arXiv:2609.39046
Fixed-Point Bifurcations and Path-Information Bounds in Nonlinear Exceptional-Point Sensing
Xue-Hao Yu, Cong-Feng Qiao
Abstract
Nonlinear exceptional points (NEPs) were proposed to circumvent the noise--responsivity cancellation that limits linear exceptional-point sensors. At an NEP, the stationary frequency response can be singular while the Hamiltonian Petermann factor remains finite, suggesting a genuine signal-to-noise ratio (SNR) enhancement. Yet recent model-specific analyses of nonlinear fluctuations have reached conflicting conclusions. We prove that any singular stationary frequency response requires a fixed-point bifurcation of the nonlinear dynamics. In local fluctuation regimes, the singularities in responsivity and long-time frequency uncertainty are both set by the critical relaxation rate, leading to exact cancellation in the SNR and precluding any divergent enhancement. Furthermore, we derive a path-information bound that applies across both local and nonlinear fluctuation regimes and places a finite ceiling on the long-time SNR enhancement. Notably, this upper bound is attainable in the local fluctuation regime, implying that frequency readout under such parameter settings is information-optimal.
AI Summary
Can nonlinear exceptional-point sensors evade the noise–responsivity cancellation of linear EP sensing? This paper connects singular stationary frequency response to bifurcations and bounds the information obtainable from frequency readout. For globally $\mathrm{U}(1)$-symmetric self-oscillators described by classical Langevin equations, the representation $R=\alpha\alpha^\dagger$ removes the global phase and maps rotating solutions to fixed points. Differentiating the fixed-point condition shows that singular response requires a singular dynamical Jacobian. For one real critical mode with relaxation rate $\lambda_c\to0$, center-manifold reduction gives $\mu_\omega\sim-g_cF_c/\lambda_c$. Local Ornstein–Uhlenbeck fluctuations yield long-time frequency-noise intensity $N_A=g_c^2\Gamma_c/|\lambda_c|^2$, so the critical amplification cancels: $\mathrm{SNR}/(\sqrt{T}|\Delta\varepsilon|)\to|F_c|/\sqrt{\Gamma_c}$. Beyond linearized fluctuations, a one-dimensional stochastic reduction and Girsanov’s theorem give $|\partial_\varepsilon\bar\omega|^2/N_A\leq\langle|\partial_\varepsilon f|^2/q\rangle$. For additive parameter forcing and constant diffusion, this ceiling is $|F_c|^2/\Gamma_c$, attained by time-averaged frequency readout in the local regime. Three-mode Kerr-sensor simulations test cubic-response scaling, $|\Delta\omega|\propto|\varepsilon|^{1/3}$ and $N_A\propto|\varepsilon|^{-4/3}$, and the nonlinear crossover. The bounds concern local parameter estimation and asymptotically long observation times, assuming stationary ergodic dynamics and parameter dependence through the drift. The analysis focuses on a single critical mode; multiple critical modes, nonstationary attractors, scattering measurements, and detector noise require extensions. Finite practical enhancements and finite-time benefits are not excluded.
Central claim may not hold: The frequency-information bound applies to parameter-independent statistics of the reduced trajectory, whereas the paper's frequency functional explicitly depends on the parameter and the eliminated global phase can carry information absent from that trajectory. The deterministic local center-manifold argument does not establish the exact one-dimensional stochastic reduction needed to extend the claimed bound to general nonlinear fluctuation regimes. The noisy CAM fixed-point interpretation conflates rank-one trajectory matrices with a potentially full-rank ensemble matrix whose nonlinear dynamics generally require an unestablished moment closure.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 7 pages, 3 figures