Condition for $1/f$ noise to occur along with an example for a diffusion equation

Technical Summary: Condition for 1/f Noise to Occur Along with an Example for a Diffusion Equation

Problem Statement
Despite the ubiquity of $1/f$ noise across diverse physical systems—from electronic devices and fluid flows to biological and astronomical phenomena—the underlying physics governing its occurrence remains uncertain. While $1/f$ noise is often defined by a power-law spectral intensity $I(f) \propto 1/f^\alpha$ with $\alpha \approx 1.0$, the wide variety of sources suggests it is not universal in a strict physical sense. However, the authors posit that a statistical universality may exist. A primary challenge in the field is identifying a necessary and sufficient condition for stationary $1/f$ noise that explains its prevalence without relying on system-specific details. Furthermore, existing models often require specific assumptions (e.g., specific pulse shapes or optimized distributions) that may not be fundamental.

Methodology
The paper employs a dual approach combining analytical derivation with general statistical formulation:

1. Analytical Example: The authors analyze a standard one-dimensional diffusion equation ($\partial q/\partial t = D \partial^2 q/\partial x^2$) with a stationary boundary condition at $x=0$. The boundary quantity $q_0(t)$ is assumed to fluctuate with an exponentially decaying correlation function (Lorentzian spectrum). The study derives the spectral intensity of the diffusion flux $j_0(t)$ at the boundary.
2. General Formulation: Drawing on scaling theories from fluid turbulence (specifically the work of Perry and Abell), the authors construct a general statistical argument. They consider a stationary random process characterized by a mean amplitude and two distinct constant timescales (or frequencies), $\tau_{small} \ll \tau_{large}$ (or $f_{low} \ll f_{high}$).
3. Application to Pulse Models: The derived condition is applied to the "random superposition of pulses" model, a common framework for $1/f$ noise in electrical devices. The authors test whether specific pulse shapes or optimized distributions are strictly necessary under their proposed condition.

Key Contributions and Results

• Analytical Derivation of $1/f$ Noise in Diffusion:
  The authors demonstrate that for a diffusion process where the boundary concentration $q_0(t)$ has a Lorentzian spectrum (characterized by a correlation time $\tau_c$), the resulting diffusion flux $j_0(t)$ exhibits $1/f$ noise.

  • The spectral intensity of the flux, $I_{j_0}(\omega)$, is derived as proportional to $\omega / (1 + (\omega\tau_c)^2)$.
  • In the frequency range $\omega \gg 1/\tau_c$, this scales asymptotically as $I_{j_0}(\omega) \propto 1/\omega$ (or $1/f$).
  • Crucially, this result arises without introducing ad-hoc parameters; the scaling is determined solely by the diffusion coefficient $D$, the mean amplitude of the boundary fluctuations, and the correlation time $\tau_c$.

• Necessary and Sufficient Condition:
  The paper formulates a general condition for the occurrence of stationary $1/f$ noise:

  1. The noise process must be characterized by a mean amplitude $j_c$.
  2. The fluctuations must be defined by two constant frequencies (or timescales), $f_{low} \ll f_{high}$ (or $\tau_{large} \gg \tau_{small}$).
  3. Crucially, there must be no other constant parameters within the frequency range $f_{low} \ll f \ll f_{high}$ that influence the shape of the spectral intensity.

  Under these conditions, dimensional analysis dictates that the spectral intensity must scale as $I(\omega) \propto j_c^2 / \omega$. The presence of the two cutoffs ensures the integral of the intensity (total power) remains finite, while the absence of other parameters forces the $1/f$ scaling in the intermediate range.

• Generalization of Pulse Models:
  Applying this condition to the random pulse model, the authors show that $1/f$ noise emerges from a random superposition of pulses regardless of the specific pulse shape (e.g., exponential or Gaussian) or the specific distribution of pulse widths, provided the distribution spans a wide range between $\tau_{small}$ and $\tau_{large}$ and no other constants intervene. This generalizes previous models that required "optimal" distributions or specific shapes.

Significance and Claims
The paper claims that the proposed condition is statistical, simple, and rigorous, offering a unified framework for understanding the ubiquity of $1/f$ noise.

• Ubiquity: The simplicity of the condition—requiring only two timescales and no other constants—explains why $1/f$ noise appears in such a diverse array of systems (fluids, electronics, geophysics) despite their differing physical mechanisms.
• Specificity: The condition also explains why $1/f$ noise is absent in some similar systems; if the separation between timescales is insufficient or if an additional constant parameter exists within the frequency band, the $1/f$ scaling breaks down.
• Extension to $\alpha \neq 1.0$: The authors suggest that while the condition strictly yields $\alpha = 1.0$, deviations ($\alpha \neq 1.0$) observed in some systems may be attributed to intermittency (spatial or temporal inhomogeneity of fluctuations), analogous to corrections in fluid turbulence scaling laws. In such cases, the statistical condition holds as an approximation.

The study concludes that the physics of $1/f$ noise can be understood through this statistical framework, independent of the specific microscopic details of the system, provided the scaling constraints are met.