Dynamic Synchronization and Resonance as the Origin of 1/f Fluctuations—Amplitude Modulation Across Music and Nature Authors: Akika Nakamichi, Izumi Uesaka, Masahiro Morikawa Published: 2025 ABSTRACT AbstractIn natural systems, astrophysics, biological physics, and social physics, 1/f fluctuations are observed across a wide range of systems. Focusing on the case of music, we propose and verify a physical mechanism for generating these fluctuations. This mechanism is based on amplitude modulation (AM) and demodulation (DM), where the 1/f spectral law appears not in the raw waveform but in its demodulated amplitude envelope. Two distinct yet complementary processes generate the required AM: (i) stochastic synchronization among oscillators, modeled via an extended Kuramoto framework that captures perpetual synchronization–desynchronization cycles, and (ii) frequency-selective resonance, modeled by spectral accumulation of eigenmodes in acoustic or structural environments. Numerical simulations demonstrate that both mechanisms, acting alone or in combination, robustly generate 1/f spectra spanning several digits when demodulation is applied and that the classical Kuramoto critical point is not essential for its emergence. While this analysis focuses on 1/f fluctuations in musical performance and acoustics, we also note that 1/f fluctuations inherent in musical scores may be similarly described by the AM/DM mechanism. 1. Introduction Also known as pink noise, 1/f fluctuation is a ubiquitous phenomenon that is widely observed and is characterized by the low-frequency power law S(ω)∝ω−β in the power spectral density (PSD), where ω is frequency and β=0.5∼1.5. These 1/f fluctuations appear everywhere in nature, such as in semiconductors, thin metals, bio-membranes, crystal oscillators, heartbeat rates, magnetoencephalography (MEG), and electroencephalograms (EEG) (brain), etc. [1]. Furthermore, artificial systems, such as music sound data, often exhibit typical 1/f fluctuations. This ubiquity strongly suggests the existence of a common underlying physical mechanism, yet its precise nature remains a subject of debate. Since the first discovery of 1/f fluctuations in 1925 in the voltage fluctuations in a vacuum tube [2], diverse theoretical explanations were proposed, including self-organized criticality, fractional Brownian motion, percolation models, and aggregation–relaxation processes [1]. While these frameworks successfully describe certain systems, they lack universality: Each model tends to be domain specific and often requires fine-tuning to reproduce empirical scaling exponents. A single, simple, and scalable physical principle that can generate 1/f spectra across such varied contexts has yet to be established. We argue that such ubiquitous phenomena must stem from simple and universal physical principles rather than complex, system-specific ones. In our previous publications [3,4], we proposed that the origin of 1/f fluctuations is the wave beat, or amplitude modulation (AM). In particular, if many waves with frequencies closer to each other can exhibit signals with arbitrarily low frequencies. Then, we attempted to verify and extend this proposal by applying it to earthquakes [5], solar flares [6], electric currents [7], and other phenomena. These systems were too complex to directly verify the simple mechanism. In this paper, we focus on music exhibiting typical 1/f fluctuations. There have been a tremendous number of works on music 1/f fluctuations after the initial work [8,9]. We examine the extent to which the above unified framework of AM applies to music. Music is generally composed of three layers: (A) Structure: Musical score is the composer’s creative output and describes the basic structure of the music. (B) Physiology: Players’ performance emits sound based on the score and exhibits their interpretation of the music. (C) Physics: The emitted sound resonates and propagates toward the listeners in the hall. We analyze the final output signal after all these layers in relation to the layers (B) and (C). Although some research reports the presence of 1/f fluctuations in the first layer (A), this layer is closely related to human brain activity and falls beyond our present scope [10]. We will extend the discussion on how to analyze this layer at the end of this paper. The key physical requirement for the AM of 1/f fluctuations is frequency accumulation, i.e., the presence of many oscillatory components with closely spaced frequencies, leading to slow envelope modulations via wave beating. We identify two distinct yet complementary physical processes that naturally generate such frequency accumulation:Synchronization: Stochastic synchronization among interacting oscillators leads to recurrent cycles of phase alignment and dispersion. We model this process using an extended stochastic Kuramoto framework that produces persistent low-frequency envelope variations without requiring fine-tuning of the classical synchronization threshold.Resonance: Resonance-driven spectral shaping, in which environmental or structural eigenmodes selectively amplify certain frequencies, creates envelope modulations even in the absence of direct coupling among oscillators. Synchronization: Stochastic synchronization among interacting oscillators leads to recurrent cycles of phase alignment and dispersion. We model this process using an extended stochastic Kuramoto framework that produces persistent low-frequency envelope variations without requiring fine-tuning of the classical synchronization threshold. Synchronization: Stochastic synchronization among interacting oscillators leads to recurrent cycles of phase alignment and dispersion. We model this process using an extended stochastic Kuramoto framework that produces persistent low-frequency envelope variations without requiring fine-tuning of the classical synchronization threshold. Resonance: Resonance-driven spectral shaping, in which environmental or structural eigenmodes selectively amplify certain frequencies, creates envelope modulations even in the absence of direct coupling among oscillators. Resonance: Resonance-driven spectral shaping, in which environmental or structural eigenmodes selectively amplify certain frequencies, creates envelope modulations even in the absence of direct coupling among oscillators. Our approach offers three major departures from previous works: First, it unifies the synchronization-based and resonance-based origins of 1/f noise within a single dynamical framework. Second, it shows that criticality is not a prerequisite: 1/f spectra arise over broad parameter ranges. And third, it demonstrates the cross-domain applicability of the AM/DM mechanism through numerical simulations and analyses of real-world data from orchestral and solo musical performances, seismic records, and astrophysical time series. The remainder of this paper is organized as follows: Section 2 summarizes empirical observations and unresolved puzzles regarding 1/f fluctuations in musical signals. These turn out to be good starting points for finding a simple physics for the origin of 1/f fluctuations. Section 3 introduces the stochastic Kuramoto model as a description of synchronization-induced AM and presents numerical results. Section 4 develops a resonance-based AM model and validates it against measured acoustic environments. Section 5 integrates the two mechanisms into a unified framework and explores their interplay in generating 1/f fluctuations. Section 6 presents possible approaches to the music layer (A) and part of (B) that build upon the successful application demonstrated in the previous sections. Finally, Section 7 discusses implications, limitations, and potential extensions of the present theory. 2. A Simple Origin of 1/f Fluctuations—Amplitude Modulation For a typical example of musical sound, we show (1) an orchestra music concert [11], and (2) soprano and chamber music concert [12]. Their PSDs exhibit typical 1/f fluctuations, with indices of −1.21 in Figure 1 and −0.98 in Figure 2, to more than four digits in both cases. Later, these examples turn out to be typical realizations of frequency accumulation—synchronization and resonance, respectively. Here we briefly describe our PSD analysis. We do not apply window functions before PSD analysis. Window functions are primarily designed to suppress spectral leakage arising from discontinuities at the boundaries of finite-length signals. While effective for identifying narrowband spectral features, they can distort broadband scale-free spectra by convolving the true power-law spectrum with the window’s spectral response, particularly affecting the low-frequency regime. For the purpose of estimating global power-law scaling, such as 1/f noise, the use of window functions may introduce systematic bias in the inferred exponent, and therefore, is not necessarily advantageous. As a matter of fact, we have checked Hann and Hamming windows and obtained almost the same power index—within several percent. To reduce statistical fluctuations while preserving scale-free behavior, the raw PSD was averaged over logarithmically spaced frequency bins. The error bars represent the variance within the bin. Power-law scaling was assessed by performing sequential linear regressions on the log–log binned PSD, disregarding several of the lowest-frequency points, starting from the lowest frequency and progressively including higher-frequency bins. For each cumulative fitting range, the power-law exponent and the coefficient of determination R2 were evaluated, and the range yielding the maximum R2 (subject to R2≥0.5 and a stable exponent) was selected. The red line was thus selected as the best fit within this range. The corresponding exponent was adopted as the final estimate of the spectral scaling. The remaining PSD graphs in this paper were obtained using the same method. The uniformity of the time-series data may be a concern. In general, the data is not uniform, and the 1/f fluctuation property is not steady; thus, the PSD here describes the average properties of the system. However, this non-uniformity itself is a powerful tool for observing the system much further. See Section 7 and the figure therein. We now examine the fundamental features of 1/f fluctuations in music. There are at least three unique properties of 1/f fluctuations in the music. Low-frequency signal continues without limit: The power −1 in PSD continues without bound. This implies a divergence in total power, as each octave contributes equally to the energy. Furthermore, if stationary, the system appears to possess an infinitely long memory, according to the Wiener–Khinchin theorem, which relates the PSD to the time correlation function. In particular, the orchestral music exhibits 1/f fluctuations throughout its entire performance, for example, for more than 30 min in Figure 1. This is contrary to the case in layer (A), where the domain of the power −1 in the PSD is one second to a few minutes (0.01 to 1 Hz) [10].Arbitrary low-frequency signal arises from a tiny system. In the ordinary argument, the system size determines the limiting frequency by a general-order estimate. From the system size l and the typical wave speed v, the maximum correlation time scale is of the order l/v. However, the 1/f fluctuation in music violates this general rule. The 100 m music hall and the sound speed only yield a characteristic timescale of 0.3 s or several Hz. On the contrary, the fluctuation of music 1/f continues for an hour or more. Therefore, we speculate that 1/f fluctuation in music is not an intrinsic property but something secondary among many waves, such as interference between them.In the music, 1/f fluctuation appears in the PSD for the data squared or zero-crossing time series; the original data never shows 1/f fluctuations. This fact has been widely appreciated since the pioneering work [8], as if it were self-evident. However, this is strange, as the original layer (A) directory exhibits 1/f fluctuations without any manipulation. Sometimes extra 1/f fluctuations may be generated in the later layers (B) and (C). The same property, the necessity of square operation, appears in economic data and some astrophysical 1/f fluctuations, while many others exhibit direct 1/f fluctuations. What is the difference? Low-frequency signal continues without limit: The power −1 in PSD continues without bound. This implies a divergence in total power, as each octave contributes equally to the energy. Furthermore, if stationary, the system appears to possess an infinitely long memory, according to the Wiener–Khinchin theorem, which relates the PSD to the time correlation function. In particular, the orchestral music exhibits 1/f fluctuations throughout its entire performance, for example, for more than 30 min in Figure 1. This is contrary to the case in layer (A), where the domain of the power −1 in the PSD is one second to a few minutes (0.01 to 1 Hz) [10]. Low-frequency signal continues without limit: The power −1 in PSD continues without bound. This implies a divergence in total power, as each octave contributes equally to the energy. Furthermore, if stationary, the system appears to possess an infinitely long memory, according to the Wiener–Khinchin theorem, which relates the PSD to the time correlation function. In particular, the orchestral music exhibits 1/f fluctuations throughout its entire performance, for example, for more than 30 min in Figure 1. This is contrary to the case in layer (A), where the domain of the power −1 in the PSD is one second to a few minutes (0.01 to 1 Hz) [10]. Arbitrary low-frequency signal arises from a tiny system. In the ordinary argument, the system size determines the limiting frequency by a general-order estimate. From the system size l and the typical wave speed v, the maximum correlation time scale is of the order l/v. However, the 1/f fluctuation in music violates this general rule. The 100 m music hall and the sound speed only yield a characteristic timescale of 0.3 s or several Hz. On the contrary, the fluctuation of music 1/f continues for an hour or more. Therefore, we speculate that 1/f fluctuation in music is not an intrinsic property but something secondary among many waves, such as interference between them. Arbitrary low-frequency signal arises from a tiny system. In the ordinary argument, the system size determines the limiting frequency by a general-order estimate. From the system size l and the typical wave speed v, the maximum correlation time scale is of the order l/v. However, the 1/f fluctuation in music violates this general rule. The 100 m music hall and the sound speed only yield a characteristic timescale of 0.3 s or several Hz. On the contrary, the fluctuation of music 1/f continues for an hour or more. Therefore, we speculate that 1/f fluctuation in music is not an intrinsic property but something secondary among many waves, such as interference between them. In the music, 1/f fluctuation appears in the PSD for the data squared or zero-crossing time series; the original data never shows 1/f fluctuations. This fact has been widely appreciated since the pioneering work [8], as if it were self-evident. However, this is strange, as the original layer (A) directory exhibits 1/f fluctuations without any manipulation. Sometimes extra 1/f fluctuations may be generated in the later layers (B) and (C). The same property, the necessity of square operation, appears in economic data and some astrophysical 1/f fluctuations, while many others exhibit direct 1/f fluctuations. What is the difference? In the music, 1/f fluctuation appears in the PSD for the data squared or zero-crossing time series; the original data never shows 1/f fluctuations. This fact has been widely appreciated since the pioneering work [8], as if it were self-evident. However, this is strange, as the original layer (A) directory exhibits 1/f fluctuations without any manipulation. Sometimes extra 1/f fluctuations may be generated in the later layers (B) and (C). The same property, the necessity of square operation, appears in economic data and some astrophysical 1/f fluctuations, while many others exhibit direct 1/f fluctuations. What is the difference? Let us first examine Property 3. We perform other operations on the original sound signal, such as arbitrary-power operations, rectification, and thresholding on the original orchestra music data of Figure 1. Then, we found the following results of arbitrary power operation, as shown in Table 1. These results naturally suggest that the square operation in Property 3 is a kind of demodulation (DM) operation of the encoded 1/f fluctuations. Rapidly alternating positive and negative values in the original sound data cancel each other out, leaving no information about the encoding. Suppose that two signals with near frequencies (0<λ≪ω) are superposed,(1)g=sin((λ+ω)t)+sin((−λ+ω)t)=2cos(λt)︸A(t)sin(ωt), and the low-frequency signal A(t) is on the sine transportation wave: A(t)sin(tω). The n-th power of this becomes (n=1,2,…)(2)g=A(t)sin(tω)g2=12A(t)21−cos(2tω)g3=14A(t)33sin(tω)−sin(3tω)g4=18A(t)43−4cos(2tω)+cos(4tω)g5=116A(t)510sin(ωt)−5sin(3tω)+sin(5tω)…. The isolated term of A(t)n exists in the even-n-th power, while A(t)n mixes with the transportation wave in the odd-n-th power. The binomial expansion sin(x)n=((eix−e−ix)/(2i))n generally proves this property. Furthermore, this even/odd-power property is also true for signals that include overtones in their timbre. The similar expansion (sin(x)+sin(2x))n proves this fact. Thus, we can naturally infer that the origin of 1/f fluctuation is the wave beat or the amplitude modulation (AM). We further hypothesize that 1/f fluctuations appear after any demodulation operation (DM). We verified other DM operations, such as absolute value and thresholding based on the mean absolute value of the data, as shown in Table 2. All possible DMs applied to the original sound data appear to extract 1/f fluctuations successfully. If AM-DM were the origin of 1/f fluctuations, Properties 1 and 2 are naturally obtained as a result of the wave beat associated with AM. Although the original data, as shown in Equation (1), exhibits only two peaks in the PSD at the original frequencies, the square of the data exhibits an extra beat peak at an arbitrary low frequency at 2λ. The beat frequency λ can be arbitrarily small. A typical example is the oldest electric instrument, the theremin, which has two nearly identical high-frequency generators: One is of fixed frequency, and the other is slightly controlled by adjusting the capacitance between the player’s hand and the antenna. The wave beat between them yields an audible sound. As pointed out in Property 3 above, music data often exhibits pink noise when squared, and even when analyzing the PSD of the zero-crossing time series derived from the original data’s zero crossings, it still appears to show pink noise. The former can be explained by frequency accumulation causing beat effects, AM. On the other hand, the latter zero-crossing time series is equivalent to frequency modulation (FM). Based on Equation (1), let us consider FM as follows. The data g slowly undulates due to the AM of A(t). Near the peaks of this undulation, the zero crossings occur at roughly constant intervals. However, at troughs of this undulation, A(t) slowly changes its signature, causing the zero crossing intervals nonuniform in that vicinity. In other words, the AM created by the linear superposition of multiple components is strongly coupled to the phase. Thus, the zero-crossing mechanism can be considered as an operation that extracts the nodes (envelope sign reversal) of AM. Consequently, AM and FM are consistent with each other generating 1/f fluctuations. However, a simple beat is not sufficient to yield a power-law PSD. We further need a systematic frequency accumulation for 1/f fluctuations. The most typical accumulation of frequencies would be the exponential form: ω=e−κξ, where κ is a constant and ω is the frequency [4]. The positive variable ξ is uniformly random within some range with a constant probability p. Then we immediately have the frequency distribution function inversely proportional to the frequency ω:(3)P(ω)=pdξdω=pκ1ω, and the frequency difference (Δω) distribution function Q(Δω) becomes(4)Q(Δω)=∫ω1ω2dωP(ω+Δω)P(ω)=p2λ2Δωlnω2ω1+Δωω1ω2+Δω, which is inversely proportional to Δω, if we set the lower limit ω1 of the integration sufficiently small. Further, in the case of power law frequency accumulation ω=t−α, the frequency difference distribution function Q(Δω) becomes Q(Δω)∝Δω−1−(2/α): the fiducial power is still −1 [4]. Based on this principle, we superpose many (N) sine waves with frequencies approaching each other exponentially,(5)ϕ(t)=∑i=1Nsin2πω1+ce−ξit, where ξi is a uniform positive random variable, and ω,c are constants. Numerical demonstration shows that the PSD of this wave superposition itself ϕ(t) does not display 1/f fluctuations, while ϕ(t)2 does. Similar frequency accumulation is also expected for the power-law approach [4]. 3. Synchronization Mechanism: Orchestral Unison and the Stochastic Kuramoto Model To model how 1/f fluctuations emerge from the collective behavior of multiple musical performers in unison, we employ a stochastic version of the Kuramoto model (SKM). This model captures the essential dynamics of weakly coupled oscillators that tend to synchronize their phases despite individual differences in natural frequencies and noise perturbations. 3.1. The Kuramoto Framework The classical Kuramoto model [13] describes the evolution of the phase θi(t) of N coupled oscillators as:(6)dθi(t)dt=ωi+KN∑j=1Nsin(θj(t)−θi(t)), where ωi is the natural frequency of the i-th oscillator, and K is the global coupling strength. In a musical analogy, each θi represents the phase of a performer’s sound production, and the coupling term represents mutual auditory adjustment among performers. The collective mode in the Kuramoto model is the mean of the phases:(7)reiψ=1N∑i=1Neiθi. For the large N limit, the Kuramoto model has a critical point Kc. The system does not synchronize at all (r=0) for KK yields smaller power indices, resembling Brownian motion, while stronger synchronization N