Time-Delayed Dynamics in Regular Kuramoto Networks with Inertia: Multistability, Traveling Waves, Chimera States, and Transitions to Seizure-Like Activity Esmaeil Mahdavi, Philipp Hövel, Mina Zarei, Farhad Shahbazi Abstract This study examines the complex interplay between inertia and time delay in regular rotor networks within the framework of the second-order Kuramoto model. By combining analytical and numerical methods, we demonstrate that intrinsic time delays – arising from finite information transmission speeds – induce multistability among fully synchronized phase-locked states. Unlike systems without inertia, the presence of inertia destabilizes these phase-locked states, reduces their basin of attraction, and gives rise to nonlinear phase-locked dynamics over specific inertia ranges. In addition, we show that time delays promote the emergence of turbulent chimera states, while inertia enhances their spatial extent. Notably, the combined influence of inertia and time delay produces dynamic patterns reminiscent of partial epileptic seizures. These findings provide new insights into synchronization phenomena by revealing how inertia and time delay fundamentally reshape the stability and dynamics of regular rotor networks, with broader implications for neuronal modeling and other complex systems. I Introduction Synchronization is a fundamental phenomenon in both natural and engineered systems, where individual components collectively exhibit coherent dynamics. Since its introduction in 1975, the Kuramoto model has served as a cornerstone for studying synchronization, modeling interactions through first-order nonlinear coupled oscillators Kuramoto 1975 . This framework has provided key insights into synchronization in chemical, biological, and mechanical systems Tyson 1973 ; Mirollo and Strogatz 1990 ; Nijmeijer and Rodriguez-Angeles 2003 . However, the first-order Kuramoto model has notable limitations: it often converges rapidly to partially synchronized states and fails to capture frequency adaptation observed in certain biological systems, such as firefly populations Ermentrout 1991 . To address these shortcomings, Ermentrout introduced an extension incorporating inertia, giving rise to the second-order Kuramoto model Ermentrout 1991 ; Tanaka, Lichtenberg, and Oishi 1997a . This modification fundamentally changes the character of phase transitions Tanaka, Lichtenberg, and Oishi 1997b and provides a more realistic framework for analyzing complex systems, including Josephson junction arrays, power grids, and neuronal networks Trees, Saranathan, and Stroud 2005 ; Filatrella, Nielsen, and Pedersen 2008 ; Rohden et al. 2012 ; Rohden et al. 2014 ; Grzybowski, Macau, and Yoneyama 2016 . The second-order Kuramoto model is especially relevant for neuroscience, where simplified neuronal prototypes can be expressed as second-order nonlinear differential equations Dolan, Majtanik, and Tass 2005 ; Majtanik, Dolan, and Tass 2006 ; Sakyte and Ragulskis 2011 . Furthermore, Josephson junction dynamics have been shown to closely mimic neuronal behavior Mishra et al. 2021 . Another critical factor influencing synchronization is time delay, which arises naturally from finite signal transmission speeds in systems such as neural networks, electronic circuits, and laser arrays Kozyreff, Vladimirov, and Mandel 2000 ; Reddy, Sen, and Johnston 2000 ; Yeung and Strogatz 1999 ; Kerszberg and Zippelius 1990 ; Waibel et al. 2013 . Time delays can profoundly affect system dynamics by inducing multistability Schuster and Wagner 1989 ; Mahdavi, Zarei, and Shahbazi 2025 ; Yeung and Strogatz 1999 ; Choi et al. 2000 ; Madadi Asl, Valizadeh, and Tass 2018 , chimera states, and first-order phase transitions Ameli, Karimian, and Shahbazi 2021 . In neuronal networks, time delays have been shown to foster synchronization between distinct populations Vicente et al. 2008 or even trigger transitions from synchronized to desynchronized states Lehnert et al. 2011 . Incorporating time delay into the Kuramoto model with inertia substantially increases the complexity of network dynamics. In binary rotor systems, time delay induces multistability, whereas greater inertia reduces the stability of phase-locked states. Their combined influence can generate periodic or chaotic non-phase-locked solutions, with time delay itself acting as a control parameter for chaos Mahdavi, Zarei, and Shahbazi 2025 . In globally coupled rotor networks, the interplay between inertia and time delay gives rise to both subcritical (first-order) and supercritical (second-order) bifurcations between synchronized and incoherent states Métivier, Wetzel, and Gupta 2020 . Analytical investigations of the second-order Kuramoto model with delay have further explored these phenomena Mahdavi, Zarei, and Shahbazi 2025 ; Métivier, Wetzel, and Gupta 2020 ; Hong, Jeon, and Choi 2002 ; Prousalis and Wetzel 2022 ; Dai et al. 2018 . The interaction between inertia and time delay adds significant complexity to network dynamics. Previous studies have shown that time delay can induce multistability in binary rotor systems and regular networks, as well as alter the stability of phase-locked states Schuster and Wagner 1989 ; Ameli et al. 2025 . Together, inertia and delay can generate periodic or chaotic behavior Mahdavi, Zarei, and Shahbazi 2025 and modulate phase transitions between synchronized and incoherent states. Although the influence of time delay has been widely investigated in globally coupled networks and first-order models, its role in regular networks with local interactions, and within the second-order Kuramoto framework remains insufficiently understood. This work addresses this gap by examining the joint effects of inertia and time delay on synchronization dynamics in regular rotor networks governed by the second-order Kuramoto model. Using analytical derivations and numerical simulations, we investigate their impact on multistability, phase-locked states, and chimera patterns. Our results provide novel insights into the stability boundaries of synchronized states and reveal dynamic regimes reminiscent of partial epileptic seizures. These findings advance the understanding of synchronization in complex systems and offer implications for neuronal modeling and related applications. The paper is structured as follows: Section II introduces the mathematical model and methods used to analyze fully synchronized phase-locked solutions and their stability. Section III presents numerical results that illustrate the system’s dynamics, and Section IV concludes with a summary of the main findings and their broader implications. II Model an Method We use the second-order Kuramoto model to study the effects of time-delayed interactions in a network of coupled oscillators. For a system of N identical rotors, the model is given by Tanaka, Lichtenberg, and Oishi 1997a : m\ddot{\theta}_{i}(t)+\dot{\theta}_{i}(t)=\omega+\frac{\alpha}{k_{i}}\sum_{j=1}^{N}a_{ij}\sin(\theta_{j}(t-\tau)-\theta_{i}(t)), (1) where \theta_{i} denotes the phase of the i th rotor, m the inertia, \omega the intrinsic frequency, \alpha the coupling strength, \tau the time delay, and k_{i} the degree of node i . The adjacency matrix a_{ij} specifies the network connectivity, with a_{ij}=1 indicating a link between rotors i and j , and a_{ij}=0 otherwise. Throughout this study, we consider a regular ring topology with P neighbors to either side of a node and use the following parameters: N=1000 , P=5 , i.e. , k_{i}=\langle k\rangle=2P=10 for i=1,\dots,N , \omega=1 , and \alpha=1 . To analyze the stability of the fully synchronized state and determine the corresponding phase-locked frequency \Omega_{f} , we derive the master stability function. In this state, all rotors share a common temporal phase evolution Mahdavi, Zarei, and Shahbazi 2025 : {\theta}_{i}(t)=\Omega_{f}t+\beta, (2) where \beta is a constant phase offset. To assess the stability of this solution, we introduce a small perturbation \xi_{i} to the rotor phases and linearize the system, obtaining: m\ddot{\bm{\xi}}+\dot{\bm{\xi}}={\alpha}\cos({\Omega_{f}}\tau)(\bm{\xi}^{\tau}-\bm{\xi})-\frac{{\alpha}}{\langle k\rangle}\cos({\Omega_{f}}\tau)\bm{L}\bm{\xi}^{\tau}. (3) where \bm{\xi}(t)=[\xi_{1}(t),\dots,\xi_{N}(t)]^{T} and \bm{\xi}^{\tau}(t)=[\xi_{1}(t-\tau),\dots,\xi_{N}(t-\tau)]^{T} denote the perturbation vectors at times t and t-\tau , respectively. Here, \bm{L} denotes the Laplacian matrix with elements l_{ij}=\delta_{ij}k_{i}-a_{ij} . Diagonalizing \bm{L} yields m\ddot{\bm{\eta}}+\dot{\bm{\eta}}={\alpha}\cos({\Omega_{f}}\tau)(\bm{\eta}^{\tau}-\bm{\eta})-\frac{{\alpha}}{\langle k\rangle}\cos({\Omega_{f}}\tau)\bm{\Lambda}\bm{\eta}^{\tau}, (4) where \bm{\Lambda} is a diagonal matrix containing the Laplacian eigenvalues \lambda_{i} . Assuming solutions of the form \eta_{i}(t)\propto\exp(z_{i}t) leads to the characteristic equation governing the stability of each node: \displaystyle mz^{2}_{i}+z_{i} \displaystyle={\alpha}\cos({\Omega_{f}}\tau)e^{-z_{i}\tau}\left(1-e^{z_{i}\tau}-\frac{\lambda_{i}}{\langle k\rangle}\right). (5) Equation ( 5 ) constitutes the master stability function of the system. The phase-locked state is stable if all z_{i} have negative real parts; otherwise, it is unstable. The phase-locked frequency \Omega_{f} in the fully synchronized state is obtained from the roots of the transcendental equation \Omega_{f}-\omega+{\alpha}\sin(\Omega_{f}\tau)=0. (6) The degree of synchrony in the network is characterized by the Kuramoto order parameter r(t) , defined as \displaystyle r(t) \displaystyle= \displaystyle\frac{1}{N}\left|\sum_{j=1}^{N}e^{i\theta_{j}(t)}\right|, (7) where 0\leq r(t)\leq 1 , with r(t)=1 denoting perfect synchrony and r(t)=0 indicating complete incoherence. In addition, the local order parameter R_{l}(t) quantifies the synchrony between rotor l and its 2M nearest neighbors: R_{l}(t)=\frac{1}{2M+1}\left|\sum_{j=l-M}^{l+M}e^{i\theta_{j}(t)}\right|, (8) with 0\leq R_{l}(t)\leq 1 . Here, R_{l}(t)=1 indicates perfect local synchrony, while R_{i}(t)=0 denotes complete incoherence between rotor i and its neighborhood. In the following, we use M=25 , that is, we calculate the local order via an average over 2M=50 node. To characterize the spatial organization of phase relationships within the network, we use the cosine similarity matrix S given by \displaystyle S_{ij} \displaystyle= \displaystyle\lim_{\Delta t\to\infty}\frac{1}{\Delta t}\int_{t_{s}}^{t_{s}+\Delta t}\cos(\theta_{i}(t)-\theta_{j}(t))\;\mathrm{d}t, (9) where -1\leq S_{ij}\leq 1 , and t_{s} denotes the transient time. A value of S_{ij}=1 indicates that rotors i and j are perfectly in phase, whereas S_{ij}=-1 corresponds to a perfectly anti-phase relationship. For the temporal similarity and periodicity, we also calculate the average autocorrelation function C defined as C(\tau^{\prime})=\left\langle\left|\frac{1}{N}\sum_{j=1}^{N}e^{i\theta_{j}(t)}e^{-i\theta_{j}(t-\tau^{\prime})}\right|\right\rangle. (10) Furthermore, we classify the observed chimera states following the methodology of Kemeth et al. Kemeth et al. 2016 , which combines a spatial coherence function with a measure of temporal correlation. The spatial coherence function is obtained by first evaluating the local curvature of the spatial phase profile using a discrete Laplacian operator. Specifically, the discrete Laplacian \mathbf{D} is rescaled and applied to the spatial phase data f at time t : \hat{\mathbf{D}}f=\Delta x^{2}\mathbf{D}f=f(x+\Delta x,t)-2f(x,t)+f(x-\Delta x,t), (11) where the phases are represented as complex numbers on the unit circle. The maximum value of |\hat{\mathbf{D}}f| is D_{m}=4 , and the spatial coherence function g_{0}(t) is subsequently computed as: g_{0}(t)=\int_{0}^{0.01D_{m}}g(t,|\hat{\mathbf{D}}f|)d|\hat{\mathbf{D}}f|, (12) denotes the normalized probability density function of |\hat{\mathbf{D}}f| at time t . A value of g_{0}(t)=1 corresponds to a fully synchronized state, g_{0}(t)=0 indicates complete incoherence, and intermediate values 00 , \{\forall t:00 , \{\forall t:00 , \{\forall t:0\tau , the interactions at time t depend on the rotor states at the earlier time t-\tau , thereby naturally incorporating the system’s history into its dynamics. Figure 2 provides a systematic numerical characterization of the delay-dependent stationary states across different inertia regimes. The panel layout highlights parameter-dependent transitions, with each row corresponding to a specific inertia value and columns representing four key aspects of stationary collective dynamics: (i) the stationary Kuramoto order parameter \langle r\rangle_{\infty} , which quantifies the degree of global phase synchronization in the long-time limit; (ii) \mathrm{\sigma}_{r} , capturing fluctuations of the order parameter in the stationary regime; (iii) the stationary ensemble-averaged mean rotor frequency \langle\dot{\theta}\rangle_{\infty} ; and (iv) \mathrm{\sigma}_{\dot{\theta}} , measuring the dispersion of individual rotor frequencies in the asymptotic state. To ensure statistical robustness, we performed 60 independent realizations for each parameter set, each with distinct random initial conditions. The results show both individual trajectories (red dots) and ensemble statistics (blue dots with error bars corresponding to \pm\sigma/\sqrt{60} standard error). This dual representation captures both the distribution of possible stationary states across different initial conditions and their mean behavior with quantified confidence. By combining multiple metrics across well-sampled realizations, this analysis allows a comprehensive identification of synchronization transitions and stability thresholds as a function of \tau . The first column shows the steady-state order parameter as a function of time delay for different inertia values. An intermediate time-delay regime emerges in which synchronization is significantly weakened, approaching an effectively asynchronous state. Within this regime, the order parameter is largely insensitive to ensemble variations. Notably, the width of this desynchronized region increases systematically with inertia. At the left and the right of this intermediate region, the system exhibits bistability, alternating between full synchronization and complete desynchronization, with the final state strongly dependent on initial conditions. This multistability underscores the system’s sensitivity to its initial preparation. The second column quantifies the temporal evolution of the order parameter via its standard deviation over time, averaged across ensembles. In phase-locked regimes – both synchronized and desynchronized – the order parameter remains constant, yielding a zero standard deviation, in excellent agreement with the analytical predictions in Fig. 1 (a). In contrast, within the intermediate time-delay region, the order parameter shows pronounced temporal fluctuations, which are largest near the region boundaries and increase markedly with inertia. In the third column, we show the ensemble-averaged mean oscillator frequency in the steady state. Among the synchronous solution branches identified analytically (Fig. 1 (a)), phase-locked states near the intrinsic frequency dominate numerically, reflecting their larger basins of attraction. However, within the intermediate parameter region, which broadens with increasing inertia, no phase-locked solutions are observed across the 60 ensemble realizations. This absence arises because the basins of attraction for these synchronous states are extremely narrow, requiring precise initial conditions. As a result, increasing inertia expands the regime dominated by non-phase-locked dynamics. At the boundaries of this intermediate region, inertia-free systems exhibit abrupt jumps in mean frequency between solutions. With increasing inertia, these transitions smooth out, suggesting that inertia regularizes frequency dynamics. Notably, the frequencies corresponding to the fully synchronized ( \langle r\rangle_{\infty}=1 ) and fully asynchronous ( \langle r\rangle_{\infty}=0 ) states are very close, supporting the proposition that linear phase-locked solutions in both states can attain similar values, as demonstrated in the supplementary mathematical proof and illustrated in Figs. SF1 and SF2. To further examine individual oscillator behavior, we analyze the statistical distribution of rotor frequencies over steady-state trajectories. The fourth column presents the standard deviation of this distribution. In phase-locked states, the dispersion vanishes, indicating complete frequency entrainment. Within the intermediate region, finite standard deviations reflect inter-oscillator frequency differences, temporal variations, or both. Frequency dispersion peaks near the region boundaries, consistent with the previously observed maximal order parameter fluctuations. Importantly, substantial frequency heterogeneity emerges among oscillators within this region, particularly near its edges. For m=1 , a particularly notable dynamical regime emerges when the time delay exceeds \tau\approx 4.85 . In this range, we observe solutions covering the full interval r\in[0,1] that appear uniquely for this inertia value. Multiple stable solutions coexist, each maintaining a time-invariant order parameter despite ongoing variations in individual rotor frequencies (see Fig. 2 (c)). This transition coincides with the destabilization of the second frequency branch of phase-locked solutions shown in Fig. 1 (a) (iii), which previously dominated the dynamics due to their large basins of attraction near the intrinsic frequency. The loss of stability of these solutions allows the emergence of novel states not captured by linear approximations. This phenomenon will be examined in greater detail later. Since the inertial state m=1 displays richer and more complex dynamics than other inertia values, we focused our analysis on this regime to examine the local behavior of network oscillators. Specifically, we computed the distribution of angular velocities, the temporal evolution of these velocities, and the cosine similarity matrices under varying time delays. The corresponding results are shown in Figs. 3 – 5 . In Fig. 3 , the distribution functions of angular velocities in panels (a), (b), and (f) are obtained in the form of the Dirac delta function, indicating that the angular velocity of all oscillators remains constant in time and equal to each other. This case demonstrates the existence of a linear phase-locked solution in the network, and the time evolution of the corresponding angular velocities in Fig. 4 confirms this result. This also demonstrates that the oscillator’s angular velocity depends on the time delay. Additionally, the corresponding cosine similarity matrices in Fig. 5 reveal that these linear phase-locked solutions converge either to \langle r\rangle_{\infty}=1 (panel a) or to \langle r\rangle_{\infty}=0 (panels b and f), with the latter corresponding to helical phase patterns. Panels (c), and (e) of Fig. 3 exhibit two-peaked distribution functions, whose corresponding time evolution in Fig. 4 indicates the variation of angular velocities of the oscillators between the two peaks. This behavior, where each rotor evolves independently and separately from the others, results in the vanishing similarity between them, as seen in Fig. 5 . The results in Fig. 3 (d) and also show unimodal angular frequency distribution functions with finite width. As shown by the time evolution of the corresponding angular velocities in Fig. 4 , some oscillators evolve with identical angular velocities while others change their angular velocities over time. This behavior leads to the formation of two groups of correlated and uncorrelated oscillators, as illustrated in the corresponding similarity matrices in Fig. 5 . Finally, Figs. 3 (g) and 3 (h) display U -shaped distribution functions of angular velocities for \tau= 5 and 6.2, respectively. This frequency distribution has not been previously observed in regular networks of first-order Kuramoto oscillators Ameli et al. 2025 . The U -shaped distribution of angular velocities indicates the periodic variation of the angular frequency of oscillators over time. It has to be noted that the nature of the angular velocity evolution depends on the value of the order parameter. When \langle r\rangle_{\infty} equals zero or one, all oscillators evolve identically (Fig 4 (g) and see the supplementary video SV1). However, when \langle r\rangle_{\infty} has a value between zero and one (as seen in Figs. 4 (h) and 5 (h)), the rotors no longer have equal frequencies at a given time, and the angular velocity evolution diagram turns to tilted strips (Fig 4 (h)). This state is a travelling wave in which some rotors get closer together and others move apart, resulting in a partially synchronized state with 0<\langle r\rangle_{\infty}<1 (see the supplementary video SV2 for a clear illustration of the dynamics). Consequently, the similarity matrix in this case closely resembles the helical patterns with \langle r\rangle_{\infty}=0 . However, due to temporal changes in the phase difference of adjacent rotors, their elements undergo modifications, as visible in Fig. 5 (h). Figure 6 elaborates further on the case of \tau=6.2 and m=1 and displays similarity matrices, phase snapshots, time evolutions of local order parameter, and angular velocity snapshots corresponding to four distinct behaviors with \langle r\rangle_{\infty}=0.766,0.438,0.117 , and 0.017 . Panels (a)–(c) exhibit a single-cycle behavior, each with one curved phase line, whereas panel (d) shows two cycles, producing two curved lines and larger-scale periodicity. All panels exhibit traveling wave behavior, in which rotors periodically converge and diverge while the global order parameter remains constant. This is reflected in the diagonal-line patterns of the local order parameter. Figure 7 presents the autocorrelation function for m=1 and various time delays. Panel (a) represents the fully coherent states, that is, the phase-locked states with \langle r\rangle_{\infty}=0 or 1 . Panels (b) and (c) showing short-range correlation represent the chaotic states emerging at the transition points. Finally, panel (d) represents a travelling wave state with 0<\langle r\rangle_{\infty}<1 whose periodicity arises from the rotors’ regularly oscillating angular velocities. Next, we investigate the effects of time delay and inertia on chimera state formation. Using the methodology of Kemeth et al. Kemeth et al. 2016 , we employ the spatial correlation function g_{0}(t) for quantitative characterization. Both the time-averaged value and temporal standard deviation of g_{0}(t) are computed to provide a comprehensive assessment of the system’s behavior. Figure 8 presents the mean and standard deviation of g_{0}(t) across various inertia and time delay values. Regions with \langle g_{0}(t)\rangle=1 indicate non-chimera states, corresponding to either complete synchronization or the helical states. In synchronized states, all rotors share identical phases, whereas the helical states ( \langle r\rangle_{\infty}=0 ) show phases uniformly distributed around a trigonometric circle with minimal separation. In both cases, we find g_{0}(t)=1 with zero standard deviation, indicating stable spatial correlations and the absence of chimera states. For higher inertia and time-delay values, \langle g_{0}(t)\rangle approaches unity with minor deviations, indicating phase-locked states in which most rotors maintain regular spacing while a few evolve independently. The near-zero standard deviation of g_{0}(t) in these regions further confirms the absence of significant chimera states, as the system primarily exhibits modified phase-locked behavior rather than true chimera patterns. The remaining regions where the spatial correlation function satisfies 0<\langle g_{0}(t)\rangle<1 correspond to chimera states in the network, characterized by the coexistence of coherent and incoherent domains. Within these regions, the magnitude of \langle g_{0}(t)\rangle directly reflects the spatial extent of coherence, with larger values indicating more extensive synchronized domains and consequently larger chimera states. Fig. 8 (a) reveals two distinct parameter regions supporting chimera state formation, both of which expand with increasing inertia m . Notably, inertia modifies the characteristic time delays for chimera emergence: at the left region shifts toward shorter delays while at the right zone moves towards higher delays as m increases. This demonstrates the complementary regulatory roles of inertia and time delay in chimera state formation. The temporal characteristics of g_{0}(t) oscillations, shown in Fig. 8 (b), provide further classification criteria. Regular periodic oscillations signify breathing chimera states, whereas irregular, chaotic fluctuations indicate turbulent chimeras. In our final investigation of chimera states, Fig. 9 presents the temporal evolution of local order parameters, oscillator frequencies, g_{0}(t) dynamics, h_{0} values, and phase distributions across varying inertia ( m ) and time delay ( \tau ) parameters. The persistent condition 0