Abstract
In this study, a bifurcation analysis of the piezoelectric neuron system (PNS) is conducted utilizing the timedomain minimum residual method (TMRM). From a biophysical standpoint, integrating piezoelectric devices into a nonlinear neuron circuit enables the conversion of external pressure or sound wave signals into electrical signals. The primary content of this research are as follows: initially, the piezoelectric neuron system is analytically solved by employing the TMRM, yielding highprecision periodic solutions. Subsequently, a detailed bifurcation analysis is performed leveraging these highprecision solutions. The findings demonstrate that the neuronal system experiences a periodicdoubling bifurcation when the amplitude of external stimuli surpasses a specific threshold. With further increases in the stimulus, this periodicdoubling response evolves into a periodic response. Moreover, changes in cell membrane resistance will lead to complex nonlinear behaviors such as stable or unstable period1 solutions and stable or unstable and critically stable period3 solutions in the system. These findings offer a theoretical foundation for the advanced design of intelligent functional neuron arrays and sensors.
Similar content being viewed by others
Explore related subjects
Discover the latest articles, news and stories from top researchers in related subjects.Avoid common mistakes on your manuscript.
1 Introduction
The nervous system, an intricate and highly nonlinear component of biological systems, poses significant challenges for traditional neuroscience research due to limitations in experimental techniques and conditions [1]. Exploring the intricate patterns of electrical activity induction and cooperative mechanisms between neurons from a purely experimental standpoint is inherently tricky [2]. Thus, to gain insights into the physiological and cognitive behaviors of the nervous system, researchers often turn to mathematical, physical, or computer technologies for assistance [3, 4]. In recent decades, leveraging these advanced methodologies, researchers have developed numerous nonlinear neuron models based on clinical electrophysiological experiments, facilitating a deeper understanding of signal transmission within the nervous system. For instance, Hodgkin and Huxley formulated the HodgkinHuxley (HH) neuron model using neural stimulation potential data from squids in 1952 [5]. Subsequently, Fitzhugh [6] proposed the FitzHughNagumo (FHN) model by simplifying the HH model, delineating fast subsystems to represent changes in neuronal membrane potential and slow subsystems for channel inactivation. Aibara et al. [7, 8] investigated the various responses of giant squid axons under external sinusoidal currents through numerical simulations and experiments. Similarly, Hindmarsh et al. [9] drawing on extensive data from voltageclamp experiments conducted in 1984, introduced the HindmarshRose (HR) model. Notably, these neuron models can be translated into corresponding nonlinear circuits using appropriate nonlinear devices, facilitating the simulation of neural systems [10]. Integrating experimental data with mathematical modeling and technological advancements, this interdisciplinary approach has dramatically enriched our comprehension of the nervous system’s intricate dynamics [11, 12].
In the human auditory system, the sound is captured by the ear, where eardrum vibrations lead to the conversion of these vibrations into nerve signals by hair cells in the cochlea. These signals are then transmitted via the cochlear nerves and auditory pathways to the auditory cortex, resulting in sound perception [13]. Damage to hair cells caused by loud noise or drug abuse impedes the transmission of sound signals from the cochlea to the brain. Cochlear implants can bypass damaged hair cells by converting sound into electrical signals, directly stimulating the auditory nerve to restore or improve hearing [14]. However, damage to auditory nerves and neural circuits can block this conversion and signal propagation, leaving no current treatment for auditory nerve recovery. Future technologies may include artificial sensors or processors to mimic firing patterns, aiding the central auditory system [15].
Investigating how auditory neurons react to various stimuli is crucial for understanding their nonlinear dynamics and designing artificial auditory sensors. Recent biophysical research has aimed to enhance neuronal circuit functionality by incorporating physical devices into nonlinear circuits, allowing for sensing specific external signals. For instance, Liu et al. [16] explored the effects of light on neural activity patterns by integrating a photocell into the circuit. In contrast, Guo et al. [17] analyzed the synchronization stability of thermosensitive neurons. Yang et al. [18] enhanced a neuronal model by combining a photocell and thermistor to examine the influence of energy diversity on neuronal coupling. Embedding piezoelectric devices into nonlinear circuits has led to the design of auditory circuits sensitive to sound waves [19]. The PNS model using piezoelectric ceramics for acousticelectric conversion can simulate cochlear nerve excitation and has potential applications in restoring hearing [15]. This paper focuses on the bifurcation phenomenon of the PNS for external stimulus intensity and resistance parameters.
Over recent decades, extensive research on neuronal model dynamics has emerged, highlighting the significance of numerical methods for their computational efficiency, versatility, and robustness in analyzing neuronal systems’ dynamic behaviors [18, 20]. Standard numerical integration methods like the Newmark\(\beta \) method [21], Runge–Kutta (RK) method and finite difference method apply time integration to the differential equations, providing discrete numerical solutions that may not fully capture the system’s dynamics. Selecting inappropriate time steps can result in inaccurate dynamic analyses [22]. Moreover, nonlinear dynamics can exhibit complex behaviors such as multiple solutions, jumps, bifurcations, and instability, complicating quantitative analyses using numerical methods [23]. Analytical methods offer a distinct advantage by providing exact solution expressions, accurately depicting the system’s longterm behavior, and identifying stable intervals and bifurcations, thus deeply reflecting the nonlinear system’s inherent characteristics. However, obtaining simple analytical solutions for nonlinear problems is rare, with most solutions being approximate or represented as finite series [24]. As a result, analytical and semianalytical methods, including the harmonic balance method [25] and incremental harmonic balance (IHB) method [26], have gained attention.
Recently, Liu et al. [27] presented a novel semianalytical solution approach: the timedomain minimum residual method (TMRM), which theoretically enables the derivation of approximate solutions of arbitrary precision by adjusting the truncation series. This method’s application enhances the accuracy of computational approaches for nonlinear systems [28], crucial for developing reliable neuron models and exploring discharge rhythms. This study employs the TMRM to solve the piezoelectric neuron circuit system, analyzing its complex dynamics through time history, phase, and bifurcation diagrams and the stability of piezoelectric neuron system (PNS) was analyzed using Floquet theory [29]. Findings suggest that the TMRMbased solutions closely match those from the RK method, efficiently tracking unstable solutions. Moreover, by changing the bifurcation parameters or applying external stimuli, the electrical activity in the piezoelectric neuron circuit can exhibit diverse discharge patterns, revealing the potential biophysical mechanisms of auditory neurons’ multistable response under external stimuli.
This article is structured as follows: Sect. 2 outlines the principles and basic equations governing the piezoelectric neuron system (PNS). Section 3 delves into the foundational principles of the TMRM and discusses the general methodology for solving nonlinear systems. In Sect. 4, we first describe the procedure for applying the TMRM to solve the PNS, including presenting the semianalytical solution obtained. Subsequently, bifurcation analysis of the PNS is carried out by receiving the semianalytical solution. Finally, the conclusions drawn from our study are summarized in Sect. 5.
2 The equation of the piezoelectric neuron system
As delineated previously, the human auditory system captures sound through the outer ear, vibrating the tympanic membrane in the middle ear. Subsequently, the acoustic data is transformed into nerve impulses within the inner ear’s cochlea, relayed via the cochlear nerves and neural pathways to the auditory cortex of the cerebral cortex, culminating in the perception of hearing. In artificial auditory neurons, piezoelectric devices can be integrated into neural circuits to emulate the firing patterns observed in biological neurons. Upon exposure to external mechanical forces or sound wave vibrations, piezoelectric devices undergo continuous deformation. This deformation induces a potential difference across the terminals of the piezoelectric device, as described by the following equation [15]:
where F symbolizes the external mechanical force, S refers to the crosssectional area, \(\epsilon \) represents the dielectric constant, h indicates the thickness of the piezoelectric device, Q is the charge released, and \(d_{33}\) is the piezoelectric strain constant, which is associated with the properties of the piezoelectric material.
A constant external mechanical force generates a stable voltage difference, whereas nonlinear vibration results in a timevarying voltage difference. The renowned FHN neuron model, proposed by Fitzhugh and Nagumo [6, 30] effectively describes the primary characteristics of neural firing behavior in response to suitable external stimuli. Consequently, as depicted in Fig. 1, a piezoelectric neuron circuit can be constructed by integrating a piezoelectric ceramic with the FHN neuron model. Utilizing Kirchhoff’s laws for current and voltage, the subsequent circuit equations can be derived:
where C represents the capacitance in Fig. 1, and L represents the self inductance coefficient of the coil. \(V_C\) signifies the voltage across the capacitor, and \(V_{PC}\) denotes the voltage output across the piezoelectric ceramic. \(R_s\) and R are the linear resistances, while E is the constant voltage source that mimics the reversal potential in the ion channel. \(i_L\) represents the current through the inductor, and the current–voltage relationship across the nonlinear resistor \(R_N\) is given by:
where \(\rho \) is the normalization parameter, and V and \(V_0\) denote the voltage across the nonlinear resistor and the cutoff voltage, respectively. To further simplify the above equation, the following form of dimensionless parameters can be introduced: \(x_1=\frac{V_C}{V_0}, x_2=\frac{\rho i_L}{V_0}, \tau =\frac{t}{\rho C}, u_{PC}=\frac{V_{PC}}{V_0}, \varphi =\frac{\rho }{R_s}, \alpha =\frac{E}{V_0}, \delta =\frac{R}{\rho }, \mu =\frac{\rho ^2 C}{L}\). From then on, the dimensionless dynamic equation of equivalent PNS (2) can be simplified as:
where the derivative of the state variable \(\varvec{x}=[x_1, x_2]\) in relation to the dimensionless time \(\tau \) is denoted by the dot above \(\varvec{x}\). It is widely known that the excitability of neurons will change with changes in the external stimulation current \(u_{PC}\), thereby inducing neurons to produce different discharge patterns. Under certain conditions, acoustic waves converted into electrical signals by piezoelectric ceramics can select different forms of signals, which can be a single periodic signal, a composite periodic signal, or a chaotic signal [19, 31]. In this paper, to illustrate the efficacy of the TMRM in neuronal computation and dynamics analysis succinctly, here we choose periodic signals as the electrical signal of the external stimulus, i.e. \(\varphi u_{PC}=A\cos (\Omega \tau )\). The following dynamic equations can be further obtained:
For the convenience of subsequent derivation, the above equation (5) can be further simplified into matrix form:
where \(\dot{\varvec{X}}=\begin{pmatrix}{\dot{x}_1} \\ {\dot{x}_2}\\ \end{pmatrix}, \varvec{X}=\begin{pmatrix}{{x}_1}\\ {{x}_2}\\ \end{pmatrix}, \varvec{C}=\begin{pmatrix}1&{}0\\ 0&{}1\\ \end{pmatrix}, \varvec{K}=\begin{pmatrix}\varphi 1&{}1\\  \mu &{}\delta \mu \\ \end{pmatrix}, \varvec{N}(\dot{\varvec{X}},\varvec{X})=\begin{pmatrix}\frac{1}{3}x_1^3\\  \alpha \mu \\ \end{pmatrix}, \varvec{F}(\varvec{\tau })=\begin{pmatrix}A\cos (\Omega \tau )\\ 0\\ \end{pmatrix}\).
3 The process of solving piezoelectric neuron system using the TMRM
The TMRM, which blends seminumerical and semianalytical approaches, is applicable to both weak and strong nonlinear systems. The procedure for employing the TMRM to address PNS will be thoroughly explained.
3.1 Problem description
For the PNS as presented in Eq. (6), its periodic solution can be expressed as an infinite series in the following form:
where \(x_{i}(\tau )\) denotes the ith degreeoffreedom (DOF) for \(i=1\) to 2, \(a_{i0}\) is a constant term, the coefficients for the cosine and sine terms are denoted by \(b_{ik}\) and \(c_{ik}\), respectively, while \(\Omega _k\) indicates the kth order harmonic’s frequency. Typically, when the system is subject to an external excitation and the excitation frequency \(\Omega \) is near the system’s natural frequency (\(\Omega \approx \omega _0\)), the system will exhibit a primary harmonic response. Should the external force’s frequency approach n multiples of the fundamental frequency (\(\Omega \approx n\omega _0\)), given that n is a positive integer distinct from 1), the system may also exhibit a subharmonic response at a frequency of \(\Omega /n\). This relationship is captured as:
Differentiating Eq. (7) with respect to dimensionless time \(\tau \) yields the firstorder differential form of the response:
Considering computational practicality, the solution can be approximated by truncating the first N terms in Eq. (7), resulting in:
This series solution, as approximated in Eq. (10), introduces a residual when substituted back into Eq. (7):
while this approximate solution does not ensure the residual \(\varvec{R}\) vanishes across the entire time domain, selecting an appropriate N and harmonic coefficients \(\varvec{a}=[a_{i0}, b_{ik}, c_{ik}], i=1,2; k=1,2,\cdots ,N\) can minimize \(\varvec{R}\) as much as possible. Therefore, solving for a semianalytical solution to the nonlinear system as outlined in Eq. (7) becomes a quest to determine harmonic coefficients \(\varvec{a}\) that reduce the residual \(\varvec{R}=\varvec{R}(\varvec{a}, \tau )\) to its minimum across the time span \(\tau _j\in [0,\tau _1,\tau _2,\dots , T]\), defined as:
where \(\hbar (\varvec{a},\tau )\) represents the nonlinear objective function and M indicates the allowable range for the harmonic coefficients \(\varvec{a}\), and \(\varvec{R}(\varvec{a}, \tau )\) is the system’s residual vector.
3.2 Unknown harmonic coefficients are solved by iteration
For the nonlinear leastsquares optimization problem described as Eq. (12), an iterative method such as the enhanced response sensitivity approach is typically employed to solve it. This process involves initially selecting a suitable set of initial iteration values, denoted as \(\varvec{a}^{(0)}\). Subsequently, the iteration update \(\Delta \varvec{a}^{(j)}\) is calculated in a manner that minimizes \(\hbar (\bar{\varvec{a}}+\Delta \varvec{a})\) based on the current parameter values \(\bar{\varvec{a}}\in M\). The iteration proceeds according to the following methodology:
In accordance with the coefficient traversal principle, all unknown coefficients must be involved in the iterative process to incrementally correct and ultimately approximate the exact solution. Directly solving the original nonlinear objective function, as indicated by (12), presents challenges. Thus, the nonlinear objective function is linearized at discrete time nodes \(\tau _j\in [0,\tau _1,\tau _2,\dots , T]\). This process transforms the original nonlinear objective function \(\hbar (\bar{\varvec{a}},\tau )\) into an approximate linear objective function \(\hat{\hbar }(\bar{\varvec{a}}+\Delta \varvec{a}, \tau )\), delineated as follows:
where \(\cdot ^2\) signifies the \(l^2\)norm. The infinite series in Eq. (7) across the entire time domain aims to render the objective function \(\hbar (\varvec{a},\tau )\) equal to zero, hence \(\Delta \varvec{R}(\bar{\varvec{a}},\tau ):=0\varvec{R}(\bar{\varvec{a}},\tau )\). \(\varvec{S}(\bar{\varvec{a}},\tau )\) is the firstorder response sensitivity matrix, expressed as:
Due to the nonlinearity and discretization of the system equation, the iteration process may encounter illposed problems, such as having more unknowns than data points, potentially leading to an illposed approximate objective function \(\hat{\hbar }(\bar{\varvec{a}}+\Delta \varvec{a})\). To address possible illposed issues, a Tikhonov regularization process is introduced, modifying Eq. (14) as follows:
where \(\varvec{I}\) denotes the identity matrix. The regularization parameter \(\lambda (\lambda \ge 0)\), which influences the iterative update quantity \(\Delta \varvec{a}\), can be determined using the Lcurve method [28]. The process to ascertain \(\lambda \) begins with the singular value decomposition of \(S(\bar{\varvec{a}},\tau )\):
where U and V represent orthogonal matrices, with \(\Sigma =diag(\xi _1,\xi _2,...,\xi _N)\) denoting the diagonal matrix whose diagonal elements are arranged in descending order: \(\xi _1\ge \xi _2\ge \dots \ge \xi _N\). The Lcurve technique is employed for identifying the best regularization parameters by plotting the norm of residuals \(\zeta (\lambda )=\Delta \varvec{R}(\bar{a},\tau )\varvec{S}(\bar{\varvec{a}},\tau )\Delta \varvec{a}^2\) against the norm of the iterative update quantity \(\Psi (\lambda )=\Delta \varvec{a}^2\). A suitable regularization parameter \(\lambda \) balances these two norms effectively. The expressions for \(\zeta (\lambda )\) and \(\Psi (\lambda )\) are given by:
The optimal regularization parameter \(\lambda \) is determined at the point of maximum curvature on the Lcurve, denoted as \(\varphi _{max}=\varphi (\lambda )\). The maximum curvature is calculated using:
where \(\Psi ^{\prime }\) denotes the derivative of \(\Psi \). Following these steps, the suitable iterative update quantity \(\Delta \varvec{a}_{\lambda }\) is obtained as:
The Lcurve method primarily assesses the linearized objective function \(\hat{\hbar }(\bar{\varvec{a}}+\Delta \varvec{a}, \tau )\) without direct consideration of the original nonlinear objective function \(\hbar (\bar{\varvec{a}},\tau )\). To evaluate the appropriateness of the iterative update quantity \(\Delta \varvec{a}_{\lambda }\) the agreement indicator is introduced, defined as:
\(\varpi (\Delta \varvec{a},\bar{\varvec{a}})\ge \varpi _{cr}\in [0.25, 0.75]\) indicates a strong correspondence between the linearized and original nonlinear objective functions.
4 Results and discussion
Subsequently, the TMRM described in Sect. 3 will be applied to derive the semianalytical solution for the PNS as indicated by (6), and the stability and bifurcation analysis of the PNS will be conducted based on the obtained semianalytical solution combined with the Floquet theory.
4.1 The semianalytical solution of the PNS obtained by the TMRM
Inserting the semianalytical solution approximation from Eq. (10) into PNS (5) and rearranging yields the system’s residual equation as follows:
Using the above residual equation with the semianalytical approximation, the frequency \(\Omega _k=k\Omega \) is given by Eq. (8), leading to the determination of \(4N+2\) unknowns, namely \(\varvec{a}=[a_{10},a_{20},b_{11},b_{12},\dots ,b_{1N},c_{11},c_{12}, \dots ,c_{1N},b_{21},b_{22},\dots ,b_{2N},c_{21},c_{22},\dots ,c_{2N}]\). This sets the stage for converting the task of computing the system’s approximate analytical solution \(\varvec{X}^N=[x_1^N, x_2^N]^T\) as per (6) into a leastsquares optimization challenge. The goal is to adjust the unknown \(\varvec{a}\) to minimize the residual \(\varvec{R}=[R_1, R_2]^T\) across the time intervals \(\tau _j\in [0,\tau _1,\tau _2,\dots , T]\), as shown in Eq. (12). The sensitivity of the state variable \(\varvec{X}^N, \dot{\varvec{X}}^N\) concerning the unknown variable \(a_k\in \varvec{a}\) can be expressed as follows:
For example, the sensitivity to the constant \(a_{10}\) and the harmonic coefficients \(b_{1k}\) and \(c_{1k}\) is:
For other parameters, their sensitivity can be obtained in the same way. Since then, the sensitivity of residual \(\varvec{R}=[R_1, R_2]^T\) for unknown parameters \(a_k\in \varvec{a}\), there are:
From Eq. (25), the firstorder response sensitivity matrix \(\varvec{S} (\varvec{a},\tau )\) is derived. The parameters for the PNS as detailed in (6) are established at \(\alpha =0.7, \delta =0.8, \mu =0.1, \varphi =0.15, A=1\) and \(\Omega =2\) [15]. In the TMRM framework, initial constants are selected with \(a_{10}=1, a_{20} = 0.5\), while the first set of harmonic coefficient initial values are determined as \(b_{11}=0, c_{12}=0.5, b_{11}=0, c_{12}=0\), and zeroes are assigned as the initial values for the remaining parameters.
Firstly, Fig. 2 has shown the system time histories and phase diagrams obtained by the TMRM with truncated series \(N=10\). Figure 2a, b represent the time histories of \(x_1\) and \(x_2\), respectively, while Fig. 2c, d represent the corresponding phase diagrams. Additionally, the system’s numerical solution was computed using the fourthorder Runge–Kutta method (4th RK method, ’ode45’ function in MATLAB), and the comparison shown in the figure indicates a close agreement between the semianalytical solution derived from the TMRM and the numerical solution represented by black dots. Table 1 details the values of the harmonic coefficients for the semianalytical solution calculated through the TMRM.
Furthermore, to assess the precision of the semianalytical solution generated by the TMRM more precisely, we plot the system’s residual curve. This involves reinserting the semianalytical solutions from Table 1 into the system Eq. (6) and computing the curve of residuals. The proximity of the residual \(\varvec{R}=[R_1, R_2]^T\) to zero directly correlates with the solution’s precision. For this purpose, truncation parameters \(N=5, N=10\) and \(N=15\) were chosen for the residual curve illustration in Fig. 3a–c. Figure 3a demonstrates that with only the 5th harmonic included, the residual for the TMRM is on the order of \(10^{4}\). By including additional harmonics, the solution’s precision is enhanced further. When \(N=10\), it reaches the order of magnitude of \(10^{9}\) (displayed in Fig. 3b), while when \(N=15\), it has already reached the order of magnitude of \(10^{13}\) (displayed in Fig. 3c), which fully meets the accuracy requirements in engineering. An interesting point can also be observed in Fig. 3, where the residual of \(x_1\) (red lines) is significantly greater than that of \(x_2\) (blue lines), as the nonlinearity of the PNS (5) mainly comes from \(x_1\).
It is important to note that besides the TMRM for deriving the semianalytical solution of the PNS as outlined in (6), alternative approaches like the incremental harmonic balance (IHB) method [32] is also capable of yielding semianalytical solutions in the format presented in Eq. (10). The reason for selecting TMRM in this paper is that this method has good convergence (without the need for deliberately selecting initial values of unknown parameters), computational efficiency, and the accuracy of its semianalytical solution is not lower than other methods mentioned above. Nevertheless, we still calculated the semianalytical solution of the PNS using the IHB method under the same set of conditions (including system parameters and truncation order). Figure 3 displays the residual curves for both the TMRM and the IHB method at \(N=5, N=10\) and \(N=15\). Observations from the figure reveal that the residual from the TMRM (Fig. 3a–c) is an order of magnitude smaller compared to the IHB method’s residual (Fig. 3d–f), suggesting a superior accuracy of the semianalytical solution derived via the TMRM under the specified conditions.
4.2 Stability and bifurcation analysis
In this section, a bifurcation analysis of the PNS is performed, leveraging the semianalytical solutions derived via the TMRM in conjunction with Floquet theory. Floquet theory plays a crucial role in identifying bifurcation types and assessing the stability of periodic solutions. Illustrated in Fig. 4a, a periodic solution is considered asymptotically stable if all Floquet multipliers are within the unit circle in the complex plane. Should a single Floquet multiplier be positioned on the unit circle, while the others are within it, the solution is deemed to be critically stable. On the other hand, if all Floquet multipliers fall outside the unit circle, the solution is regarded as unstable. Furthermore, Floquet theory aids in identifying the bifurcation type of periodic solutions. A saddlenode bifurcation leading to instability arises when a Floquet multiplier traverses the point (1, 0) and moves beyond the boundary of the unit circle, as indicated by the pink arrow in Fig. 4a, while the remaining multipliers stay inside the circle. When a conjugate pair of Floquet multipliers crosses the unit circle, with the remaining multipliers staying inside, the system experiences either a Hopf or a quadratic Hopf bifurcation, illustrated by the blue arrow. Lastly, a periodicdoubling bifurcation is observed when at least one Floquet multiplier crosses through the point \((1, 0)\) with all others inside the unit circle, marked by the red arrow in Fig. 4a.
In next analysis, the system parameters are set as \(\alpha =1.1, \delta =0.8, \mu =0.3, \varphi =0.15, \Omega =0.942\), with the amplitude A of the external excitation serving as the bifurcation parameter for examining the PNS’s bifurcation behavior. Figure 4b illustrates the variation of the Floquet multiplier in relation to A within the range \(A\in [0.5, 2]\). As evident from Fig. 4b, for \(A\in [0.5, 0.94)\), every Floquet multiplier is positioned within the unit circle, signifying stability in the system’s periodic solution. Nonetheless, once A surpasses 0.94, a Floquet multiplier exits the unit circle, passing through the point \((1, 0)\) on the complex plane, signifying the system’s entrance into the first periodicdoubling bifurcation, resulting in the loss of stability for the periodic solution. As A further increases beyond 1.46, the Floquet multiplier reenters the unit circle through the point \((1,0)\), suggesting the system undergoes a reverse periodicdoubling bifurcation, thereby restoring stability to the previously unstable periodic solution.
The findings from the preceding analysis are corroborated by constructing the bifurcation diagram of the PNS, as depicted in Fig. 5. This diagram, generated from the highprecision semianalytical solutions derived through the TMRM, plots the extreme values of \(x_1\) (Fig. 5a) and \(x_2\) (Fig. 5b) on the vertical axis. In Fig. 5a, the stable periodic1 solutions are indicated by a solid black line, whereas a dashed line denotes the unstable periodic1 solutions, and the stable periodic2 solutions are marked with blue dots. The diagram reveals that the stable periodic1 solution undergoes bifurcation into a stable periodic2 solution when A surpasses 0.94. Notably, while the system retains a periodic1 solution, this solution becomes unstable. It is essential to highlight that while numerical methods can obtain stable periodic2 solutions, unstable periodic1 solutions require semianalytical approaches like the TMRM. The bifurcation diagram for \(x_2\) presented in Fig. 5b leads to analogous conclusions, thereby aligning perfectly with the observations in Fig. 4b.
Figure 5 illustrates that the system exhibits a subharmonic response when the amplitude A falls within the range (0.94, 1.46). Specifically, at \(A=1.3\), the stable periodic2 solution can be determined using the TMRM. The time histories and the corresponding phase diagram for the PNS are showcased in Fig. 6. Here, the solid red (\(x_1\)) and blue (\(x_2\)) lines depict the outcomes computed by the TMRM with a truncation series of \(N=30\). Meanwhile, the black dots denote the responses acquired via the 4th RK method. Observations from the figure reveal that both \(x_1\) and \(x_2\) exhibit dual extreme values within their time histories, correlating to the two blue dots during the interval \(A\in (0.94, 1.46)\) in Fig. 5. The phase diagram in Fig. 6b demonstrates two rotations, indicative of the periodic2 behavior. Further analysis involves performing a spectral transformation on the system’s periodic2 response to scrutinize the harmonic content. Figure 7 displays a logarithmic spectrogram density analysis of \(x_2\), derived from executing a fast Fourier transform (FFT) on the stable periodic2 response and subsequently applying a logarithmic scale to the amplitude. The analysis reveals that besides the primary harmonic response at a frequency of \(\Omega \), a subharmonic response emerges at a frequency of \(\frac{\Omega }{2}\). This subharmonic frequency, corresponding to the stable periodic2 solution as expressed by \(\Omega _k=\frac{k\Omega }{2}\), underscores the intricate dynamical behavior of the system within the specified amplitude range.
With \(\varphi \) is selected as the bifurcation parameter from Eq. (5) for system stability analysis, the other parameters are set as \(\alpha =0, \delta =0.8, \mu =0.1, A=1\), and \(\Omega =0.942\), with \(\varphi \) decreasing progressively from 0.6 to 0. This approach mirrors the bifurcation analysis conducted for the external excitation amplitude A, where the semianalytical solution is initially derived through the TMRM. Subsequently, a bifurcation diagram depicting the response of \(x_1\) as a function of \(\varphi \) is generated, as illustrated in Fig. 8. This diagram reveals that as \(\varphi \) diminishes, the PNS displays increasingly complex dynamical behavior. Specifically, a unique stable periodic1 solution (represented by black dots) is observed as \(\varphi \) decreases from 0.6 to 0.394 (highlighted by the red point \(P_1\) in Fig. 8b). Below \(\varphi =0.394\), a critically stable periodic3 solution emerges (depicted by pink red dots), coexisting with the stable periodic1 solution.
As \(\varphi \) crosses 0.374 (marked by the black point \(P_2\)), the PNS generates a saddlenode bifurcation. Specifically, the critically stable periodic3 solution bifurcates, yielding two periodic3 solutions with differing amplitudes: one stable (blue dots) and the other unstable (green dots). Thus, in the range \(\varphi \in (0.330, 0.374)\), the system simultaneously exhibits three periodic solutions: a stable periodic1 solution, along with both stable and unstable periodic3 solutions. The point \(\varphi =0.330\) (the pink point \(P_3\)) is identified as a critical juncture where the stable periodic1 solution becomes unstable. When \(\varphi \) decreases from 0.330 to 0, the system’s stability status remains consistent, maintaining an unstable periodic1 solution alongside both a stable and an unstable periodic3 solution.
The stability of these solutions is further elucidated by analyzing the corresponding Floquet multipliers. Figure 9a showcases the trend of Floquet multipliers for the periodic1 solution, indicating stability with all multipliers inside the unit circle for \(\varphi >0.330\). Upon crossing point \(P_3\), the Floquet multipliers move outside the unit circle, indicating that all subsequent periodic1 solutions become unstable. Figure 9b focuses on the Floquet multiplier within \(\varphi \in (0.374, 0.393)\), where the system maintains one multiplier on the unit circle, especially at (1,0) on the complex plane, while the other remains inside, suggesting the periodic3 solution’s critical stability at this juncture. Figure 9c is based on the stable periodic3 solution depicted in Fig. 8, with Floquet multipliers for the blue dots entirely within the unit circle, affirming these solutions’ asymptotic stability. Calculation of Floquet multipliers for the periodic3 solution (green dots in Fig. 8) reveals that the modulus of the maximum Floquet multiplier consistently exceeds 1 within this range (Fig. 9d), indicating the instability of these periodic3 solutions.
To further elucidate the system’s dynamics, phase diagrams of \(x_1\) for \(\varphi =0.38, \varphi =0.35\) and \(\varphi =0.20\) are depicted in Fig. 10. For \(\varphi =0.38\), Fig. 10a, b illustrate the stable periodic1 solution and the critically stable periodic3 solution, as identified in Fig. 8. When \(\varphi \) is adjusted to 0.35, Fig. 10c–e display a trio of solutions coexisting within the system: a stable periodic1 solution, a stable periodic3 solution, and an unstable periodic3 solution. These are visually represented by a black solid line, a blue solid line, and a green dashed line, respectively, corresponding to the black, blue, and green dots in Fig. 8 for \(\varphi =0.35\). Similarly, Figs. 10f–h portray the conditions at \(\varphi =0.20\), showcasing the coexistence of an unstable periodic1 solution, a stable periodic3 solution, and an unstable periodic3 solution. The phase diagrams are complemented by the numerical solutions calculated using the 4th RK method, highlighted with black dots. The correspondence of the semianalytical solutions obtained through the TMRM with numerical results highlights the compatibility of the methods. Notably, the amplitude of the periodic1 solution exhibits minor variations across different \(\varphi \) values, in contrast to the periodic3 solution, where significant amplitude discrepancies are observed, highlighting the system’s sensitive response to changes in \(\varphi \) (cell membrane resistance). Due to the fact that the PNS model of piezoelectric ceramics for acousticelectric conversion can simulate cochlear nerve excitation, the bifurcation phenomenon of PNS under the influence of resistance parameters will have great potential applications in restoring hearing.
5 Conclusion
This paper presents the highaccuracy semianalytical solution of the piezoelectric neuron system (PNS) via the timedomain minimum residual method (TMRM) and performs an indepth bifurcation analysis of the PNS utilizing the derived semianalytical solution. Specifically, we analyzed the periodicdoubling bifurcation phenomenon that occurs when the amplitude of external excitation (the amplitude of external stimuli) changes, as well as the periodic instability phenomenon caused by changes in parameter \(\varphi \) (cell membrane resistance). Based on the analysis results, the following conclusions are drawn:

The TMRM efficiently yields the periodic solution of the PNS, with the precision of the semianalytical solution enhancing as the truncation series is extended. With identical parameter configurations, the solution’s accuracy via the TMRM surpasses that achieved through the IHB method.

As the amplitude of the external excitation increases, the system exhibits a periodicdoubling bifurcation phenomenon, which means that in addition to an unstable periodic1 solution, there is also a stable periodic2 solution with the subharmonic frequencies \(\Omega _k=\frac{k\Omega }{2}\), and the relevant Floquet multiplier similarly exits the unit circle, passing through the point \((1, 0)\) on the complex plane.

As the amplitude of the external excitation further increases, the Floquet multiplier moves back into the unit circle via the point \((1,0)\), indicating the system experiences a periodicdoubling bifurcation in reverse.

The reduction of \(\varphi \) brings about complex dynamic phenomena. As \(\varphi \) decreases, the system first exhibits a critical stable periodic3 solution. As \(\varphi \) further decreases, the PNS exhibits a saddlenode bifurcation; that is, the system has generated a stable periodic3 solution and an unstable periodic3 solution.

The change of \(\varphi \) has further led to the transformation of the stable periodic1 solution into unstable, and the corresponding periodic1 solution will cross the unit circle. Currently, the system has stable/unstable periodic1 solutions and periodic3 solutions.
Data availability
All data included in this study are available upon request by contact with the corresponding author.
References
Gu, H.G., Pan, B.B.: A fourdimensional neuronal model to describe the complex nonlinear dynamics observed in the firing patterns of a sciatic nerve chronic constriction injury model. Nonlinear Dynamics 81(4), 2107–2126 (2015). https://doi.org/10.1007/s1107101521297
Lu, Q.S., Gu, H.G., Yang, Z.Q., Shi, X., Duan, L.X., Zheng, Y.H.: Dynamics of firing patterns, synchronization and resonances in neuronal electrical activities: experiments and analysis. Acta Mechanica Sinica 24(6), 593–628 (2008). https://doi.org/10.1007/s1040900802048
Dayan, P., Abbott, L.F.: Theoretical neuroscience: computational and mathematical modeling of neural systems. The MIT press, Cambridge, MA (2005)
Wu, F.Q., Ma, J., Zhang, G.: A new neuron model under electromagnetic field. Applied Mathematics and Computation 347, 590–599 (2019). https://doi.org/10.1016/j.amc.2018.10.087
Hodgkin, A.L., Huxley, A.F.: A quantitative description of membrane current and its application to conduction and excitation in nerve. The Journal of Physiology 117(4), 500 (1952). https://doi.org/10.1113/jphysiol.1952.sp004764
FitzHugh, R.: Mathematical models of threshold phenomena in the nerve membrane. The Bulletin of Mathematical Biophysics 17, 257–278 (1955). https://doi.org/10.1007/BF02477753
Aihara, K., Matsumoto, G., Ikegaya, Y.: Periodic and nonperiodic responses of a periodically forced hodgkinhuxley oscillator. Journal of theoretical biology 109(2), 249–269 (1984). https://doi.org/10.1016/S00225193(84)800053
Aihara, K., Matsumoto, G.: Chaotic oscillations and bifurcations in squid giant axons. Chaos 12, 257–269 (1986). https://doi.org/10.1515/9781400858156.257
Hindmarsh, J.L., Rose, R.M.: A model of neuronal bursting using three coupled first order differential equations, Proceedings of the Royal society of London. Series B. Biological sciences 221 (1222) (1984) 87–102. https://doi.org/10.1098/rspb.1984.0024
Tamaševičius, A., Mykolaitis, G., Tamaševičiūtė, E., Bumelienė, S.: Twoterminal feedback circuit for suppressing synchrony of the fitzhughnagumo oscillators. Nonlinear Dynamics 81(1–2), 783–788 (2015). https://doi.org/10.1007/s110710152028y
Yan, B., Panahi, S., He, S.B., Jafari, S.: Further dynamical analysis of modified fitzhughnagumo model under the electric field. Nonlinear Dynamics 101, 521–529 (2020). https://doi.org/10.1007/s1107102005816y
Li, R.H., Wang, Z.H., Dong, E.Z.: A new locally active memristive synapsecoupled neuron model. Nonlinear Dynamics 104(4), 4459–4475 (2021). https://doi.org/10.1007/s11071021065741
Saadatzi, M., Saadatzi, M.N., Banerjee, S.: Modeling and fabrication of a piezoelectric artificial cochlea electrode array with longitudinal coupling. IEEE Sensors Journal 20(19), 11163–11172 (2020). https://doi.org/10.1109/JSEN.2020.2996192
Kim, Y., Kim, J.S., Kim, G.W.: A novel frequency selectivity approach based on travelling wave propagation in mechanoluminescence basilar membrane for artificial cochlea. Scientific Reports 8, 12023 (2018). https://doi.org/10.1038/s41598018306330
Zhou, P., Yao, Z., Ma, J., Zhu, Z.G.: A piezoelectric sensing neuron and resonance synchronization between auditory neurons under stimulus. Chaos, Solitons & Fractals 145, 110751 (2021). https://doi.org/10.1016/j.chaos.2021.110751
Liu, Y., Xu, Y., Ma, J.: Synchronization and spatial patterns in a lightdependent neural network. Communications in Nonlinear Science and Numerical Simulation 89, 105297 (2020). https://doi.org/10.1016/j.cnsns.2020.105297
Guo, Y.Y., Wang, C.N., Yao, Z., Xu, Y.: Desynchronization of thermosensitive neurons by using energy pumping. Physica A: Statistical Mechanics and Its Applications 602, 127644 (2022). https://doi.org/10.1016/j.physa.2022.127644
Yang, F.F., Wang, Y., Ma, J.: Creation of heterogeneity or defects in a memristive neural network under energy flow. Communications in Nonlinear Science and Numerical Simulation 119, 107127 (2023). https://doi.org/10.1016/j.cnsns.2023.107127
Xie, Y., Zhou, P., Yao, Z., Ma, J.: Response mechanism in a functional neuron under multiple stimuli. Physica A: Statistical Mechanics and its Applications 607, 128175 (2022). https://doi.org/10.1016/j.physa.2022.128175
Hua, H.T., Gu, H.G.: Bifurcations underlying sigh and eupnea rhythmic transition in a prebötzinger complex model. The European Physical Journal Special Topics 231(22–23), 4109–4116 (2022). https://doi.org/10.1140/epjs/s11734022006315
Liu, G., Wang, L., Luo, W.L., Liu, J.K., Lu, Z.R.: Parameter identification of fractional order system using enhanced response sensitivity approach. Communications in Nonlinear Science and Numerical Simulation 67, 492–505 (2019). https://doi.org/10.1016/j.cnsns.2018.07.026
Skufca, J.D.: Analysis still matters: A surprising instance of failure of rungekuttafelberg ode solvers. SIAM review 46(4), 729–737 (2004). https://doi.org/10.1137/S003614450342911X
Liu, G., Liu, J.K., Wang, L., Lu, Z.R.: Timedomain minimum residual method combined with energy balance for nonlinear conservative systems. Mechanical Systems and Signal Processing 170, 108818 (2022). https://doi.org/10.1016/j.ymssp.2022.108818
Liu, G., Liu, J.K., Lu, Z.R.: Highprecision semianalytical solution for the quasiperiodic nanobeam system based on the weight timedomain minimum residual method. Composite Structures 323, 117457 (2023). https://doi.org/10.1016/j.compstruct.2023.117457
Chen, Y.M., Liu, Q.X., Liu, J.K.: Harmonic balancebased approach for optimal time delay to control unstable periodic orbits of chaotic systems. Acta Mechanica Sinica 36(4), 918–925 (2020). https://doi.org/10.1007/s1040902000966z
Kim, K., Ri, K., Yun, C., Jong, Y., Han, P.: Nonlinear forced vibration and stability analysis of nonlinear systems combining the ihb method and the aft method. Computers & Structures 264, 106771 (2022). https://doi.org/10.1016/j.compstruc.2022.106771
Liu, G.: Analysis methods and parameter identification for strongly nonlinear systems based on response sensitivity. Sun Yatsen University, Thesis (2020)
Liu, G., Lu, Z.R., Wang, L., Liu, J.K.: A new semianalytical technique for nonlinear systems based on response sensitivity analysis. Nonlinear Dynamics 103(2), 1529–1551 (2021). https://doi.org/10.1007/s1107102006197y
Liu, G., Liu, J.K., Wang, L., Lu, Z.R.: A new semianalytical approach for quasiperiodic vibrations of nonlinear systems. Communications in Nonlinear Science and Numerical Simulation 103, 105999 (2021). https://doi.org/10.1016/j.cnsns.2021.105999
Nagumo, J., Arimoto, S., Yoshizawa, S.J.: An active pulse transmission line simulating nerve axon. Proceedings of the IRE 50(10), 2061–2070 (1962). https://doi.org/10.1109/JRPROC.1962.288235
Guo, Y.T., Zhou, P., Yao, Z., Ma, J.: Biophysical mechanism of signal encoding in an auditory neuron. Nonlinear Dynamics 105(4), 3603–3614 (2021). https://doi.org/10.1007/s1107102106770z
Liu, G., Lu, Z.R., Liu, J.K., Chen, Y.M.: Quasiperiodic aeroelastic response analysis of an airfoil with external store by incremental harmonic balance method. International Journal of NonLinear Mechanics 100, 10–19 (2018). https://doi.org/10.1016/j.ijnonlinmec.2018.01.004
Acknowledgements
The present investigation was performed under the support of the National Natural Science Foundation of China (No. 12302031), GuangDong Basic and Applied Basic Research Foundation (No. 2023A1515010028, No. 2023A1515240046), the State Key Laboratory of Mechanical System and Vibration (No. MSV202402), Shenzhen Science and Technology Program (No. ZDSYS20210623091808026).
Funding
The authors have not disclosed any funding.
Author information
Authors and Affiliations
Contributions
Xuejun Li: Programming, Formal analysis, Data curation, Writingoriginal draft. Yanmao Chen: Validation, Methodology, Investigation, Supervision. Jike Liu: Programming, Designing computer programs, Visualization. Guang Liu: Conceptualization, Programming, Writingoriginal draft & review & editing, Project administration, Funding acquisition.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The MATLAB implementation codes in this paper can be downloaded at: Guang Liu’s Website, Guang Liu’s Blog or Guang Liu’s ResearchGate.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author selfarchiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Li, Xj., Chen, Ym., Liu, Jk. et al. Deciphering the complex behavior of piezoelectric neuron systems through bifurcation analysis with the timedomain minimum residual method. Nonlinear Dyn 112, 17417–17431 (2024). https://doi.org/10.1007/s1107102409924x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s1107102409924x