betaFeedback

WorksPhysics and Astronomy

Weak-Signal Amplification through Stochastic Resetting

Abstract

Stochastic resetting (SR) offers a distinct route to amplifying weak periodic signals in nonlinear systems. Here, we demonstrate SR-induced resonance (SRIR) in a system with double-well potential driven by a weak periodic force, where the amplitude of the mean periodic response exhibits a pronounced maximum at an optimal SR rate. The underlying mechanism exploits dynamical instability: SR repeatedly brings the system close to an unstable state, where the phase of the periodic force at each reset biases the subsequent motion toward one of the stable states. The local instability then amplifies this weak bias into a large-scale response. The SRIR arises from a competition between rare resets that limit phase-dependent state selection and frequent resets that prevent the system from fully relaxing to the selected state. We develop a general renewal formulation for the periodic response and a coarse-grained approximation that predicts the optimal resetting rate, revealing the low-frequency scaling ropt(ω)∝ω2/3r_{\rm opt}(ω)\proptoω^{2/3}, in agreement with numerical simulations. SRIR is fundamentally distinct from conventional noise-induced resonance (NIR), which relies on noise-activated interwell transitions whose rates are periodically modulated by the external driving force. In the weak-forcing and slow-driving regime, SRIR can achieve much stronger signal amplification than NIR. SRIR persists for a finite spread of resetting positions, and our framework extends beyond bistability. We further discuss how SRIR and NIR can be selectively controlled. Our work proposes a novel approach to the classical paradigm of stochastic resonance.

main_new.pdf

Introduction

Introduction. — Stochastic resetting (SR) provides a simple way of controlling nonequilibrium dynamics by intermittently returning a system to a prescribed state [1, 2, 3]. Such renewal events can fundamentally reshape nonequilibrium stationary states and transient relaxation [4, 5, 6, 7], modify response and ergodic properties [8, 9, 10, 11], and strongly affect first passage statistics, often generating an optimal finite SR rate [12, 13, 14, 15, 16, 17]. SR has also become experimentally accessible, particularly in colloidal systems controlled by optical tweezers [18, 19, 20], with recent extensions to active particles [21, 22, 23]. The scope of SR extends further to stochastic thermodynamics [24, 25], extreme-value statistics [26], and dynamics on complex networks [27]. Despite these advances, how SR reshapes the response of nonlinear systems remains largely unexplored. This raises the central question of our work: Can SR itself induce resonant amplification of weak periodic signals?

Resonant amplification in nonlinear systems is conventionally associated with stochastic resonance, also referred to as noise-induced resonance (NIR), which arises from the interplay of multistability, noise, and weak periodic signals [28, 29, 30, 31]. In a paradigm bistable system, the periodic forcing is on its own too weak to drive transitions between the two stable states, but it modulates the potential barriers, while noise enables barrier crossings and drives the transitions between the states. NIR emerges when the noise-induced switching time becomes comparable to the periodic driving time, allowing transitions between the two stable states to follow periodic signal. NIR was originally introduced to explain the noise-assisted synchronization of glacial–interglacial climate transitions with weak periodic orbital forcing [32, 33] and later this noise-assisted amplification was demonstrated experimentally in electronic bistable systems [34], ring lasers [35], and tunnel diodes [36], as well as in biological sensory systems [37, 38]. Noise can also optimize coherence in excitable systems without periodic signal, giving rise to coherence resonance [39, 40, 41].

SR provides a fundamentally different route, with reset events representing instantaneous state-reinitialization processes, ranging from abrupt climate transitions near tipping points [42], voltage resets in bistable circuits, neuronal spike-and-reset dynamics [43] to price dynamics of barrier-type options [44, 45]. SR repeatedly restores the system near a dynamically susceptible (typically unstable) state, where the phase of the periodic drive at each reset determines the instantaneous directional bias and favors relaxation toward one of the stable states. The local instability then amplifies this weak phase-dependent bias into a large-scale response, see Figure 1.

Figure 1. Schematic of SR-assisted amplification near an unstable state in a bistable system driven by weak periodic force. (a) Reset to instability: Poisson SR repeatedly returns the system to the unstable state xr=0x_r = 0, where the instantaneous periodic bias favors relaxation toward either +xs+x_s or −xs-x_s. (b) Weak periodic bias: Near the unstable state, the periodic drive breaks the local symmetry through δx˙≈λuδx+Acos⁡ϕ\delta\dot{x} \approx\lambda_u\delta x + A\cos\phi, while the instability λu>0\lambda_u > 0 amplifies the resulting displacement. (c) Phase-dependent selection: The phase at resetting biases the relaxation direction to the stable states, with cos⁡ϕ>0\cos\phi> 0 favoring +xs+x_s and cos⁡ϕ<0\cos\phi< 0 favoring −xs-x_s.

In this Letter, we establish SR as a distinct route to resonant signal amplification in periodically driven systems, opening a new avenue for controlling response of nonlinear dynamics beyond conventional NIR. We demonstrate SR-induced resonance (SRIR) in a bistable system and uncover the physical picture: At small SR rates, rare resets provide too few opportunities for phase-dependent relaxation, whereas at large rates, frequent resets interrupt relaxation before the system can reach a selected stable state. This competition produces a pronounced maximum in the response amplitude at an intermediate SR rate. We derive an exact renewal formulation for the mean response under Poisson SR and obtain the corresponding optimality condition, while a coarse-grained approximation that predicts a robust scaling law for the optimal resetting rate. Comparison with conventional NIR further shows that SRIR can achieve substantially stronger amplification of weak periodic signals over a broad range of driving frequencies.

Model.— We consider an overdamped particle in a double-well potential subject to periodic driving,

dx(t)dt=ax(t)−bx3(t)+Acos⁡(ωt).(1)\frac{dx(t)}{dt} = ax(t) - bx^3(t) + A\cos(\omega t). \tag*{(1)}

In addition, the particle undergoes Poisson SR at rate rr. We denote the successive SR times by tnt_n, with n≥1n \ge1 and t0=0t_0 = 0, and define the waiting time between two successive SR events as τn=tn−tn−1\tau_n = t_n - t_{n-1}. The waiting times τn\tau_n are independent and identically distributed according to the exponential density ψ(τ)=re−rτ\psi(\tau) = re^{-r\tau}. At each SR time tnt_n, the particle is instantaneously returned to the reset point x(tn)=xr=0x(t_n) = x_r = 0, while the phase of the periodic driving remains unchanged. In the absence of forcing, the potential U(x)=−ax2/2+bx4/4U(x) = -ax^2/2 + bx^4/4 with a,b>0a,b > 0, possesses two stable states at x=±xsx = \pm x_s, with xs=a/bx_s = \sqrt{a/b}, separated by an unstable state xu=0x_u = 0. We restrict our analysis to the subthreshold regime, A<Ac=2a3/2/(33b)A < A_c = 2a^{3/2}/(3\sqrt{3b}), for which the periodic forcing alone cannot overcome the potential barrier [46].

Renewal formulation

— Let xϕ(τ)x_\phi(\tau) denote the uninterrupted deterministic trajectory following a reset that occurs at phase ϕ\phi. More precisely, if a reset occurs at time tnt_n, we define ϕ=ωtn(mod2π)\phi= \omega t_n \pmod{2\pi} and introduce the elapsed time τ=t−tn\tau= t - t_n measured from that reset event. The trajectory xϕ(τ)x_\phi(\tau) then evolves from the initial condition xϕ(0)=xr=0x_\phi(0) = x_r = 0 according to the deterministic dynamics

dxϕ(τ)dτ=axϕ(τ)−bxϕ3(τ)+Acos⁡(ϕ+ωτ),(2)\frac{dx_\phi(\tau)}{d\tau} = ax_\phi(\tau) - bx_\phi^3(\tau) + A\cos(\phi+ \omega\tau), \tag*{(2)}

provided no further reset occurs during the interval of duration τ\tau. We note that Eq. (1) describes the full dynamics, including a sequence of SR events, whereas Eq. (2) describes a single uninterrupted deterministic post-reset trajectory, parameterized by the reset phase ϕ\phi and the elapsed time τ\tau. Representative post-reset trajectories generated by Eq. (2) are shown in Figure A1(a). Near the unstable state xu=0x_u = 0 and for sufficiently slow driving, the post-reset dynamics Eq. (2) can be linearized as dxϕ/dτ∼αxϕ+Acos⁡ϕdx_\phi/d\tau\sim\alpha x_\phi+ A\cos\phi. This gives xϕ(τ)∼Aa(eaτ−1)cos⁡ϕx_\phi(\tau) \sim\frac{A}{a}(e^{a\tau} - 1)\cos\phi. The sign of the instantaneous periodic bias at the reset time therefore defines the initial direction of motion, while the intrinsic instability of x=0x = 0 exponentially amplifies this small phase-dependent perturbation, driving the system toward a selected stable state.

Figure A1. Phase-dependent reset-to-well dynamics underlying the coarse-grained reduction. (a) Representative exact post-reset trajectories xϕ(τ)x_\phi(\tau) for reset phases ϕ\phi, obtained from Eq. (2). Each trajectory starts from xϕ(0)=0x_\phi(0)=0 and subsequently follows the uninterrupted deterministic dynamics. Open symbols mark the first arrival at either +xs+x_s or −xs-x_s, thereby defining the reset-to-well time τs(ϕ)\tau_s(\phi) and the destination variable σϕ=±1\sigma_\phi=\pm1 used in the coarse-grained reduction (10). (b) Reset-to-well time τs(ϕ)\tau_s(\phi) as a function of the reset phase ϕ\phi, with the destination variable σϕ\sigma_\phi encoded by color: blue (σϕ=+1\sigma_\phi=+1) and red (σϕ=−1\sigma_\phi=-1) correspond to first arrival at +xs+x_s and −xs-x_s, respectively. Thus, panel (b) displays the two phase-dependent quantities {τs(ϕ),σϕ}\{\tau_s(\phi),\sigma_\phi\} entering the coarse-grained reduction description (10). Parameters are A=0.3A=0.3, ω=0.1\omega=0.1, and xs=1x_s=1.

As the most recent reset occurred at time tn=t−τt_n = t - \tau, the phase of the periodic driving at that reset event is ϕ=ω(t−τ)\phi= \omega(t-\tau), where the initial phase of the driving at t=0t = 0 is set to ϕ0=0\phi_0 = 0 [47]. The ensemble-averaged position then obeys the renewal relation [1, 14]

⟨x(t)⟩=e−rtxϕ0(t)+∫0tre−rτxω(t−τ)(τ) dτ.(3)\langle x(t)\rangle= e^{-rt}x_{\phi_0}(t) + \int_0^t re^{-r\tau}x_{\omega(t-\tau)}(\tau)\,d\tau. \tag*{(3)}

Here, the first term represents the trajectories that experience no reset before time tt, whereas the integral accounts for trajectories whose most recent reset occurred at time tn=t−τt_n = t - \tau. Specifically, xϕ0(t)x_{\phi_0}(t) denotes the deterministic trajectory evolving from the initial position with driving phase ϕ0\phi_0, while xω(t−τ)(τ)x_{\omega(t-\tau)}(\tau) denotes the deterministic trajectory evolving for a duration τ\tau from the resetting position, with driving phase ω(t−τ)\omega(t-\tau) at the reset time. The external driving phase remains unaffected by resetting. Accordingly, re−rτdτre^{-r\tau}d\tau is the probability that the most recent reset occurred within the interval (t−τ−dτ,t−τ)(t-\tau-d\tau,t-\tau) and that no subsequent reset occurred prior to time tt.

In the long-time regime, t≫r−1t \gg r^{-1}, the no-reset contribution in Eq. (3) is exponentially suppressed. As the deterministic post-reset trajectories remain bounded, the upper integration limit can also be extended to infinity, yielding ⟨x(t)⟩=∫0∞re−rτxω(t−τ)(τ) dτ\langle x(t)\rangle= \int_0^\infty re^{-r\tau}x_{\omega(t-\tau)}(\tau)\,d\tau, where the phase argument is understood modulo 2π2\pi. Expanding the 2π2\pi-periodic post-reset trajectory as xϕ(τ)=x_\phi(\tau) = ∑nhn(τ)einϕ\sum_n h_n(\tau)e^{in\phi}, with the Fourier coefficients hn(τ)h_n(\tau), gives

⟨x(t)⟩=∑n=−∞∞Cn(r)einωt,(4)\langle x(t)\rangle= \sum_{n=-\infty}^{\infty} C_n(r)e^{in\omega t}, \tag*{(4)}
Cn(r)=r∫0∞e−(r+inω)τhn(τ) dτ,(5)C_n(r) = r\int_0^\infty e^{-(r+in\omega)\tau}h_n(\tau)\,d\tau, \tag*{(5)}
hn(τ)=(2π)−1∫02πxϕ(τ)e−inϕ dϕ.(6)h_n(\tau) = (2\pi)^{-1}\int_0^{2\pi}x_\phi(\tau)e^{-in\phi}\,d\phi. \tag*{(6)}

The fundamental harmonic n=1n=1 of the mean response has the amplitude 2∣C1∣2|C_1|. We therefore define the amplification factor with respect to the driving-force magnitude AA as

G(r)=2∣C1(r)∣A=rπA∣∫02πdϕ∫0∞dτ xϕ(τ)e−rτe−i(ϕ+ωτ)∣.(7)G(r) = \frac{2|C_1(r)|}{A} = \frac{r}{\pi A}\left|\int_0^{2\pi}d\phi\int_0^\infty d\tau\,x_\phi(\tau)e^{-r\tau}e^{-i(\phi+\omega\tau)}\right|. \tag*{(7)}

This exact representation retains the complete phase-dependent post-reset dynamics of Eq. (2) without coarse graining. More generally, the same renewal structure applies to periodically driven nonlinear systems under Poisson SR, with the model-specific dynamics entering through the uninterrupted post-reset harmonics hn(τ)h_n(\tau), see Supplemental Material [48].

At the maximum of G(r)G(r), differentiation of Eq. (7) with respect to rr gives the optimality condition

ropt×τeff(ropt,ω)=1,(8)r_{\mathrm{opt}}\times\tau_{\mathrm{eff}}(r_{\mathrm{opt}},\omega)=1, \tag*{(8)}

where

τeff(r,ω)=Re⁡[∫0∞τe−(r+iω)τh1(τ) dτ∫0∞e−(r+iω)τh1(τ) dτ].(9)\tau_{\mathrm{eff}}(r,\omega)=\operatorname{Re}\left[\frac{\int_0^\infty\tau e^{-(r+i\omega)\tau}h_1(\tau)\,d\tau}{\int_0^\infty e^{-(r+i\omega)\tau}h_1(\tau)\,d\tau}\right]. \tag*{(9)}

Equation (6) gives the exact optimality condition for the resonant amplification, relating the optimal SR rate to a weighted post-reset time scale. However, this condition does not necessarily imply the existence of a nonzero optimal SR rate in general periodically driven systems; an interior maximum requires additional structure in the system’s response. A counterexample is a particle confined in a harmonic potential, U(x)=kx2/2U(x)=kx^2/2, which possesses a single stable equilibrium and exhibits maximum amplification at the boundary ropt=0r_{\mathrm{opt}}=0, see Eqs. (S24)–(S27) of Supplemental Material [48].

To verify the exact renewal theory for the amplification factor (5), we provide direct Monte Carlo simulations. Both approaches reveal a nonmonotonic dependence of the amplification on the SR rate in system (1), with excellent quantitative agreement [Fig. 2(a)]. Near the optimal rate, the ensemble-averaged response exhibits its largest first-harmonic amplitude [Fig. 2(c)], although individual trajectories remain stochastic due to a random sequence

SRIR in the bistable system

Figure 2. SRIR in the bistable system (1). (a) Amplification factor G(r)G(r) for A=0.3A=0.3 and ω=0.1\omega=0.1. Symbols denote direct simulations, and the solid line represents the exact renewal prediction (5), yielding the maximum value at SR rate ropt≈0.18r_{\mathrm{opt}}\approx0.18. The arrows mark the SR rates examined in (b)–(d). (b)–(d) Representative individual trajectories x(t)x(t) and ensemble-averaged responses for r=0.01r=0.01, r=ropt≈0.18r=r_{\mathrm{opt}}\approx0.18, and r=1r=1, respectively. The dotted curves are the periodic driving Acos⁡(ωt)A\cos(\omega t). Other parameters are dt=0.01dt=0.01, N=500N=500, and a=b=xs=1a=b=x_s=1 throughout the text.

of resetting times. The two limiting cases reveal the competing requirements underlying the SRIR. For small rr, resets are too rare to sustain phase-dependent selection despite allowing ample post-reset relaxation [Fig. 2(b)]. For too large rr, phase-selection events are frequent, but repeated resets interrupt the full relaxation before a selected stable state is reached [Fig. 2(d)]. The optimal SR rate therefore balances frequent phase selection against sufficient time for relaxation toward the selected stable state. In particular, for r=0r=0, the amplification factor G(r)G(r) approaches the finite value 1/4a2+ω21/\sqrt{4a^2+\omega^2}, as predicted by linear response theory in the weak-driving regime [29]. The SR-induced dynamics of the bistable system (1) are further illustrated by Movie S1 in Supplemental Material [48].

Coarse-grained reduction.—Although the renewal formulation above is exact, the optimal SR rate remains implicit because it depends on phase-dependent post-reset trajectories. To obtain a transparent analytical description and an explicit asymptotic prediction for roptr_{\mathrm{opt}}, we introduce a coarse-grained representation of the post-reset dynamics.

In conventional NIR, a bistable system is commonly described by a two-state reduction, in which the continuous dynamics is coarse grained into occupations of the two wells and noise-activated transitions between them [28, 29]. Such a description is appropriate when the transition time is negligible compared with the residence time in either well. In the present SR problem, however, each reset returns the particle to the unstable state, from which it undergoes a finite deterministic relaxation before reaching a selected stable state. This transient stage cannot be treated as instantaneous, because repeated SR can interrupt the relaxation before the selected response is fully established. Although the relaxation time may remain shorter than the mean reset interval, it generates the leading response loss that increases with the SR rate and therefore enters a balance determining the optimum. We thus retain the finite post-reset relaxation as an additional transient state in this simplified coarse-grained description.

Specifically, we introduce a three-state coarse-grained description in which each exact post-reset trajectory is approximated as

xϕ(τ)∼σϕxsΘ[τ−τs(ϕ)]={0,0≤τ<τs(ϕ)+xs,τ≥τs(ϕ), σϕ=+1−xs,τ≥τs(ϕ), σϕ=−1(10)x_{\phi}(\tau) \sim\sigma_{\phi}x_s\Theta[\tau-\tau_s(\phi)] = \begin{cases} 0, & 0 \le\tau< \tau_s(\phi) \\ +x_s, & \tau\ge\tau_s(\phi),\ \sigma_{\phi}=+1 \\ -x_s, & \tau\ge\tau_s(\phi),\ \sigma_{\phi}=-1 \end{cases} \tag*{(10)}

where Θ(z)=0\Theta(z)=0 for z<0z<0 and Θ(z)=1\Theta(z)=1 for z≥0z\ge0, is the Heaviside step function. τs(ϕ)=inf⁡{τ>0:∣xϕ(τ)∣=xs}\tau_s(\phi)=\inf\{\tau>0: |x_{\phi}(\tau)|=x_s\} is the reset-to-well time, with τs(ϕ)=∞\tau_s(\phi)=\infty if the trajectory never reaches either well. For τs(ϕ)<∞\tau_s(\phi)<\infty, σϕ=sgn⁡[xϕ(τs(ϕ))]=±1\sigma_{\phi}=\operatorname{sgn}[x_{\phi}(\tau_s(\phi))]=\pm1, is the index identifying the destination well. Both quantities are obtained directly from the exact deterministic dynamics in Eq. (2) and are illustrated in Figure A1. Thus, the three coarse-grained states correspond to a transient state 00 and two well states ±xs\pm x_s.

This reduction retains the phase-dependent transit time and potential-well selection, {τs(ϕ),σϕ}\{\tau_s(\phi),\sigma_{\phi}\}, but neglects the continuous position response during transit and the subsequent intrawell motion. Substituting Eq. (10) into the renewal response Eq. (4) yields the approximate amplification factor and corresponding optimality condition, Eqs. (17) and (18), respectively, as derived in End Matter.

For weak forcing and slow driving, the reset-to-well relaxation time and selected-well index have the asymptotic forms

τs(ϕ)∼32aln⁡(axsA∣cos⁡ϕ∣),σϕ∼sgn⁡(cos⁡ϕ),(11)\tau_s(\phi) \sim\frac{3}{2a}\ln\left(\frac{a x_s}{A|\cos\phi|}\right), \qquad\sigma_{\phi}\sim\operatorname{sgn}(\cos\phi), \tag*{(11)}

as derived in Eqs. (S49) and (S53) of the Supplemental Material [48]. Substituting Eq. (11) directly into the reduced amplification factor G(r)G(r), Eq. (17), and expanding in the asymptotic window τˉs≪r−1≪ω−1\bar{\tau}_s\ll r^{-1}\ll\omega^{-1}, gives

G(r)∼4xsπA(1−rτˉs−ω22r2),(12)G(r)\sim\frac{4x_s}{\pi A}\left(1-r\bar{\tau}_s-\frac{\omega^2}{2r^2}\right), \tag*{(12)}

where τˉs=32a[ln⁡(axs2A)+1]\bar{\tau}_s=\frac{3}{2a}\left[\ln\left(\frac{a x_s}{2A}\right)+1\right]. The two corrections in Eq. (12) encode the competing response losses underlying the SRIR. The term ω2/(2r2)\omega^2/(2r^2) results from insufficiently frequent phase selection events and decreases as rr increases, whereas rτˉsr\bar{\tau}_s describes the loss caused by resets that interrupt the finite post-reset relaxation and increases with rr. Maximizing Eq. (12) with respect to rr therefore yields

ropt∼(ω2τˉs)1/3.(13)r_{\mathrm{opt}}\sim\left(\frac{\omega^2}{\bar{\tau}_s}\right)^{1/3}. \tag*{(13)}

This predicts ropt(ω)∝ω2/3r_{\mathrm{opt}}(\omega)\propto\omega^{2/3}, together with a much weaker dependence on the forcing amplitude AA via τˉs(A)\bar{\tau}_s(A).

Figure 3 shows the dependence of the optimal SR rate roptr_{\mathrm{opt}} on the driving frequency ω\omega. The coarse-grained reduction captures the increase of roptr_{\mathrm{opt}} predicted by the exact renewal theory (8), with quantitative deviations appearing at higher frequencies. In the low-frequency regime ω≲τˉs−1≈0.44\omega\lesssim\bar{\tau}_s^{-1}\approx0.44 (A=0.3A=0.3, a=xs=1a=x_s=1 in Figure 3), both results approach the asymptotic prediction (13). Physically, faster driving requires more frequent SR to maintain effective phase selection, thereby shifting the resonance toward larger SR rates. The extension of this asymptotic scaling to a broader class of phase-selected responses beyond bistability, including systems with three or four stable states [49], is developed in Sec. S2 of Supplemental Material [48].

Figure 3. Optimal SR rate roptr_{\mathrm{opt}} versus driving frequency ω\omega for A=0.3A=0.3. Circles, squares, and diamonds denote the exact renewal prediction (8), the coarse-grained prediction (A4), and the weak-forcing, slow-driving asymptotic result (13), respectively.

Comparison with conventional NIR.—To compare SRIR with conventional NIR, we consider the same periodically driven bistable system without SR, r=0r=0, but subject to additive Gaussian white noise F=2D ξ(t)F=\sqrt{2D}\,\xi(t) on the right-hand side of Eq. (1), with ⟨ξ(t)⟩=0\langle\xi(t)\rangle=0 and ⟨ξ(t)ξ(t′)⟩=δ(t−t′)\langle\xi(t)\xi(t')\rangle=\delta(t-t'). For each pair (A,ω)(A,\omega), we solve the corresponding full-time periodic Fokker–Planck equation [50] and optimize the resonant amplification factor Gnoise(D)G_{\mathrm{noise}}(D) over the noise intensity DD, yielding Gnoisemax⁡=max⁡DGnoise(D)G_{\mathrm{noise}}^{\max}=\max_D G_{\mathrm{noise}}(D), see the numerical details in Sec. S4 of Supplemental Material [48]. The noise-driven case does not rely on a Kramers-rate or linear response approximation [51, 52]. We compare this directly with Gresetmax⁡=max⁡rGreset(r)G_{\mathrm{reset}}^{\max} = \max_r G_{\mathrm{reset}}(r). To provide a common reference for the two mechanisms, we also consider the limiting case of fully phase-locked switching. For an ideal two-state response satisfying ∣⟨x(t)⟩∣≤xs|\langle x(t)\rangle| \leq x_s, the largest possible first-harmonic amplitude is attained by a square wave that alternates between the two stable states every half-period of the weak drive, yielding

G≤Gsq=4xsπA.(14)G \leq G_{\mathrm{sq}} = \frac{4x_s}{\pi A}. \tag*{(14)}

The derivation is given in End Matter. This bound provides a natural reference for how closely the optimized SRIR and NIR responses approach the ideal square-wave limit.

For the system with same double-well potential and periodic driving, at their respective optimal parameters, SRIR produces an ensemble-averaged response much closer to square-wave response, whereas the NIR response remains less coherent, see Figure 4(a) and Figure 4(b). SR therefore approaches the square-wave limit more readily at finite frequencies because it bypasses activated barrier crossing and repeatedly returns the system directly to the unstable state.

SRIR versus NIR

Figure 4. SRIR versus NIR. (a),(b) Representative dynamics at A=0.1A = 0.1 and ω=0.01\omega= 0.01 for optimal SR (ropt≈0.035r_{\mathrm{opt}} \approx0.035) and optimal noise (Dopt≈0.072D_{\mathrm{opt}} \approx0.072), respectively. Thin curves show individual trajectories, thick curves the ensemble-averaged responses, black dashed curves the ideal square-wave response, and dotted curves the weak periodic input Acos⁡(ωt)A\cos(\omega t). (c) Optimized amplification factors Gresetmax⁡=max⁡rGreset(r)G_{\mathrm{reset}}^{\max} = \max_r G_{\mathrm{reset}}(r) (blue) and Gnoisemax⁡=max⁡DGnoise(D)G_{\mathrm{noise}}^{\max} = \max_D G_{\mathrm{noise}}(D) (red) as functions of the forcing amplitude AA at ω=0.01\omega= 0.01 (circles) and 0.10.1 (squares). The black dashed curve denotes the ideal square-wave response [12]. (d) Optimized amplification factors in the (A,ω)(A,\omega) plane. The black dashed curve indicates the ideal square-wave response.

Figure 4(c) further reveals quantitative differences between the two mechanisms and the ideal square-wave response. At ω=0.1\omega= 0.1, the optimized noise-induced amplification approaches an approximately AA-independent plateau, consistent with the linear-response prediction Gnoisemax⁡≈constG_{\mathrm{noise}}^{\max} \approx\mathrm{const} [29]. By contrast, the optimized SR amplification factor Gresetmax⁡G_{\mathrm{reset}}^{\max} increases as the forcing amplitude AA decreases, approaching ∝A−1\propto A^{-1} scaling of the ideal square-wave response [12]. Although both SRIR and NIR approach this reference in the limit ω→0\omega\to0, their finite-frequency behavior differs markedly. As shown in Figure 4(d), optimized SR yields substantially stronger resonant amplification over a broad range of finite driving frequencies.

Potential experimental realization and applications

The proposed SRIR could be accessible with existing experimental ingredients. Periodically driven bistable electronic circuits are well established [53], and Poisson-distributed control pulses could be used to reset the circuit voltage to the vicinity of an unstable state. Similarly, optical double-well platforms used to study stochastic resonance [54] could be combined with holographic-tweezer protocols for SR [18]. A further platform is spin-Hall-effect nanomagnets, where current pulses have been demonstrated to prepare magnetization near a metastable configuration, from which a weak magnetic bias can select a subsequent stable state [55]. In these realizations, instantaneous SR can be approached when the reset duration is much shorter than the mean reset time, the driving timescale, and the characteristic relaxation time.

Discussion.—SRIR was also reported in generalized Langevin dynamics with SR of the memory kernel [56]. In our scenario here the mechanism is qualitatively different: SR acts directly on the dynamical state. This state-space reinitialization provides a direct trajectory-level picture and lends itself naturally to controlled experimental SR. The SRIR and signal amplification mechanism studied here are robust for a finite spread of SR positions around the unstable state, see Eq. (25) and Figure C1 in End Matter, showing that precise resetting to the unstable equilibrium is not essential.

Figure C1. Robustness of SRIR to variations in the reset position (C1). The amplification factors G(r)G(r), obtained by the direct simulations from system (1), are shown for different SR widths μ/xs\mu/x_s at (a) A=0.3A=0.3 and (b) A=0.1A=0.1, with ω=0.1\omega=0.1. Dashed curves denote ideal SR (μ=0\mu=0). Parameters: a=b=xs=1a=b=x_s=1.

The results here are not restricted to the quartic double-well potential or even to bistable dynamics. The renewal representation (4) and its coherent harmonics apply to a generally periodically driven system under independent Poisson SR. The coarse-grained reduction can likewise be extended to multi-attractor systems, defining a broader class of phase-selected, finite-relaxation responses for which ropt(ω)∝ω2/3r_{\mathrm{opt}}(\omega) \propto\omega^{2/3} is valid. A detailed discussion of the generality and possible extensions is provided in Supplemental Material [48], while the corresponding SRIR in a tristable system is illustrated in Supplemental Movie S2.

In the bistable system (1) considered here, the periodic forcing may couple to both barrier modulation and post-reset stable state selection, so SRIR and NIR can coexist. An interesting question is whether SRIR and NIR can be selectively controlled, so that one resonance remains active while the other is suppressed. This possibility follows from a simple physical distinction between their symmetry-breaking mechanisms: In conventional NIR, the periodic driving breaks the left-right symmetry by alternately modulating the activation barriers, whereas in SRIR the driving breaks the symmetry of the post-reset dynamics. In Sec. S3 of Supplemental Material [48], we provide protocols to control the symmetry-breaking between barrier modulation and post-reset state selection, in which the SRIR remains active while the NIR is suppressed, and vice versa.

Future studies may investigate the interplay between SR and noise, as well as extensions to non-Poisson resetting, partial resetting [57], and optimized resetting protocols [13, 58]. Extending the present mechanism to underdamped dynamics [59, 60] may lead to qualitatively different resonance behavior depending on the resetting protocol. While resetting both position and velocity preserves the renewal structure, resetting position alone retains velocity memory, potentially modifying the amplification and the optimal resetting rate.

Acknowledgments

Acknowledgments.— K.L. acknowledges funding by the China Scholarship Council (CSC Project 202406830056). J.Z. acknowledges the support from the National Natural Science Foundation of China (Grant No. 12572037). R.M. acknowledges financial support from the German Science Foundation (DFG, Grant No. 318763901, CRC Data Assimilation and ME 1535/22-1).

Data availability

Data availability.— The code and data used to generate the simulation results and figures are available at [61].

Coarse-grained reduction of the exact renewal response.—To establish a clear connection with the exact renewal formulation, we substitute the coarse-grained post-reset trajectory representation (10) into the fundamental harmonic (6), yielding

h1(τ)=xs2π∫02πσϕe−iϕΘ[τ−τs(ϕ)] dϕ.(15)h_1(\tau)=\frac{x_s}{2\pi}\int_0^{2\pi}\sigma_\phi e^{-i\phi}\Theta[\tau-\tau_s(\phi)]\,d\phi. \tag*{(15)}

Here, τs(ϕ)=inf⁡{τ>0:∣xϕ(τ)∣=xs}\tau_s(\phi)=\inf\{\tau>0:|x_\phi(\tau)|=x_s\} and σϕ=sgn⁡[xϕ(τs(ϕ))]\sigma_\phi=\operatorname{sgn}[x_\phi(\tau_s(\phi))] are extracted directly from the exact post-reset dynamics (2), as illustrated in Figure A1. The discontinuities in τs(ϕ)\tau_s(\phi) shown in Figure A1(b) have two distinct origins. Those accompanied by a change of σϕ\sigma_\phi correspond to genuine phase-selection boundaries. The remaining jumps do not indicate a change of the selected well; they arise when the first relaxation excursion narrowly misses the prescribed position ±xs\pm x_s, delaying the first arrival until a later excursion.

Inserting Eq. (15) into the exact renewal coefficient (5) and performing the τ\tau integration yields

C1(r)=xsr2π(r+iω)∫02πσϕe−iϕe−(r+iω)τs(ϕ) dϕ,(16)C_1(r)=\frac{x_s r}{2\pi(r+i\omega)}\int_0^{2\pi}\sigma_\phi e^{-i\phi}e^{-(r+i\omega)\tau_s(\phi)}\,d\phi, \tag*{(16)}

and the corresponding resonant amplification factor is

G(r)=xsrπAr2+ω2∣∫02πσϕe−iϕe−(r+iω)τs(ϕ) dϕ∣.(17)G(r)=\frac{x_s r}{\pi A\sqrt{r^2+\omega^2}}\left|\int_0^{2\pi}\sigma_\phi e^{-i\phi}e^{-(r+i\omega)\tau_s(\phi)}\,d\phi\right|. \tag*{(17)}

The factors entering Eq. (17) have a direct physical interpretation: σϕ\sigma_\phi retains the phase-selected destination well, e−rτs(ϕ)e^{-r\tau_s(\phi)} is the probability that the reset-to-well relaxation is completed without an intervening reset, and e−iωτs(ϕ)e^{-i\omega\tau_s(\phi)} accounts for the phase accumulated during this relaxation.

Optimization of Eq. (17) gives the optimality condition in coarse-grained description

ω2ropt(ropt2+ω2)=τeff(ropt,ω),(18)\frac{\omega^2}{r_{\mathrm{opt}}(r_{\mathrm{opt}}^2+\omega^2)}=\tau_{\mathrm{eff}}(r_{\mathrm{opt}},\omega), \tag*{(18)}

where the weighted reset-to-well time is

τeff(r,ω)=Re⁡[∫02πτs(ϕ)σϕe−iϕe−(r+iω)τs(ϕ) dϕ∫02πσϕe−iϕe−(r+iω)τs(ϕ) dϕ].(19)\tau_{\mathrm{eff}}(r,\omega)=\operatorname{Re}\left[\frac{\int_0^{2\pi}\tau_s(\phi)\sigma_\phi e^{-i\phi}e^{-(r+i\omega)\tau_s(\phi)}\,d\phi}{\int_0^{2\pi}\sigma_\phi e^{-i\phi}e^{-(r+i\omega)\tau_s(\phi)}\,d\phi}\right]. \tag*{(19)}

This relation retains the same competition between SR and post-reset relaxation as the exact renewal condition (8), but is modified by the step-like coarse graining of the post-reset trajectory. The exact renewal condition (8) provides the quantitative prediction for roptr_{\mathrm{opt}}, whereas the coarse-grained reduction (A4) expresses the resonance directly in terms of the phase dependent reset-to-well time and enables analytical asymptotic scaling (13).

Ideal two-state square-wave response.—We derive the maximal fundamental harmonic response attainable within an ideal two-state switching and clarify the square-wave response used in the main text. Consider a periodic ensemble-averaged response ⟨x(t)⟩\langle x(t)\rangle with period T=2π/ωT=2\pi/\omega. In the ideal two-state switching limit, the system occupies only the two stable states ±xs\pm x_s, and therefore

∣⟨x(t)⟩∣≤xs.(20)|\langle x(t)\rangle|\le x_s. \tag*{(20)}

The coherent first harmonic amplitude can be written as

Xω=max⁡φ2T∫0T⟨x(t)⟩cos⁡(ωt−φ) dt,(21)X_\omega=\max_\varphi\frac{2}{T}\int_0^T\langle x(t)\rangle\cos(\omega t-\varphi)\,dt, \tag*{(21)}

where φ\varphi accounts for the phase lag of the response relative to the periodic drive. Using Eq. (20), we obtain

Xω≤2xsT∫0T∣cos⁡(ωt−φ)∣ dt=4xsπ.(22)X_\omega\le\frac{2x_s}{T}\int_0^T|\cos(\omega t-\varphi)|\,dt=\frac{4x_s}{\pi}. \tag*{(22)}

Thus, among all periodic responses restricted to the two states ±xs\pm x_s, the first harmonic amplitude cannot exceed 4xs/π4x_s/\pi.

The bound is attained when the response follows the sign of the fundamental harmonic, corresponding to fully phase-locked switching between the two stable states

⟨x(t)⟩sq=xssgn⁡[cos⁡(ωt−φ)]=4xsπ[cos⁡(ωt−φ)−13cos⁡3(ωt−φ)+15cos⁡5(ωt−φ)−⋯ ].(23)\langle x(t)\rangle_{\mathrm{sq}} = x_s \operatorname{sgn}[\cos(\omega t-\varphi)] = \frac{4x_s}{\pi}\left[\cos(\omega t-\varphi)-\frac{1}{3}\cos3(\omega t-\varphi)+\frac{1}{5}\cos5(\omega t-\varphi)-\cdots\right]. \tag*{(23)}

Hence, the fundamental harmonic has amplitude 4xs/π4x_s/\pi, and the corresponding amplification factor satisfies

Gsq=4xsπA,(24)G_{\mathrm{sq}}=\frac{4x_s}{\pi A}, \tag*{(24)}

which defines the two-state square-wave response. Equality corresponds to fully phase-locked switching, where the system occupies the drive favored state during each half cycle and switches once per half period.

Robustness to the resetting position.— The SRIR does not require resetting precisely to the unstable state of the unforced system. To quantify its robustness against a finite spread of SR positions around the unstable state, we replace the ideal SR condition by

x(tn)=xr+μηn,ηn∼N(0,1),(25)x(t_n)=x_r+\mu\eta_n,\qquad\eta_n\sim\mathcal{N}(0,1), \tag*{(25)}

where μ\mu characterizes the spatial uncertainty of the SR position. As shown in Figure C1, the resonant maximum of G(r)G(r) persists for finite μ\mu. Increasing μ\mu primarily reduces the peak amplification, whereas the position of the resonance remains comparatively insensitive. The suppression is more pronounced for weaker forcing.

This robustness can be understood from the linearized post-reset dynamics near the unstable state. For sufficiently slow driving, Eq. (2) reduces to dxφ/dτ∼axφ+Acos⁡φdx_\varphi/d\tau\sim ax_\varphi+A\cos\varphi, with xφ(0)=μηx_\varphi(0)=\mu\eta, yielding xφ(τ)∼μηeaτ+Acos⁡φa(eaτ−1)x_\varphi(\tau)\sim\mu\eta e^{a\tau}+\frac{A\cos\varphi}{a}(e^{a\tau}-1). For aτ≫1a\tau\gg1, this simplifies to xφ(τ)∼eaτ[μη+(A/a)cos⁡φ]x_\varphi(\tau)\sim e^{a\tau}[\mu\eta+(A/a)\cos\varphi]. Thus, fluctuations in the SR position introduce a random bias that competes with the phase-dependent bias of the periodic drive, with their relative strength characterized by

ϵr=aμA.(26)\epsilon_r=\frac{a\mu}{A}. \tag*{(26)}

For ϵr≪1\epsilon_r\ll1, the periodic bias dominates, and the resonant response is only weakly affected. As ϵr\epsilon_r approaches unity, positional fluctuations increasingly obscure the phase-dependent relaxation, explaining the greater sensitivity to resetting uncertainty at weaker forcing observed in Figure C1.

SM_new.pdf

This Supplemental Material provides the theoretical and numerical details supporting the main text. § 7 develops a general renewal formulation for periodically driven systems under stochastic resetting (SR). § 9 introduces a phase-dependent coarse-grained approximation, identifies the competing response losses underlying the optimal SR rate, and derives the low-frequency scaling of the optimal SR rate and discusses the generality beyond bistable system. § 10 discusses how SR-induced resonance (SRIR) and noise-induced resonance (NIR) can be selectively controlled. § 11 describes the numerical methods used for the SR dynamics, renewal calculations, and the comparison of SRIR with NIR. Finally, Sec. S5 describes Supplemental Movies S1 and S2, which illustrate the SRIR dynamics in bistable and tristable systems, respectively.

General renewal formulation

General periodically driven dynamics and renewal response

The exact response can be obtained without specifying a potential. Consider a dd-dimensional periodic dynamical system

dxdt=F(x,θ;A),(27)\frac{d\mathbf{x}}{dt}=\mathcal{F}(\mathbf{x},\theta;A), \tag*{(27)}
dθdt=ω,(28)\frac{d\theta}{dt}=\omega, \tag*{(28)}
F(x,θ;A)=F(x,θ+2π;A),(29)\mathcal{F}(\mathbf{x},\theta;A)=\mathcal{F}(\mathbf{x},\theta+2\pi;A), \tag*{(29)}

where θ\theta is the driving phase, ω\omega is the angular frequency, and AA parametrizes the amplitude of periodic forcing. At Poisson times with rate r>0r>0, the full dynamical state is redrawn from a normalized distribution ρR(x)\rho_R(\mathbf{x}), independently of its pre-reset value. The external phase θ\theta is not reset. Neither the inter-reset dynamics nor ρR\rho_R depends on rr. Fixed-point SR corresponds to ρR(x)=δ(x−xr)\rho_R(\mathbf{x})=\delta(\mathbf{x}-\mathbf{x}_r).

For a reset occurring at time trt_r, we define the reset phase ϕ=ϕ0+ωtr\phi=\phi_0+\omega t_r (mod 2π2\pi) and the elapsed time τ=t−tr\tau=t-t_r since that reset. We denote by Xϕ(τ;x0)\mathbf{X}_{\phi}(\tau;\mathbf{x}_0) the uninterrupted trajectory at elapsed time τ\tau, initialized at x0\mathbf{x}_0 immediately after a reset occurring at phase ϕ\phi. It satisfies

dXϕdτ=F(Xϕ,ϕ+ωτ;A),(30)\frac{d\mathbf{X}_{\phi}}{d\tau}=\mathcal{F}(\mathbf{X}_{\phi},\phi+\omega\tau;A), \tag*{(30)}
Xϕ(0;x0)=x0.(31)\mathbf{X}_{\phi}(0;\mathbf{x}_0)=\mathbf{x}_0. \tag*{(31)}

For a real observable QQ, we define the uninterrupted post-reset response as

qϕ(τ)=∫dx0 ρR(x0)Q[Xϕ(τ;x0)].(32)q_{\phi}(\tau)=\int d\mathbf{x}_0\,\rho_R(\mathbf{x}_0)Q[\mathbf{X}_{\phi}(\tau;\mathbf{x}_0)]. \tag*{(32)}

Here assume that QQ is bounded on the domain of reachable states. This is sufficient for convergence of the integrals and limiting operations below. All properties of the uninterrupted dynamics enter through qϕ(τ)q_{\phi}(\tau).

Similarly, for an initial state distribution ρinit(x0)\rho_{\mathrm{init}}(\mathbf{x}_0), the uninterrupted response from the initial preparation is

qinit(t)=∫dx0 ρinit(x0)Q[Xϕ0(t;x0)].(33)q_{\mathrm{init}}(t)=\int d\mathbf{x}_0\,\rho_{\mathrm{init}}(\mathbf{x}_0)Q[\mathbf{X}_{\phi_0}(t;\mathbf{x}_0)]. \tag*{(33)}

When the initial and reset-state distributions coincide, ρinit=ρR\rho_{\mathrm{init}}=\rho_R, one has qinit(t)=qϕ0(t)q_{\mathrm{init}}(t)=q_{\phi_0}(t).

At time tt, the probability that no reset has occurred is e−rte^{-rt}. For 0<τ<t0<\tau<t, the probability density that the most recent reset occurred at time (t−τ)(t-\tau) is re−rτre^{-r\tau}. Let Q(t)=⟨Q[x(t)]⟩Q(t)=\langle Q[\mathbf{x}(t)]\rangle denote the full ensemble-averaged response. Conditioning this on the most recent reset gives

Q(t)=e−rtqinit(t)+r∫0te−rτqϕ0+ω(t−τ)(τ) dτ.(34)Q(t)=e^{-rt}q_{\mathrm{init}}(t)+r\int_0^t e^{-r\tau}q_{\phi_0+\omega(t-\tau)}(\tau)\,d\tau. \tag*{(34)}

The reset-phase argument is always understood modulo 2π2\pi. In the long-time limit, the response approaches the periodic form

Qper(t)=r∫0∞e−rτqϕ0+ω(t−τ)(τ) dτ.(35)Q_{\mathrm{per}}(t)=r\int_{0}^{\infty}e^{-r\tau}q_{\phi_0+\omega(t-\tau)}(\tau)\,d\tau. \tag*{(35)}

For any bounded observable ∣Q∣≤Qmax⁡|Q|\le Q_{\max}, the finite-time response (S5) converges exponentially to this periodic state,

∣Q(t)−Qper(t)∣≤2Qmax⁡e−rt.(36)|Q(t)-Q_{\mathrm{per}}(t)|\le2Q_{\max}e^{-rt}. \tag*{(36)}

Thus, the periodic response follows directly from the renewal construction without requiring the uninterrupted system to possess a unique attractor.

Since qϕ(τ)q_{\phi}(\tau) is 2π2\pi-periodic in the reset phase ϕ\phi, we can use the Fourier series

qϕ(τ)=∑n=−∞∞hn(τ)einϕ,(37)q_{\phi}(\tau)=\sum_{n=-\infty}^{\infty}h_n(\tau)e^{in\phi}, \tag*{(37)}

where

hn(τ)=12π∫02πqϕ(τ)e−inϕ dϕ.(38)h_n(\tau)=\frac{1}{2\pi}\int_{0}^{2\pi}q_{\phi}(\tau)e^{-in\phi}\,d\phi. \tag*{(38)}

Substituting Eq. (S8) into Eq. (S6) gives

Qper(t)=∑n=−∞∞Cn(r;ϕ0)einωt,(39)Q_{\mathrm{per}}(t)=\sum_{n=-\infty}^{\infty}C_n(r;\phi_0)e^{in\omega t}, \tag*{(39)}

where

Cn(r;ϕ0)=einϕ0r∫0∞e−(r+inω)τhn(τ) dτ=einϕ0rh~n(r+inω),(40)\begin{aligned} C_n(r;\phi_0)&=e^{in\phi_0}r\int_{0}^{\infty}e^{-(r+in\omega)\tau}h_n(\tau)\,d\tau\\ &=e^{in\phi_0}r\widetilde{h}_n(r+in\omega), \tag*{(40)} \end{aligned}

with

h~n(s)=∫0∞e−sτhn(τ) dτ.(41)\widetilde{h}_n(s)=\int_{0}^{\infty}e^{-s\tau}h_n(\tau)\,d\tau. \tag*{(41)}

We note that the initial drive phase ϕ0\phi_0 only introduces the phase factor einϕ0e^{in\phi_0} in the nnth harmonic and therefore does not affect its amplitude. Without loss of generality, we set ϕ0=0\phi_0=0 hereafter

Cn(r;ϕ0=0):=Cn(r).(42)C_n(r;\phi_0=0):=C_n(r). \tag*{(42)}

Equations (S9)–(S12) provide the general renewal representation of the coherent harmonics. Its structure is independent of the detailed deterministic dynamics, which enters entirely through hn(τ)h_n(\tau). This is used in Eq. (4) of the main text.

Amplification factor

For the fundamental harmonic, the general renewal relation with the help of Eqs (S13) and (S12) gives

C1(r)=rh~1(r+iω).(43)C_1(r)=r\widetilde{h}_1(r+i\omega). \tag*{(43)}

Exact optimal condition

For either a general observable or the bistable position response, Eq. (S16) is valid. Taking the logarithmic derivative with respect to rr gives

dln⁡Gdr=1r+Re⁡[∂rh~1(r+iω)h~1(r+iω)].(44)\frac{d\ln G}{dr}=\frac{1}{r}+\operatorname{Re}\left[\frac{\partial_r\widetilde{h}_1(r+i\omega)}{\widetilde{h}_1(r+i\omega)}\right]. \tag*{(44)}

where we used dln⁡∣Z∣/dr=Re⁡(Z′/Z)d\ln|Z|/dr=\operatorname{Re}(Z'/Z) for a nonzero complex function Z(r)Z(r). Since h1(τ)h_1(\tau) is independent of rr, differentiation under the integral yields

∂rh~1(r+iω)=−∫0∞τe−(r+iω)τh1(τ) dτ.(45)\partial_r\widetilde{h}_1(r+i\omega)=-\int_{0}^{\infty}\tau e^{-(r+i\omega)\tau}h_1(\tau)\,d\tau. \tag*{(45)}

Therefore,

dln⁡Gdr=1r−τeff(r,ω),(46)\frac{d\ln G}{dr}=\frac{1}{r}-\tau_{\mathrm{eff}}(r,\omega), \tag*{(46)}

where

τeff(r,ω)=Re⁡∫0∞τe−(r+iω)τh1(τ) dτ∫0∞e−(r+iω)τh1(τ) dτ.(47)\tau_{\mathrm{eff}}(r,\omega)=\operatorname{Re}\frac{\int_{0}^{\infty}\tau e^{-(r+i\omega)\tau}h_1(\tau)\,d\tau}{\int_{0}^{\infty}e^{-(r+i\omega)\tau}h_1(\tau)\,d\tau}. \tag*{(47)}

Every interior stationary point for variable rr satisfies

r∗τeff(r∗,ω)=1.(48)r_*\tau_{\mathrm{eff}}(r_*,\omega)=1. \tag*{(48)}

For the bistable model, the optimal rate is selected by

ropt∈arg max⁡r≥0 G(r)(49)r_{\mathrm{opt}}\in\underset{r\ge0}{\operatorname{arg\,max}}\,G(r) \tag*{(49)}

The coherent first harmonic amplitude of the observable QQ is

Qω(r)=2∣C1(r)∣.(50)Q_{\omega}(r)=2|C_1(r)|. \tag*{(50)}

We define the corresponding amplification factor GG as the ratio of the output amplitude Qω(r)Q_{\omega}(r) to the forcing amplitude AA,

G(r)≡Qω(r)A=2∣C1(r)∣A=2rA∣h~1(r+iω)∣.(51)G(r)\equiv\frac{Q_{\omega}(r)}{A}=\frac{2|C_1(r)|}{A}=\frac{2r}{A}\left|\widetilde{h}_1(r+i\omega)\right|. \tag*{(51)}

Using the definition of h1(τ)h_1(\tau), the amplification factor can equivalently be written as

G(r)=rπA∣∫02πdϕ∫0∞dτ qϕ(τ)e−rτ−i(ϕ+ωτ)∣.(52)G(r)=\frac{r}{\pi A}\left|\int_{0}^{2\pi}d\phi\int_{0}^{\infty}d\tau\,q_{\phi}(\tau)e^{-r\tau-i(\phi+\omega\tau)}\right|. \tag*{(52)}

The detailed uninterrupted dynamics enters entirely through qϕ(τ)q_{\phi}(\tau), while the resetting statistics enter through the exponential age weight re−rτre^{-r\tau}. These results and definitions are used for Eq. (7) in the main text.

using the appropriate no-reset response at the boundary r=0r = 0.

If the maximum is interior, it must obey Eq. (48); roots of that condition must be checked against one another and against boundary values. Numerical integration of the exact response, followed by one-dimensional optimization or root finding with this check, yields the prediction without an additional dynamical approximation. Equation (S22) is only a condition for an interior extremum; it neither guarantees that such an extremum exists nor that it is a maximum. The following simple example illustrates this point.

The existence of an interior optimum is not guaranteed by the renewal structure alone. As a simple counterexample, consider the driven linear equation

dx(t)dt=−κx(t)+Acos⁡(ωt),(53)\frac{dx(t)}{dt}=-\kappa x(t)+A\cos(\omega t), \tag*{(53)}

where κ>0\kappa> 0 and the reset position is xr=0x_{\mathrm r}=0. For a reset occurring at phase ϕ\phi, the uninterrupted trajectory is

xϕ(τ)=ARe⁡[eiϕeiωτ−e−κτκ+iω],(54)x_{\phi}(\tau)=A\operatorname{Re}\left[e^{i\phi}\frac{e^{i\omega\tau}-e^{-\kappa\tau}}{\kappa+i\omega}\right], \tag*{(54)}

so that

h1(τ)=A2eiωτ−e−κτκ+iω.(55)h_1(\tau)=\frac{A}{2}\frac{e^{i\omega\tau}-e^{-\kappa\tau}}{\kappa+i\omega}. \tag*{(55)}

Substitution this into the exact renewal relation Eqs. (43) and (51) gives

C1(r)=A2(r+κ+iω),G(r)=1(r+κ)2+ω2.(56)C_1(r)=\frac{A}{2(r+\kappa+i\omega)},\qquad G(r)=\frac{1}{\sqrt{(r+\kappa)^2+\omega^2}}. \tag*{(56)}

Hence dG/dr<0dG/dr < 0 for all r≥0r \ge0, and the maximum occurs at the boundary ropt=0r_{\mathrm{opt}}=0. Thus, a finite post-reset relaxation time alone is insufficient to produce SRIR. An interior optimum requires additional response structure, such as the competition between reset-triggered phase selection and finite post-reset relaxation, as developed in § 9.

Coarse-Grained Reduction

The exact renewal formulation developed in § 7 applies to general periodically driven systems under Poisson resetting, without presupposing or guaranteeing resonance. We now identify a more restrictive response-level structure that gives rise to competing effects responsible for an optimal resetting rate. This construction requires neither bistability nor multistability; it does not assume resetting near an unstable state either. Local dynamical instability provides one natural mechanism for realizing such a response structure.

Optimal rate scaling

Suppose that a reset occurring at forcing phase ϕ\phi selects a well-defined post-reset response Qsel(ϕ)Q_{\mathrm{sel}}(\phi) where the subscript "sel" denotes "selection". This response is established after a finite, phase-dependent relaxation time τs(ϕ)\tau_s(\phi). At the level of the observable entering the coherent response, the reduction is

qϕ(τ)=Qsel(ϕ)Θ[τ−τs(ϕ)].(57)q_{\phi}(\tau)=Q_{\mathrm{sel}}(\phi)\Theta[\tau-\tau_s(\phi)]. \tag*{(57)}

Here Θ[z]\Theta[z] is the Heaviside function, Qsel(ϕ)Q_{\mathrm{sel}}(\phi) denotes the selected observable output and τs(ϕ)\tau_s(\phi) characterizes the time required for this output to develop following a reset. We note that Eq. (57) is a coarse-grained approximation in which the transient evolution toward the selected state is neglected, and the observable is assumed to reach its selected value only after the relaxation time τs(ϕ)\tau_s(\phi).

Equation (S28) is a response-level description and does not require a particular state-space geometry. For a system with KK phase-selectable branches, Qsel(ϕ)Q_{\mathrm{sel}}(\phi) may take the values {Q1,…,QK}\{Q_1,\ldots,Q_K\}. The essential requirement is that, once established, the selected output remains approximately persistent on time scale relevant to the renewal response. Systems whose post-reset observable continues to undergo an intrinsic oscillation after the initial relaxation need not belong to this class, even though the exact renewal formulation in § 7 remains applicable.

Substituting Eq. (57) into the general renewal representation (S16) gives

G(r)=2rAr2+ω2∣M(r+iω)∣,(58)G(r)=\frac{2r}{A\sqrt{r^2+\omega^2}}\left|\mathcal{M}(r+i\omega)\right|, \tag*{(58)}
M(z)=12π∫02πQsel(ϕ)e−iϕe−zτs(ϕ) dϕ.(59)\mathcal{M}(z)=\frac{1}{2\pi}\int_0^{2\pi}Q_{\mathrm{sel}}(\phi)e^{-i\phi}e^{-z\tau_s(\phi)}\,d\phi. \tag*{(59)}

These relations are exact within the reduced step-response description. The factor r/(r+iω)r/(r+i\omega) describes the renewal of the reset phase, whereas M(z)\mathcal{M}(z) contains the phase-selected post-reset relaxation dynamics.

To characterize M(z)\mathcal{M}(z), we introduce the response weighted moments

Mk=12π∫02πQsel(ϕ)e−iϕτsk(ϕ) dϕ,k={0,1,…},(60)M_k=\frac{1}{2\pi}\int_0^{2\pi}Q_{\mathrm{sel}}(\phi)e^{-i\phi}\tau_s^k(\phi)\,d\phi,\qquad k=\{0,1,\ldots\}, \tag*{(60)}

which are generally complex. We assume M0≠0M_0\ne0 and write

M1M0=τb+iτp,(61)\frac{M_1}{M_0}=\tau_b+i\tau_p, \tag*{(61)}

where

τb=Re⁡(M1M0),τp=Im⁡(M1M0).(62)\tau_b=\operatorname{Re}\left(\frac{M_1}{M_0}\right),\qquad\tau_p=\operatorname{Im}\left(\frac{M_1}{M_0}\right). \tag*{(62)}

We refer to τb\tau_b as the effective relaxation time. It differs from an ordinary phase-averaged relaxation time because the weight Qsel(ϕ)e−iϕQ_{\mathrm{sel}}(\phi)e^{-i\phi} retains specifically those post-reset events that contribute coherently to the fundamental harmonic. A positive τb\tau_b means that finite post-reset relaxation produces a leading suppression of the coherent response as the resetting rate increases, as shown below.

For fixed forcing parameters, both Qsel(ϕ)Q_{\mathrm{sel}}(\phi) and τs(ϕ)\tau_s(\phi) are determined entirely by the uninterrupted post-reset dynamics and are therefore independent of the resetting rate rr. Consequently, τb\tau_b is also independent of rr, although it may in general depend on AA, ω\omega, and other system parameters.

To control higher-order corrections, we introduce the second absolute weighted moment

M^2=12π∫02π∣Qsel(ϕ)∣τs2(ϕ) dϕ.(63)\widehat{M}_2 = \frac{1}{2\pi}\int_0^{2\pi} \lvert Q_{\mathrm{sel}}(\phi)\rvert\tau_s^2(\phi)\,\mathrm{d}\phi. \tag*{(63)}

and define

χ2=M^2∣M0∣+∣M1M0∣2.(64)\chi_2 = \frac{\widehat{M}_2}{\lvert M_0\rvert}+\left\lvert\frac{M_1}{M_0}\right\rvert^2 . \tag*{(64)}

We focus on phase-selected responses satisfying the convergence

M^2<∞.(65)\widehat{M}_2 < \infty. \tag*{(65)}

The first condition ensures a positive relaxation loss with increasing resetting rate, while the second controls the higher-order corrections. In particular, for M^2<∞\widehat{M}_2 < \infty, the transform M(z)\mathcal{M}(z) admits, as z→0z\to0,

M(z)=M0−zM1+O(∣z∣2M^2).(66)\mathcal{M}(z)=M_0-zM_1+O\left(\lvert z\rvert^2\widehat{M}_2\right). \tag*{(66)}

Setting z=r+iωz=r+i\omega gives

∣M(r+iω)∣∣M0∣=1−rτb+ωτp+O[(r2+ω2)χ2].(67)\frac{\lvert\mathcal{M}(r+i\omega)\rvert}{\lvert M_0\rvert}=1-r\tau_b+\omega\tau_p+O\left[(r^2+\omega^2)\chi_2\right]. \tag*{(67)}

The term ωτp\omega\tau_p modifies the low-frequency reference response, but it is independent of rr to this order. In contrast, rτbr\tau_b is the leading relaxation loss relevant to optimization over the resetting rate.

From Eq. (S29a), it follows that

G(r)G0=rr2+ω2∣M(r+iω)∣∣M0∣,G0=2∣M0∣A.(68)\frac{G(r)}{G_0}=\frac{r}{\sqrt{r^2+\omega^2}}\frac{\lvert\mathcal{M}(r+i\omega)\rvert}{\lvert M_0\rvert},\qquad G_0=\frac{2\lvert M_0\rvert}{A}. \tag*{(68)}

For any finite ω>0\omega>0, sufficiently rare resetting gives

rr2+ω2∼rω→0,(69)\frac{r}{\sqrt{r^2+\omega^2}}\sim\frac{r}{\omega}\to0, \tag*{(69)}

so that the resetting-generated coherent response is suppressed because phase-selection events occur too infrequently.

At larger resetting rates, finite post-reset relaxation produces the competing loss rτbr\tau_b. For slow driving, these two limits leave an intermediate asymptotic window

ω≪r≪τb−1,(70)\omega\ll r\ll\tau_b^{-1}, \tag*{(70)}

where τb\tau_b is the effective relaxation time defined in Eq. (62). Within this regime, we have

rr2+ω2=1−ω22r2+O(ω4r4).(71)\frac{r}{\sqrt{r^2+\omega^2}}=1-\frac{\omega^2}{2r^2}+O\left(\frac{\omega^4}{r^4}\right). \tag*{(71)}

Combining this expression with Eq. (67) gives

G(r)G0=1+ωτp−ω22r2−rτb+O[ω4r4+(r2+ω2)χ2+ω2r2(rτb+∣ωτp∣)].(72)\frac{G(r)}{G_0}=1+\omega\tau_p-\frac{\omega^2}{2r^2}-r\tau_b +O\left[\frac{\omega^4}{r^4}+(r^2+\omega^2)\chi_2+\frac{\omega^2}{r^2}(r\tau_b+\lvert\omega\tau_p\rvert)\right]. \tag*{(72)}

The two leading rr-dependent losses in Eq. (72) are therefore

L(r)=ω22r2+rτb.(73)L(r)=\frac{\omega^2}{2r^2}+r\tau_b. \tag*{(73)}

The first is the finite-rate phase-selection loss: resetting events that are too rare cannot update the selected response sufficiently rapidly relative to the external phase. The second is the interruption loss: overly frequent resets suppress the finite post-reset response relaxation. This is one of the main technical sides of the paper.

As

L′(r)=−ω2r3+τb,L′′(r)=3ω2r4>0,(74)L'(r)=-\frac{\omega^2}{r^3}+\tau_b,\qquad L''(r)=\frac{3\omega^2}{r^4}>0, \tag*{(74)}

the leading loss has a unique minimum at

ropt(ω)∼(ω2τb)1/3.(75)r_{\mathrm{opt}}(\omega)\sim\left(\frac{\omega^2}{\tau_b}\right)^{1/3}. \tag*{(75)}

If the higher-order moments remaining suitably bounded, Eq. (65), one can yield the asymptotic scaling ropt∝ω2/3r_{\mathrm{opt}}\propto\omega^{2/3}. The exponent therefore follows from the competing response losses rather than from a specific potential or resetting location. This is one of the key results, claimed in Eq. (13) of the main text.

Bistable realization

We now apply the general coarse-grained reduction to the bistable system Eqs. (1) and (2) of the main text. The reduced post-reset response is

xϕ(τ)=σϕxΘ[τ−τs(ϕ)].(76)x_\phi(\tau)=\sigma_\phi x_\Theta[\tau-\tau_s(\phi)]. \tag*{(76)}

where σϕ=±1\sigma_\phi=\pm1 denotes the selected well and τs(ϕ)\tau_s(\phi) is the corresponding reset-to-well relaxation time.

For weak forcing and sufficiently slow driving, near the unforced unstable state, the post-reset dynamics becomes

dxϕ(τ)dτ∼axϕ(τ)+Acos⁡ϕ(77)\frac{dx_{\phi}(\tau)}{d\tau} \sim ax_{\phi}(\tau) + A\cos\phi \tag*{(77)}

with the solution

xϕ(τ)∼Acos⁡ϕa(eaτ−1).(78)x_{\phi}(\tau) \sim\frac{A\cos\phi}{a}\left(e^{a\tau}-1\right). \tag*{(78)}

This gives the approximate selection rule of the response (S46)

σϕ∼sgn⁡(cos⁡ϕ).(79)\sigma_{\phi} \sim\operatorname{sgn}(\cos\phi). \tag*{(79)}

Using Qsel(ϕ)=xssgn⁡(cos⁡ϕ)Q_{\mathrm{sel}}(\phi)=x_s\operatorname{sgn}(\cos\phi) in the general moments (S30) to (S32), yields

τˉb=τˉs=14∫02π∣cos⁡ϕ∣τs(ϕ) dϕ.(80)\bar{\tau}_b=\bar{\tau}_s=\frac{1}{4}\int_{0}^{2\pi}|\cos\phi|\tau_s(\phi)\,d\phi. \tag*{(80)}

Thus, the effective relaxation time τb\tau_b reduces to the first-harmonic-weighted reset-to-well time τˉs\bar{\tau}_s.

For motion toward the selected well, let

Fϕ=A∣cos⁡ϕ∣,ϵϕ=Fϕaxs.(81)F_{\phi}=A|\cos\phi|,\qquad\epsilon_{\phi}=\frac{F_{\phi}}{ax_s}. \tag*{(81)}

The relaxation time of the system (2) in the main text is

τs(ϕ)∼∫0xsdxax−bx3+Fϕ=1a∫01dyy−y3+ϵϕ,(82)\tau_s(\phi)\sim\int_{0}^{x_s}\frac{dx}{ax-bx^3+F_{\phi}}=\frac{1}{a}\int_{0}^{1}\frac{dy}{y-y^3+\epsilon_{\phi}}, \tag*{(82)}

where y=x/xsy=x/x_s. For ϵϕ≪1\epsilon_{\phi}\ll1, the neighborhoods of both y=0y=0 and y=1y=1 contribute to the reset-to-well time τs(ϕ)\tau_s(\phi) logarithmically, yielding

τs(ϕ)∼32aln⁡(axsA∣cos⁡ϕ∣).(83)\tau_s(\phi)\sim\frac{3}{2a}\ln\left(\frac{ax_s}{A|\cos\phi|}\right). \tag*{(83)}

The factor 3/(2a)3/(2a) contains both the unstable departure contribution a−1ln⁡(1/ϵϕ)a^{-1}\ln(1/\epsilon_{\phi}) and the near-well contribution (2a)−1ln⁡(1/ϵϕ)(2a)^{-1}\ln(1/\epsilon_{\phi}).

The logarithmic divergence at the isolated phase-selection boundaries cos⁡ϕ=0\cos\phi=0 is integrable, so the second absolute weighted moment used in Sec. S2.1 remains finite. Substituting Eq. (S53) into Eq. (S50) gives

τˉs∼32a[ln⁡(axs2A)+1].(84)\bar{\tau}_s\sim\frac{3}{2a}\left[\ln\left(\frac{ax_s}{2A}\right)+1\right]. \tag*{(84)}

Hence τb=τˉs>0\tau_b=\bar{\tau}_s>0, and the condition (S35) required by the general phase-selected reduction, is satisfied.

Applying the amplification factor (S42) gives

G(r)G0∼1−ω22r2−rτˉs,G0=4xsπA,(85)\frac{G(r)}{G_0}\sim1-\frac{\omega^2}{2r^2}-r\bar{\tau}_s,\qquad G_0=\frac{4x_s}{\pi A}, \tag*{(85)}

and therefore taking the maximum of G(r)G(r) yields

ropt(ω)∼(ω2τˉs)1/3.(86)r_{\mathrm{opt}}(\omega)\sim\left(\frac{\omega^2}{\bar{\tau}_s}\right)^{1/3}. \tag*{(86)}

The scaling ropt(ω)∝ω2/3r_{\mathrm{opt}}(\omega)\propto\omega^{2/3} follows from the general two-loss balance, whereas the logarithmic amplitude dependence τˉs(A)\bar{\tau}_s(A) originates from the specific reset-to-well relaxation of the bistable dynamics. Here, we refer to the description of Figure 3 in the main text.

Three-stable-state realization

To demonstrate that the phase-selected finite-relaxation structure is not restricted to a bistable or one-dimensional geometry, we consider a two-dimensional system with three symmetry-related attracting states. In polar coordinates, the unforced potential is

U(ρ,ϑ)=ρ44−ρ22−ϵρ3cos⁡(3ϑ),(87)U(\rho,\vartheta)=\frac{\rho^4}{4}-\frac{\rho^2}{2}-\epsilon\rho^3\cos(3\vartheta), \tag*{(87)}

and the periodically driven dynamics is

dxdt=−∇U(x)+A(cos⁡(ωt)sin⁡(ωt)),(88)\frac{d\mathbf{x}}{dt}=-\nabla U(\mathbf{x})+A \begin{pmatrix} \cos(\omega t)\\ \sin(\omega t) \end{pmatrix}, \tag*{(88)}

with Poisson resetting to the origin. In Cartesian coordinates, one has

dxdt=x−(x2+y2)x+3ϵ(x2−y2)+Acos⁡(ωt),\frac{dx}{dt}=x-(x^2+y^2)x+3\epsilon(x^2-y^2)+A\cos(\omega t),
dydt=y−(x2+y2)y−6ϵxy+Asin⁡(ωt).(89)\frac{dy}{dt}=y-(x^2+y^2)y-6\epsilon xy+A\sin(\omega t). \tag*{(89)}

In the absence of forcing, the three attractors (A) lie at

ϑj=2π(j−1)3,j={1,2,3},(90)\vartheta_j=\frac{2\pi(j-1)}{3},\qquad j=\{1,2,3\}, \tag*{(90)}

with radius

ρA=3ϵ+9ϵ2+42.(91)\rho_A=\frac{3\epsilon+\sqrt{9\epsilon^2+4}}{2}. \tag*{(91)}

They are separated by three saddle (S) at ϑ=(2j−1)π/3\vartheta=(2j-1)\pi/3 with radius

ρS=−3ϵ+9ϵ2+42.(92)\rho_S=\frac{-3\epsilon+\sqrt{9\epsilon^2+4}}{2}. \tag*{(92)}

The origin is a repelling point of the unforced system, see Figure 1 in the main text. We choose the forcing sufficiently weak that, after the initial post-reset branch selection, the deterministic trajectory remains associated with the selected attracting branch until another reset occurs. Let j(ϕ)∈{1,2,3}j(\phi)\in\{1,2,3\} denote the branch selected by an uninterrupted trajectory initiated at the origin at forcing phase ϕ\phi. Introducing

τs(ϕ)=inf⁡{τ>0:ρϕ(τ)=ρA},(93)\tau_s(\phi)=\inf\{\tau>0:\rho_{\phi}(\tau)=\rho_A\}, \tag*{(93)}

and assign the complex selected output

Qsel(ϕ)=ρAexp⁡(2πi3(j(ϕ)−1)).(94)Q_{\mathrm{sel}}(\phi)=\rho_A\exp\left(\frac{2\pi i}{3}\left(j(\phi)-1\right)\right). \tag*{(94)}

Equations (S63) and (S64) have exactly the structure assumed in Eq. (57). These quantities enter directly into the moments (S30) in § 9, giving τb=Re⁡(M1/M0)\tau_b=\operatorname{Re}(M_1/M_0). For the parameters considered in Figure S1, τb>0\tau_b>0 and the second absolute weighted moment remains finite, so that the conditions of the phase-selected reduction are satisfied.

Figure S1. Three-stable-state realization of SRIR. (a) Reset-induced phase dependence of the selected attractor. Different colors denote three attracting branches. (b) Amplification factor G(r)G(r), showing a maximum at the finite optimal SR rate ropt≈0.18r_{\mathrm{opt}}\approx0.18 (dashed line). The solid line represents the prediction from the exact renewal formulation, while the open circles denote direct Monte Carlo simulations. (c) Optimal SR rate versus driving frequency. Open circles denote direct simulations and the dashed line shows the theoretical prediction (S45). Parameters are ϵ=1\epsilon=1 and A=0.3A=0.3; panels (a) and (b) use ω=0.1\omega=0.1.

Figure S1 shows the phase-dependent relaxation toward the three attracting branches. Each reset phase selects one of the three attracting branches. The Poisson SR dynamics also exhibits a pronounced maximum of the coherent response at a finite resetting rate, as shown in Figure S1. Figure S1 compares the numerically determined roptr_{\mathrm{opt}} with the theoretical prediction (S45). The numerical results follow this predicted scaling ropt(ω)∝ω2/3r_{\mathrm{opt}}(\omega)\propto\omega^{2/3} over the frequency range considered.

More three- and four-stable-state rotational systems considered in [4] also exhibit the same optimal-rate scaling although these results are not shown here. These findings demonstrate that the competition between repeated resetting and finite post-reset relaxation extends beyond bistability.

Selective control of SRIR and NIR

SRIR and conventional NIR both exploit a periodically induced symmetry breaking, but the driving acts on different dynamical objects. In conventional NIR, it modulates noise-activated escape from the wells, whereas in SRIR it biases the outgoing motion after reinjection near an unstable state. We distinguish these two coupling mechanism and construct here the examples in which one is present while the other is absent.

Consider a symmetric bistable potential U0(x)U_0(x) with stable states x±x_{\pm} and an intermediate unstable state xux_u, satisfying

U0′(x±)=U0′(xu)=0,U_0'(x_{\pm})=U_0'(x_u)=0,
U0′′(x±)>0,U0′′(xu)<0,(95)U_0''(x_{\pm})>0,\quad U_0''(x_u)<0, \tag*{(95)}

where the prime and double prime denote the first and second derivatives with respect to xx, respectively. Introduce a weak periodic perturbation through

U(x,t)=U0(x)−Ag(x)cos⁡(ωt),U(x,t)=U_0(x)-Ag(x)\cos(\omega t),
dxdt=−U0′(x)+Ag′(x)cos⁡(ωt).(96)\frac{dx}{dt}=-U_0'(x)+Ag'(x)\cos(\omega t). \tag*{(96)}

The main equation (1) of the main text considered corresponds to g(x)=xg(x)=x. In the following, We compare noise-free dynamics with Poisson resetting to xux_u against dynamics without resetting, but with additive Gaussian white noise of intensity DD.

Distinct symmetry-breaking mechanisms

The different physical origins of the two resonances can be made explicit for the general periodic coupling introduced above. We first consider the dynamics immediately after resetting to the unstable state xux_u. Writing x=xu+δxx=x_u+\delta x, the initial post-reset motion in the quasistatic regime obeys, to leading order,

d(δx)dt∼λuδx+Ag′(xu)cos⁡ϕ,(97)\frac{d(\delta x)}{dt}\sim\lambda_u\delta x+Ag'(x_u)\cos\phi, \tag*{(97)}

where λu=−U0′′(xu)>0\lambda_u=-U_0''(x_u)>0 and ϕ\phi is the drive phase at resetting. The phase-dependent bias of the outgoing motion is therefore controlled by the parameter

χSRIR≡g′(xu).(98)\chi_{\mathrm{SRIR}}\equiv g'(x_u). \tag*{(98)}

For χSRIR≠0\chi_{\mathrm{SRIR}}\ne0, the periodic signal biases the direction in which the system leaves the unstable reset state. Provided that the outgoing branches lead to distinct attracting basins, this local bias selects the subsequent attractor and is amplified by the instability. A finite optimal resetting rate additionally requires the competition between phase selection and interruption of response build-up discussed in the main text. If instead g′(xu)=0g'(x_u)=0 the leading signal-induced bias at the reset state vanishes.

Conventional stochastic resonance probes a different aspect of the same periodic coupling. Let xu(t)x_u(t) and x±(t)x_\pm(t) denote the instantaneous saddle and local minima, which approach xux_u and x±x_\pm as A→0A\to0. For weak forcing,

xi(t)=xi+δxi(t),δxi(t)=O(A),(99)x_i(t)=x_i+\delta x_i(t),\qquad\delta x_i(t)=O(A), \tag*{(99)}

where i={u,+,−}i=\{u,+,-\} and U0′(xi)=0U_0'(x_i)=0. Expanding the potential at the shifted extrema gives

U[xi(t),t]=U0[xi+δxi(t)]−Ag[xi+δxi(t)]cos⁡(ωt)=U0(xi)+U0′(xi)δxi(t)−A[g(xi)+g′(xi)δxi(t)]cos⁡(ωt)+O(A2)=U0(xi)−Ag(xi)cos⁡(ωt)+O(A2),(100)\begin{aligned} U[x_i(t),t]&=U_0[x_i+\delta x_i(t)]-Ag[x_i+\delta x_i(t)]\cos(\omega t)\\ &=U_0(x_i)+U_0'(x_i)\delta x_i(t)\\ &\quad-A\left[g(x_i)+g'(x_i)\delta x_i(t)\right]\cos(\omega t)+O(A^2)\\ &=U_0(x_i)-Ag(x_i)\cos(\omega t)+O(A^2), \tag*{(100)} \end{aligned}

where we used U0′(xi)=0U_0'(x_i)=0 and δxi(t)=O(A)\delta x_i(t)=O(A), so that Ag′(xi)δxi(t)=O(A2)Ag'(x_i)\delta x_i(t)=O(A^2). The instantaneous activation barriers are therefore

ΔU±(t)=U[xu(t),t]−U[x±(t),t]=U0(xu)−U0(x±)−A[g(xu)−g(x±)]cos⁡(ωt)+O(A2).(101)\begin{aligned} \Delta U_\pm(t)&=U[x_u(t),t]-U[x_\pm(t),t]\\ &=U_0(x_u)-U_0(x_\pm)\\ &\quad-A[g(x_u)-g(x_\pm)]\cos(\omega t)+O(A^2). \tag*{(101)} \end{aligned}

The directional barrier asymmetry is

ΔU+(t)−ΔU−(t)=AχNIRcos⁡(ωt)+O(A2),(102)\Delta U_+(t)-\Delta U_-(t)=A\chi_{\mathrm{NIR}}\cos(\omega t)+O(A^2), \tag*{(102)}

where

χNIR≡g(x+)−g(x−).(103)\chi_{\mathrm{NIR}}\equiv g(x_+)-g(x_-). \tag*{(103)}

Thus χNIR≠0\chi_{\mathrm{NIR}}\ne0 values periodically favor noise-induced escape from one well over the other and provides the conventional route to coherent interwell switching. When g(x+)=g(x−)g(x_+)=g(x_-), the leading directional barrier modulation vanishes. This does not imply that the complete noise-driven response vanishes, because common barrier modulation, curvature-dependent prefactors, intrawell motion, and higher-order barrier changes may still contribute.

The two resonance mechanisms therefore probe different properties of periodic coupling, measured by the quantities (S68) and (S73) respectively. The former is determined locally by forcing near the unstable reset state, whereas the latter reflects the global energetic asymmetry between the two wells. These quantities can be varied independently, allowing the post-reset phase-selection mechanism to remain active when the leading barrier-asymmetry mechanism is suppressed, or vice versa.

For the standard additive forcing as discussed in the main text, g(x)=xg(x)=x, one has

χSRIR=1,χNIR=x+−x−,(104)\chi_{\mathrm{SRIR}}=1,\qquad\chi_{\mathrm{NIR}}=x_+-x_-, \tag*{(104)}

so that both mechanisms are present. Thus, in the standard periodically tilted bistable system, the same forcing can simultaneously bias noise-induced barrier crossing and post-reset attractor selection.

The examples below realize two different cases: the first retains the post-reset phase-selection mechanism while suppressing the leading barrier asymmetry, i.e.,

χSRIR≠0,χNIR=0(105)\chi_{\mathrm{SRIR}}\ne0,\qquad\chi_{\mathrm{NIR}}=0 \tag*{(105)}

whereas the second retains the leading barrier-asymmetry mechanism while suppressing post-reset phase selection, so that

χSRIR=0,χNIR≠0.(106)\chi_{\mathrm{SRIR}}=0,\qquad\chi_{\mathrm{NIR}}\ne0. \tag*{(106)}

Example 1: Active SRIR and suppressed NIR

Consider the quartic double well

U0(x)=−a2x2+b4x4,(107)U_0(x)=-\frac{a}{2}x^2+\frac{b}{4}x^4, \tag*{(107)}

where xu=0x_u=0, x±=±xsx_\pm=\pm x_s, xs=a/bx_s=\sqrt{a/b}. We choose first

g(x)=x(1−x2xs2).(108)g(x)=x\left(1-\frac{x^2}{x_s^2}\right). \tag*{(108)}

and the driven deterministic equation becomes

dxdt=ax−bx3+A(1−3x2xs2)cos⁡(ωt).(109)\frac{dx}{dt}=ax-bx^3+A\left(1-\frac{3x^2}{x_s^2}\right)\cos(\omega t). \tag*{(109)}

Then we have

χSRIR=1,χNIR=0.(110)\chi_{\mathrm{SRIR}}=1,\qquad\chi_{\mathrm{NIR}}=0. \tag*{(110)}

Near x=0x=0, its quasistatic form is x˙∼ax+Acos⁡ϕ\dot{x}\sim ax+A\cos\phi. Hence the same leading local seed as in the main-text model is retained. Meanwhile, for sufficiently weak forcing, ΔU±(t)=ΔU0+O(A2)\Delta U_{\pm}(t)=\Delta U_0+O(A^2). Both activation barriers have no first-order height modulation because g(x)g(x) in Eq. (108) vanishes at all three unperturbed extrema. This construction therefore preserves the local SRIR seeding channel while removing the leading barrier-height modulation.

Figure S2(a) shows a case in which SR produces a pronounced resonant enhancement. The SRIR gain develops a clear maximum at an intermediate resetting rate, whereas the corresponding NIR response remains much weaker over the same parameter range. This contrast indicates that the strong response is not a generic consequence of stochastic fluctuations. Rather, it originates from the reset-induced phase selection, through which reinitialization repeatedly places the system in a dynamically susceptible configuration where the weak periodic bias can efficiently select and amplify the subsequent response.

Example 2: Suppressed SRIR and active NIR

For the same quartic potential, we choose now instead

g(x)=x33xs2.(111)g(x)=\frac{x^3}{3x_s^2}. \tag*{(111)}

The deterministic dynamics is

dxdt=ax−bx3+Ax2xs2cos⁡(ωt).(112)\frac{dx}{dt}=ax-bx^3+A\frac{x^2}{x_s^2}\cos(\omega t). \tag*{(112)}

Then we have

χSRIR=0,χNIR=2xs3.(113)\chi_{\mathrm{SRIR}}=0,\qquad\chi_{\mathrm{NIR}}=\frac{2x_s}{3}. \tag*{(113)}

At the ideal reset point, x˙=0\dot{x}=0, for every driving phase. Thus a trajectory reset exactly to xr=0x_r=0 remains there for all subsequent times in the noise-free dynamics. The deterministic phase-selection mechanism is absent, not merely small to first order. Nevertheless, the activation barriers obey ΔU+(t)−ΔU−(t)=2Axs3cos⁡(ωt)+O(A2)\Delta U_+(t)-\Delta U_-(t)=\frac{2Ax_s}{3}\cos(\omega t)+O(A^2). The conventional directional barrier-modulation channel therefore remains present, and noise-assisted switching can respond to the driving. Whether the chosen response measure has a pronounced maximum as DD is varied must be checked at the specified forcing and frequency. Finite reset-position uncertainty or additional noise removes the exact trapping at x=0x=0 and can permit a reset response not captured by the ideal seeding criterion.

The complementary behavior is illustrated in Figure S2. Here the noise-driven system exhibits a clear resonant maximum at a finite noise intensity, while the resetting-induced coherent response vanishes identically for all SR rates. In this case the prescribed reset state carries no phase-dependent bias, so resetting does not select a coherent outgoing response even though the underlying nonlinear system supports NIR. This example shows that SR alone is insufficient to generate resonance. A phase-sensitive post-reset response is an essential ingredient of SRIR.

Figure S2. Comparison of SRIR and NIR for two nonlinear systems (Eq. (96)). (a) For g(x)g(x) given by Eq. (108), SRIR (blue circles) exhibits a pronounced maximum as a function of SR rate rr, while NIR (red triangles) is much weaker. (b) For g(x)g(x) given by Eq. (111), NIR displays a clear maximum as a function of the noise intensity DD, whereas the SRIR response vanishes identically, G=0G=0, because resetting generates no phase-dependent bias. The two examples demonstrate that SRIR and NIR are dynamically distinct and that the existence of one does not imply the other. Parameters: A=0.3,ω=0.1,a=b=xs=1A=0.3,\omega=0.1,a=b=x_s=1.

Numerical Methods

Direct Monte Carlo simulation

Between successive resets, the noise-free bistable dynamics obey

dxdt=ax−bx3+Acos⁡(ωt).(114)\frac{dx}{dt}=ax-bx^{3}+A\cos(\omega t). \tag*{(114)}

At each Poisson reset time tnt_n, the position is returned to x(tn)=0x(t_n)=0, while the external drive phase remains unchanged.

The probability of at least one SR event during a numerical step Δt\Delta t is

pr=1−e−rΔt.(115)p_r=1-e^{-r\Delta t}. \tag*{(115)}

Equivalently, continuous Poisson event times can be generated from tn+1−tn=−ln⁡(Un/r)t_{n+1}-t_n=-\ln(U_n/r), where the UnU_n are independent uniform random variables on (0,1)(0,1).

The response is measured over an integer number of driving periods, Tobs=NperTT_{\mathrm{obs}}=N_{\mathrm{per}}T, where T=2π/ωT=2\pi/\omega. For realization jj, we define

Qc(j)=2Tobs∫t0t0+Tobsxj(t)cos⁡(ωt) dt,Q_c^{(j)}=\frac{2}{T_{\mathrm{obs}}}\int_{t_0}^{t_0+T_{\mathrm{obs}}}x_j(t)\cos(\omega t)\,\mathrm{d}t,
Qs(j)=2Tobs∫t0t0+Tobsxj(t)sin⁡(ωt) dt.(116)Q_s^{(j)}=\frac{2}{T_{\mathrm{obs}}}\int_{t_0}^{t_0+T_{\mathrm{obs}}}x_j(t)\sin(\omega t)\,\mathrm{d}t. \tag*{(116)}

For NensN_{\mathrm{ens}} independent realizations, the ensemble-averaged components are Q‾α=Nens−1∑j=1NensQα(j)\overline{Q}_{\alpha}=N_{\mathrm{ens}}^{-1}\sum_{j=1}^{N_{\mathrm{ens}}}Q_{\alpha}^{(j)}, with α={c,s}\alpha=\{c,s\}. Thus, we set

C1=Q‾c−iQ‾s2,G(r)=Q‾c 2+Q‾s 2A.(117)C_1=\frac{\overline{Q}_c-i\overline{Q}_s}{2},\qquad G(r)=\frac{\sqrt{\overline{Q}_c^{\,2}+\overline{Q}_s^{\,2}}}{A}. \tag*{(117)}

The Fourier components are therefore ensemble averaged before their magnitude is taken.

For the three-stable-state system (S58), the periodic response is extracted from the complex position Zj(t)=xj(t)+iyj(t)Z_j(t)=x_j(t)+iy_j(t). For each realization, we calculate the Fourier coefficient

C1(j)=1Tobs∫t0t0+TobsZj(t)e−iωt dt.(118)C_1^{(j)}=\frac{1}{T_{\mathrm{obs}}}\int_{t_0}^{t_0+T_{\mathrm{obs}}}Z_j(t)e^{-i\omega t}\,\mathrm{d}t. \tag*{(118)}

The ensemble-averaged coefficient and the corresponding amplification factor are

C1=Nens−1∑j=1NensC1(j),G(r)=∣C1∣A.(119)C_1=N_{\mathrm{ens}}^{-1}\sum_{j=1}^{N_{\mathrm{ens}}}C_1^{(j)},\qquad G(r)=\frac{\lvert C_1\rvert}{A}. \tag*{(119)}

Post-reset trajectories

The renewal and coarse-grained reduction use the same uninterrupted post-reset trajectories. The reset phase is

sampled on the midpoint grid

ϕj=2π(j+1/2)Nϕ,j={0,…,Nϕ−1}.(120)\phi_j=\frac{2\pi(j+1/2)}{N_\phi},\qquad j=\{0,\ldots,N_\phi-1\}. \tag*{(120)}

For each ϕj\phi_j, the trajectory is obtained by integrating

dxϕjdτ=axϕj−bxϕj3+Acos⁡(ϕj+ωτ),(121)\frac{dx_{\phi_j}}{d\tau}=ax_{\phi_j}-bx_{\phi_j}^{3}+A\cos(\phi_j+\omega\tau), \tag*{(121)}
xϕj(0)=0,(122)x_{\phi_j}(0)=0, \tag*{(122)}

with a fourth-order Runge–Kutta scheme. Here τ\tau is the elapsed time since the reset, and no further reset is applied while computing xϕj(τ)x_{\phi_j}(\tau).

The first phase harmonic is

h1(τ)=12π∫02πxϕ(τ)e−iϕ dϕ∼1Nϕ∑j=0Nϕ−1xϕj(τ)e−iϕj.(123)\begin{aligned} h_1(\tau)=\frac{1}{2\pi}\int_{0}^{2\pi}x_{\phi}(\tau)e^{-i\phi}\,\mathrm{d}\phi\\ &\sim\frac{1}{N_\phi}\sum_{j=0}^{N_\phi-1}x_{\phi_j}(\tau)e^{-i\phi_j}. \tag*{(123)} \end{aligned}

The exact renewal representation then gives

C1(r)=r∫0∞e−(r+iω)τh1(τ) dτ,C_1(r)=r\int_{0}^{\infty}e^{-(r+i\omega)\tau}h_1(\tau)\,\mathrm{d}\tau,
G(r)=2∣C1(r)∣A.(124)G(r)=\frac{2\lvert C_1(r)\rvert}{A}. \tag*{(124)}

For numerical evaluation, the age integral is truncated at τmax⁡\tau_{\max} and discretized as

C1(r)∼r∑k=0Nτwke−(r+iω)τkh1(τk),(125)C_1(r)\sim r\sum_{k=0}^{N_\tau}w_ke^{-(r+i\omega)\tau_k}h_1(\tau_k), \tag*{(125)}

where τk=kΔτ\tau_k=k\Delta\tau, τmax⁡=NτΔτ\tau_{\max}=N_\tau\Delta\tau, and wkw_k are the quadrature weights.

For the bistable realization, the complete trajectory is replaced by

xϕ(τ)=σϕxsΘ[τ−τs(ϕ)],xs=a/b,(126)x_{\phi}(\tau)=\sigma_{\phi}x_s\Theta[\tau-\tau_s(\phi)],\qquad x_s=\sqrt{a/b}, \tag*{(126)}

where

τs(ϕ)=inf⁡{τ>0:∣xϕ(τ)∣=xs},\tau_s(\phi)=\inf\{\tau>0:\lvert x_{\phi}(\tau)\rvert=x_s\},
σϕ=sgn⁡[xϕ(τs(ϕ))].(127)\sigma_{\phi}=\operatorname{sgn}[x_{\phi}(\tau_s(\phi))]. \tag*{(127)}

If the first threshold crossing occurs between τk\tau_k and τk+1\tau_{k+1}, its time is refined by linear interpolation

τs(ϕj)∼τk+Δτσϕjxs−xϕj(τk)xϕj(τk+1)−xϕj(τk).(128)\tau_s(\phi_j)\sim\tau_k+\Delta\tau\frac{\sigma_{\phi_j}x_s-x_{\phi_j}(\tau_k)}{x_{\phi_j}(\tau_{k+1})-x_{\phi_j}(\tau_k)}. \tag*{(128)}

with σϕj=sgn⁡[xϕj(τk+1)]\sigma_{\phi_j}=\operatorname{sgn}[x_{\phi_j}(\tau_{k+1})]. Substitution into the renewal formula yields

C1(r)=xsr2π(r+iω)∫02πσϕe−iϕ−(r+iω)τs(ϕ) dϕ.(129)C_1(r)=\frac{x_sr}{2\pi(r+i\omega)}\int_{0}^{2\pi}\sigma_{\phi}e^{-i\phi-(r+i\omega)\tau_s(\phi)}\,\mathrm{d}\phi. \tag*{(129)}

Consequently, the phase-grid approximation to the reduced gain is

G(r)∼2xsrAr2+ω2∣1Nϕ∑j=0Nϕ−1σϕje−iϕj−(r+iω)τs(ϕj)∣.(130)G(r) \sim\frac{2x_{\mathrm{s}}r}{A\sqrt{r^2+\omega^2}}\left|\frac{1}{N_{\phi}}\sum_{j=0}^{N_{\phi}-1}\sigma_{\phi_j}e^{-i\phi_j-(r+i\omega)\tau_{\mathrm{s}}(\phi_j)}\right|. \tag*{(130)}

The SR rate is varied without recomputing the uninterrupted trajectories. Scanning and maximization give the optimal rates for the renewal and give rise to reduced response.

For the three-stable-state system (S58), the two-dimensional dynamics can be expressed in terms of the complex position Z=x+iyZ=x+iy as

dZdt=Z−∣Z∣2Z+3ϵZ‾ 2+Aeiωt,(131)\frac{dZ}{dt}=Z-|Z|^2Z+3\epsilon\overline{Z}^{\,2}+Ae^{i\omega t}, \tag*{(131)}

with Poisson SR to Z=0Z=0.

The renewal formulation applies directly to the complex post-reset trajectories Zϕ(τ)Z_{\phi}(\tau). The first-harmonic coefficient and amplification factor are given by

h1(τ)=12π∫02πZϕ(τ)e−iϕ dϕ,(132)h_1(\tau)=\frac{1}{2\pi}\int_{0}^{2\pi}Z_{\phi}(\tau)e^{-i\phi}\,d\phi, \tag*{(132)}
C1(r)=r∫0∞e−(r+iω)τh1(τ) dτ,(133)C_1(r)=r\int_{0}^{\infty}e^{-(r+i\omega)\tau}h_1(\tau)\,d\tau, \tag*{(133)}
G(r)=∣C1(r)∣A.(134)G(r)=\frac{\lvert C_1(r)\rvert}{A}. \tag*{(134)}

Time-Periodic Fokker–Planck Calculations for NIR

For comparison with conventional NIR, resetting is removed and additive Gaussian white noise is included [1]

dxdt=ax−bx3+Acos⁡(ωt)+2D ξ(t),(135)\frac{dx}{dt}=ax-bx^3+A\cos(\omega t)+\sqrt{2D}\,\xi(t), \tag*{(135)}

where ⟨ξ(t)⟩=0\langle\xi(t)\rangle=0 and ⟨ξ(t)ξ(t′)⟩=δ(t−t′)\langle\xi(t)\xi(t')\rangle=\delta(t-t'). The probability density satisfies [3]

∂P(x,t)∂t=−∂[(ax−bx3+Acos⁡ωt)P(x,t)]∂x+D∂2P(x,t)∂x2.(136)\frac{\partial P(x,t)}{\partial t}=-\frac{\partial\left[(ax-bx^3+A\cos\omega t)P(x,t)\right]}{\partial x}+D\frac{\partial^2P(x,t)}{\partial x^2}. \tag*{(136)}

The periodic density is approximated by the truncated harmonic expansion

P(x,t)∼∑n=−NhNhpn(x)einωt,(137)P(x,t)\sim\sum_{n=-N_{\mathrm{h}}}^{N_{\mathrm{h}}}p_n(x)e^{in\omega t}, \tag*{(137)}

Here, p−n(x)=pn∗(x)p_{-n}(x)=p_n^*(x), NhN_{\mathrm{h}} denotes the Fourier truncation order corresponding to the highest harmonic retained in the expansion, and the superscript (∗)(*) denotes complex conjugation.

Substituting Eq. (137) into Eq. (136) one can yield

inωpn=−∂Jn∂x(138)in\omega p_n=-\frac{\partial J_n}{\partial x} \tag*{(138)}

where

Jn=(ax−bx3)pn+A2(pn−1+pn+1)−D∂pn∂x.(139)J_n=(ax-bx^3)p_n+\frac{A}{2}(p_{n-1}+p_{n+1})-D\frac{\partial p_n}{\partial x}. \tag*{(139)}

The harmonic truncation is given by pNh+1=p−Nh−1=0p_{N_{\mathrm{h}}+1}=p_{-N_{\mathrm{h}}-1}=0.

On the computational domain [xL,xR][x_L,x_R], zero-flux boundary conditions and zero-mode normalization are imposed via

Jn(xL)=Jn(xR)=0,J_n(x_L)=J_n(x_R)=0,
∫xLxRp0(x) dx=1.(140)\int_{x_L}^{x_R}p_0(x)\,dx=1. \tag*{(140)}

The second condition obeys as for n≠0n\ne0, integrating (S105) over the domain [xL,xR][x_L,x_R] together with the zero-flux boundary conditions gives ∫xLxRpn(x) dx=0\int_{x_L}^{x_R}p_n(x)\,dx=0. Thus, only the zero-frequency mode carries the total probability.

The spatial discretization employs a conservative finite-volume scheme. We divide the computational domain [xL,xR][x_L,x_R] into spatial cells [xi−1/2,xi+1/2][x_{i-1/2},x_{i+1/2}] of width Δxi=xi+1/2−xi−1/2\Delta x_i=x_{i+1/2}-x_{i-1/2}. The cell-averaged Fourier coefficient is defined as

p‾n,i=1Δxi∫xi−1/2xi+1/2pn(x) dx.(141)\overline{p}_{n,i}=\frac{1}{\Delta x_i}\int_{x_{i-1/2}}^{x_{i+1/2}}p_n(x)\,dx. \tag*{(141)}

Integrating the harmonic conservation equation (S105) over each cell yields

inωp‾n,i=−Jn,i+1/2−Jn,i−1/2Δxi,(142)in\omega\overline{p}_{n,i}=-\frac{J_{n,i+1/2}-J_{n,i-1/2}}{\Delta x_i}, \tag*{(142)}

where Jn,i±1/2J_{n,i\pm1/2} denote the numerical probability fluxes through the cell interfaces, approximating the continuous harmonic current Jn(xi±1/2)J_n(x_{i\pm1/2}) defined in Eq. (139). The same interface flux is shared by adjacent cells, so that the internal flux contributions cancel when the discrete equations are summed over the computational domain. The coupled harmonic equations (S109) form a linear system for the cell-averaged Fourier coefficients p‾n,i\overline{p}_{n,i}. Solving this system under zero-flux boundary conditions Jn,1/2=Jn,Nx+1/2=0J_{n,1/2}=J_{n,N_x+1/2}=0 and the normalization constraint ∑ip‾0,iΔxi=1\sum_i\overline{p}_{0,i}\Delta x_i=1 yields the Fourier coefficients of the time-periodic stationary probability density.

The first-harmonic coefficient of the mean response is obtained from

C1,noise=∫−∞∞xp1(x) dx∼∑ixip‾1,iΔxi.(143)C_{1,\mathrm{noise}}=\int_{-\infty}^{\infty}xp_1(x)\,dx\sim\sum_i x_i\overline{p}_{1,i}\Delta x_i. \tag*{(143)}

where xix_i denotes the cell center and the numerical sum extends over the computational domain. The corresponding amplification factor is

Gnoise(D;A,ω)=2∣C1,noise∣A.(144)G_{\mathrm{noise}}(D; A, \omega) = \frac{2|C_{1,\mathrm{noise}}|}{A}. \tag*{(144)}

For fixed driving amplitude AA and frequency ω\omega, the maximum amplification is obtained by optimizing over the noise intensity DD,

Gnoisemax⁡(A,ω)=max⁡D>0Gnoise(D;A,ω).(145)G_{\mathrm{noise}}^{\max}(A, \omega) = \max_{D>0} G_{\mathrm{noise}}(D; A, \omega). \tag*{(145)}

This approach directly solves the time-periodic Fokker–Planck equation without invoking a Kramers-rate or two-state approximation. This provide the results of NIR in Figure 4 of the main text.

Dynamical Visualization of SRIR

Supplemental Movies S1 and S2 illustrate the SRIR in the bistable and tristable systems discussed in Sec. 2.2 and Sec. 2.3, respectively.

Supplemental Movie S1: Bistable system

The upper panel displays the periodically driven double-well potential and particle positions, while the middle panel compares an individual trajectory with the ensemble-averaged response. The lower panel shows the amplification factor G(r)G(r) as a function of the resetting rate.

In Movie S1, particles are reset to the origin (x=0x = 0), an unstable state of the undriven bistable system. For rare SR (r=0.01r = 0.01), the individual particle (white sphere) remains predominantly trapped in one of the stable wells, while the ensemble-averaged position (yellow sphere) exhibits only weak periodic oscillations. At the optimal SR rate (r≈0.178r \approx0.178), the individual particle undergoes irregular transitions between the two wells following SR, whereas the ensemble-averaged position exhibits pronounced, resonant oscillations. For frequent SR (r=1r = 1), repeated returns to the unstable point interrupt relaxation toward the stable wells, confining the individual particle closer to the origin and strongly suppressing the ensemble-averaged oscillations.

Supplemental Movie S2: Tristable system

The upper panel presents a three-dimensional rendering of the time-dependent potential landscape, illustrating the three stable attractors and the evolution of particle positions, with different colors distinguishing the three attractors. The lower-left panels display the xx and yy components of an individual trajectory and the ensemble-averaged response, while the lower-right panel shows the corresponding amplification factor G(r)G(r).

In Movie S2, particles are reset to the origin (x=y=0x = y = 0), an unstable state of the undriven system. For rare SR (r=0.01r = 0.01), the individual particle (white sphere) spends long periods close to a single attractor, whereas the ensemble-averaged position (yellow sphere) shows only a weak response to the rotating periodic drive. At the optimal SR rate (r≈0.185r \approx0.185), the individual particle is randomly redirected among the three attractors through SR, while the ensemble average develops periodic large-scale oscillations in both spatial components, following the rotating drive. At high SR rates (r=1r = 1), frequent reinitialization prevents the particle from fully approaching the selected attractors, resulting in a substantially reduced ensemble response. The movie demonstrates that SRIR extends beyond bistability, producing resonant two-dimensional dynamics even though individual trajectories remain stochastic.

References

  1. [1]M. R. Evans, S. N. Majumdar, and G. Schehr, Stochastic resetting and applications, J. Phys. A: Math. Theor. 53, 193001 (2020).
  2. [2]A. Pal, S. Kostinski, and S. Reuveni, The inspection paradox in stochastic resetting, J. Phys. A: Math. Theor. 55, 021001 (2022).DOI
  3. [3]S. Gupta and A. M. Jayannavar, Stochastic resetting: A (very) brief review, Front. Phys. 10, 789097 (2022).DOI
  4. [4]M. R. Evans and S. N. Majumdar, Diffusion with stochastic resetting, Phys. Rev. Lett. 106, 160601 (2011).DOI
  5. [5]S. N. Majumdar, S. Sabhapandit, and G. Schehr, Dynamical transition in the temporal relaxation of stochastic processes under resetting, Phys. Rev. E 91, 052131 (2015).
  6. [6]Y. Liang, Q. Wei, W. Wang, and A. G. Cherstvy, Ultra-slow diffusion processes under stochastic resetting, Phys. Fluids 37, 032014 (2025).
  7. [7]T. Sandev, V. Domazetovski, L. Kocarev, R. Metzler, and A. Chechkin, Heterogeneous diffusion with stochastic resetting, J. Phys. A: Math. Theor. 55, 074003 (2022).DOI
  8. [8]I. M. Sokolov, Linear response and fluctuation-dissipation relations for Brownian motion under resetting, Phys. Rev. Lett. 130, 067101 (2023).DOI
  9. [9]V. Stojkoski, T. Sandev, L. Kocarev, and A. Pal, Autocorrelation functions and ergodicity in diffusion with stochastic resetting, J. Phys. A: Math. Theor. 55 104003 (2022).DOI
  10. [10]E. Barkai, R. Flaquer-Galmes, and V. Méndez, Ergodic properties of Brownian motion under stochastic resetting, Phys. Rev. E 108, 064102 (2023).arxiv.org/abs/2306.13621
  11. [11]W. Wang, A. G. Cherstvy, R. Metzler, and I. M. Sokolov, Restoring ergodicity of stochastically reset anomalous-diffusion processes, Phys. Rev. Res. 4, 013161 (2022).DOI
  12. [12]S. Reuveni, Optimal stochastic restart renders fluctuations in first passage times universal, Phys. Rev. Lett. 116, 170601 (2016).DOI
  13. [13]A. Pal and S. Reuveni, First passage under restart, Phys. Rev. Lett. 118, 030603 (2017).DOI
  14. [14]A. V. Chechkin and I. M. Sokolov, Random search with resetting: A unified renewal approach, Phys. Rev. Lett. 121, 050601 (2018).DOI
  15. [15]A. Pal, I. Eliazar, and S. Reuveni, First passage under restart with branching, Phys. Rev. Lett. 122, 020602 (2019).DOI
  16. [16]B. De Bruyne, J. Randon-Furling, and S. Redner, Optimization in first-passage resetting, Phys. Rev. Lett. 125, 050602 (2020).DOI
  17. [17]S. Reuveni, M. Urbach, and J. Klafter, Role of substrate unbinding in Michaelis-Menten enzymatic reactions, Proc. Natl. Acad. Sci. USA 111, 4391 (2014).DOI
  18. [18]O. Tal-Friedman, A. Pal, A. Sekhon, S. Reuveni, and Y. Roichman, Experimental realization of diffusion with stochastic resetting, J. Phys. Chem. Lett. 11, 7350 (2020).
  19. [19]B. Besga, A. Bovon, A. Petrosyan, S. N. Majumdar, and S. Ciliberto, Optimal mean first-passage time for a Brownian searcher subjected to resetting: Experimental and theoretical results, Phys. Rev. Res. 2, 032029(R) (2020).DOI
  20. [20]F. Faisant, B. Besga, A. Petrosyan, S. Ciliberto, and S. N. Majumdar, Optimal mean first-passage time of a Brownian searcher with resetting in one and two dimensions: Experiments, theory and numerical tests, J. Stat. Mech. 11, 113203 (2021).DOI
  21. [21]R. Vatash and Y. Roichman, Many-body colloidal dynamics under stochastic resetting: Competing effects of particle interactions on the steady-state distribution, Phys. Rev. Res. 7, L032020 (2025).arxiv.org/abs/2504.10015
  22. [22]R. Goerlich, T. D. Keidar, and Y. Roichman, Resetting as a swift equilibration protocol in an anharmonic potential, Phys. Rev. Res. 6, 033162 (2024).arxiv.org/abs/2401.09958
  23. [23]K. S. Olsen, H. Löwen, and L. Caprini, Optimal area exploration by resetting active particles, Phys. Rev. Lett. 137, 057102 (2026).DOI
  24. [24]K. S. Olsen, D. Gupta, F. Mori, and S. Krishnamurthy, Thermodynamic cost of finite-time stochastic resetting, Phys. Rev. Res. 6, 033343 (2024).DOI
  25. [25]D. Gupta, C. A. Plata, and A. Pal, Work fluctuations and Jarzynski equality in stochastic resetting, Phys. Rev. Lett. 124, 110608 (2020).DOI
  26. [26]J. M. Meylahn, S. Sabhapandit, and H. Touchette, Large deviations for Markov processes with resetting, Phys. Rev. E 92, 062148 (2015).DOI
  27. [27]F. H. Gonzalez, A. P. Riascos, and D. Boyer, Diffusive transport on networks with stochastic resetting to multiple nodes, Phys. Rev. E 103, 062126 (2021).DOI
  28. [28]B. McNamara and K. Wiesenfeld, Theory of stochastic resonance, Phys. Rev. A 39, 4854 (1989).pubmed.ncbi.nlm.nih.gov/9901841
  29. [29]L. Gammaitoni, P. Hänggi, P. Jung, and F. Marchesoni, Stochastic resonance, Rev. Mod. Phys. 70, 223 (1998).
  30. [30]T. Wellens, V. Shatokhin, and A. Buchleitner, Stochastic resonance, Rep. Prog. Phys. 67, 45 (2004).
  31. [31]A. Pikovsky, M. Rosenblum, and J. Kurths, Synchronization: A Universal Concept in Nonlinear Sciences (Cambridge University Press, Cambridge, 2001).
  32. [32]R. Benzi, A. Sutera, and A. Vulpiani, The mechanism of stochastic resonance, J. Phys. A: Math. Gen. 14, L453 (1981).DOI
  33. [33]R. Benzi, G. Parisi, A. Sutera, and A. Vulpiani, Stochastic resonance in climatic change, Tellus 34, 10 (1982).
  34. [34]S. Fauve and F. Heslot, Stochastic resonance in a bistable system, Phys. Lett. A 97, 5 (1983).DOI
  35. [35]B. McNamara, K. Wiesenfeld, and R. Roy, Observation of stochastic resonance in a ring laser, Phys. Rev. Lett. 60, 2626 (1988).
  36. [36]R. N. Mantegna and B. Spagnolo, Stochastic resonance in a tunnel diode, Phys. Rev. E 49, R1792 (1994).pubmed.ncbi.nlm.nih.gov/9961504
  37. [37]J. E. Levin and J. P. Miller, Broadband neural encoding in the cricket cercal sensory system enhanced by stochastic resonance, Nature 380, 165 (1996).pubmed.ncbi.nlm.nih.gov/8600392
  38. [38]J. K. Douglass, L. Wilkens, E. Pantazelou, and F. Moss, Noise enhancement of information transfer in crayfish mechanoreceptors by stochastic resonance, Nature 365, 337 (1993).pubmed.ncbi.nlm.nih.gov/8377824
  39. [39]A. Pikovsky and J. Kurths, Coherence resonance in a noise-driven excitable system, Phys. Rev. Lett. 78, 775 (1997).
  40. [40]G. Hu, T. Ditzinger, C. Z. Ning, and H. Haken, Stochastic resonance without external periodic force, Phys. Rev. Lett. 71, 807 (1993).
  41. [41]A. Longtin, Autonomous stochastic resonance in bursting neurons, Phys. Rev. E 55, 868 (1997).DOI
  42. [42]T. M. Lenton, H. Held, E. Krieger, J. W. Hall, W. Lucht, S. Rahmstorf, and H. J. Schellnhuber, Tipping elements in the Earth’s climate system, Proc. Natl. Acad. Sci. USA 105, 1786 (2008).
  43. [43]R. Brette and W. Gerstner, Adaptive exponential integrate-and-fire model as an effective description of neuronal activity, J. Neurophysiol. 94, 3637 (2005).
  44. [44]D. Vinod, A. G. Chervstov, W. Wang, R. Metzler, and I. M. Sokolov, Nonergodicity of reset geometric Brownian motion, Phys. Rev. E 105, L012106 (2022).DOI
  45. [45]V. Stojkoski, T. Sandev, L. Kocarev, and A. Pal, Geometric Brownian motion under stochastic resetting: A stationary yet nonergodic process, Phys. Rev. E 104, 014121 (2021).DOI
  46. [46]The threshold follows from the saddle-node conditions \(a x_c - b x_c^3 + F_c = 0\) and \(a - 3 b x_c^2 = 0\), yielding \(A_c = |F_c| = 2 a^{3/2}/(3\sqrt{3b})\).
  47. [47]In the long-time periodic state, the initial drive phase \(\phi_0\) only multiplies the \(n\)th harmonic coefficient by the phase factor \(e^{in\phi_0}\) and therefore does not affect its amplitude \(|C_n(r)|\). Hence, without a loss of generality, we set \(\phi_0 = 0\).
  48. [48]See Supplemental Material for the general renewal formulation, coarse-grained reduction, optimal-rate scaling beyond bistability, selective control of SRIR and NIR, numerical details, and Movies S1 and S2 illustrating the reset-induced dynamics in bistable and tristable systems, respectively.
  49. [49]J. Zhu, F. Zhao, Y. Li, and X. Liu, Rotational stochastic resonance in multistable systems, Physica A 634, 129473 (2024).DOI
  50. [50]H. Risken, The Fokker–Planck Equation: Methods of Solution and Applications, 2nd ed., Springer Series in Synergetics Vol. 18 (Springer-Verlag, Berlin, 1989).
  51. [51]H. A. Kramers, Brownian motion in a field of force and the diffusion model of chemical reactions, Physica 7, 284 (1940).DOI
  52. [52]P. Hänggi, P. Talkner, and M. Borkovec, Reaction-rate theory: Fifty years after Kramers, Rev. Mod. Phys. 62, 251 (1990).
  53. [53]L. Gammaitoni, F. Marchesoni, E. Menichella-Saetta, and S. Santucci, Stochastic resonance in bistable systems, Phys. Rev. Lett. 62, 349 (1989).pubmed.ncbi.nlm.nih.gov/10040210
  54. [54]P. Jop, A. Petrosyan, and S. Ciliberto, Work and dissipation fluctuations near the stochastic resonance of a colloidal particle, Europhys. Lett. 81, 50005 (2008).DOI
  55. [55]D. Bhowmik, L. You, and S. Salahuddin, Spin Hall effect clocking of nanomagnetic logic without a magnetic field, Nat. Nanotechnol. 9, 59 (2014).DOI
  56. [56]P. Jolakoski, L. Basnarkov, L. Kocarev, A. Popovska-Mitrovikj, V. Bakeva, and T. Sandev, Response to an external field of a generalized Langevin equation with stochastic resetting of the memory kernel, Phys. Rev. E 112, 024120 (2025).arxiv.org/abs/2504.06687
  57. [57]C. Di Bello, A. V. V. Chechkin, A. K. Hartmann, Z. Palmowski, and R. Metzler, Time-dependent probability density function for partial resetting dynamics, New J. Phys. 25, 082002 (2023).arxiv.org/abs/2305.13722
  58. [58]A. S. Bodrova, A. V. Chechkin, and I. M. Sokolov, Scaled Brownian motion with renewal resetting, Phys. Rev. E 100, 012120 (2019).DOI
  59. [59]D. Gupta, Stochastic resetting in underdamped Brownian motion, J. Stat. Mech. 3 033212 (2019).DOI
  60. [60]H. Franke and I. M. Sokolov, Resetting of underdamped Brownian motion, Phys. Rev. E 113, 064125 (2026).DOI
  61. [61]See the numerical codes available at Zenodo, https://doi.org/10.5281/zenodo.23798786.

Paper details

vlnc:W_94vencrr
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents