biology, DNA methylation dynamics reconcile long-term epigenetic memory with rapid stochastic switching; in climate, ecology, finance, and engineered networks, abrupt tipping points can arise from the same mechanism. By demonstrating that phase transitions can emerge from non-normal amplification rather than eigenvalue instabilities, we introduce a predictive, compact framework for sudden transitions in complex systems, establishing non-normality as a defining principle of a new universality class of phase transitions.

Figures (6)

• Figure 1: Comparison between theoretical predictions and numerical simulations of the dynamics \ref{['eq:numerical']}. (Top) Escape rate $\Gamma$ as a function of $\kappa/\kappa_c$. black dots: numerical simulations of the full two-dimensional dynamics. Solid line: theoretical prediction from the renormalized Kramers formula $\Gamma \sim \exp\!\left(-\Delta U_\text{eff}/\delta_\text{eff}\right)$, with $U_\text{eff}$ and $\delta_\text{eff}$ defined in \ref{['eq:eff_dyn']}. (Bottom) Standard deviation of the reaction variable $x$ as a function of $\kappa/\kappa_c$. For $\kappa < \kappa_c$, the variance matches that of an Ornstein--Uhlenbeck process localized near a stable equilibrium. For $\kappa > \kappa_c$, the system transits between $x=\pm 1$ with equal probability, leading to $\sqrt{\mathrm{Var}(x)} \approx 1$. We used $\omega=1$, $\delta=0.01$, and $\kappa_c=10$, and simulations are performed for $\kappa$ ranging from $\kappa_c/10$ to $5\kappa_c$. Each simulation runs over a total time $T=10^{5}$ with integration step $\Delta t=0.1$, corresponding to $N=10^{6}$ data points. The escape rate is estimated as $\Gamma = 1/\langle\tau\rangle$, where $\langle\tau\rangle$ is the mean first-passage time from $x>0$ to $x<0$ (or vice versa).
• Figure 2: Simulation of a nonlinear two-dimensional system described in Section \ref{['sec:apx_numerical']}, with parameters $\omega=1$, $\delta=0.01$, and $\kappa_c=10$. The left panels correspond to $\kappa=\kappa_c/10$, while the right panels correspond to $\kappa=5\kappa_c$. Top panels: dynamics in phase space, where the horizontal axis denotes the reaction coordinate ($x$) and the vertical axis the non-normal mode ($y$). Red dots mark the stable equilibria $(x,y)=(\pm 1,1)$, blue arrows indicate the force vector field, and the dashed red line at $x=0$ denotes the unstable manifold along the reaction direction. Bottom panels: time series of the reaction variable ($x$). Continuous red lines mark the stable equilibria at $x=\pm 1$, and the dashed red line marks the unstable equilibrium at $x=0$. All simulations are performed over a time horizon $T=500$ with integration step $\Delta t=0.1$, corresponding to $N=5000$ time steps.
• Figure 3: Escape rate (left panel) and standard deviation (right panel) of the reaction variable ($x$) for the dynamics described in Section \ref{['sec:apx_numerical']}, with parameters $\omega=1$, $\delta=0.01$, and $\kappa_c=10$, as a function of $\kappa$. Simulations are performed for $\kappa$ ranging from $\kappa_c/10$ to $5\kappa_c$. Each simulation runs over a total time $T=10^{5}$ with integration step $\Delta t=0.1$, corresponding to $N=10^{6}$ data points. The escape rate is estimated as $\Gamma = 1/\langle\tau\rangle$, where $\langle\tau\rangle$ is the mean first-passage time from $x>0$ to $x<0$ (or vice versa). Black dots denote numerical measurements, and the red curve shows the theoretical escape rate given by Eq. \ref{['eq:escape_apx']}. The vertical dashed blue line marks the critical value $\kappa=\kappa_c$. In the right panel, the lower horizontal blue line indicates the theoretical standard deviation of an Ornstein-Uhlenbeck approximation near equilibrium, $\sqrt{\delta/\omega}=0.1$, while the upper blue line ($=1$) corresponds to the variance of a process equally likely to be near $x=\pm 1$.
• Figure 4: Simulation of a nonlinear two-dimensional system described in the $(z_1,z_3)$ by \ref{['eq:apx_z_dyn_nn']}, with parameters $z_{3,0}=-0.2$, $z_{3,+}=0.6$, $z_{3,-}=-0.6$$\omega_1=\omega_3=1$, $\delta=0.001$, and $\kappa_c=10$. The left panels correspond to $\kappa=\kappa_c/10$, while the right panels correspond to $\kappa=2\kappa_c$. Top panels: dynamics in phase space $(y_1,y_3)$\ref{['eq:apx_z_to_y']}, where the horizontal axis denotes the ratio of methylated site ($y_3$) and the vertical axis the ratio of unmethylated site ($y_1$). Red dots mark the stable equilibria, blue arrows indicate the force vector field, and the dashed red line the axis $z_1$ and the blue dashed line the axis $z_3$, which crosses each other at the unstable equilibrium. Bottom panels: time series of the reaction variable ($z_3$). Continuous red lines mark the stable equilibria at $z_{3,\pm}=\pm 0.6$, and the dashed red line marks the unstable equilibrium at $z_{3,0}=-0.2$. All simulations are performed over a time horizon $T=500$ with integration step $\Delta t=0.1$, corresponding to $N=5000$ time steps.
• Figure 5: Escape rate of the reaction variable $z_3$ as a function of $\kappa/\kappa_c$. Plain dots ("$\cdot$") denote transitions from $z_3>z_{3,0}$ to $z_3<z_{3,0}$, while crosses ("$+$") denote the reverse transitions. The dynamics follow \ref{['eq:apx_z_dyn_nn']} with parameters $z_{3,0}=-0.2$, $z_{3,+}=0.6$, $z_{3,-}=-0.6$, $\omega_1=\omega_3=1$, $\delta=0.001$, and $\kappa_c=10$. Simulations are performed for $\kappa$ ranging from $\kappa_c/10$ to $10\kappa_c$, each run covering a total time $T=10^{5}$ with integration step $\Delta t=0.1$, corresponding to $N=10^{5}$ sampled points. The escape rate is computed as $\Gamma = 1/\langle\tau\rangle$, where $\langle\tau\rangle$ is the mean first-passage time across the unstable saddle $z_{3,0}$.