Growing interfaces in the Kardar-Parisi-Zhang universality class develop roughness over time — the surface height at a fixed point fluctuates as material is deposited. These fluctuations are correlated in time: the height now depends on the height a moment ago. The question is what the temporal power spectrum looks like.
This paper answers it for discrete KPZ models in the stationary regime. The power spectrum follows 1/f^α scaling with a spectral exponent of exactly 5/3. This is not 1/f noise (exponent 1) or white noise (exponent 0) or Brownian motion (exponent 2), but a specific intermediate value determined by the KPZ scaling exponents.
Note (2026-11-15; read against the paper's full text): the essay says the spectral exponent is 'exactly' 5/3 and 'determined by the KPZ scaling exponents'. The paper's wording is that it 'observes' 1/f^alpha-type scaling with spectral exponent 5/3 in simulations of discrete KPZ models, for small systems where the stationary state can be reached, supported by scaling-theoretical arguments. The paper also reports that an earlier study by Takeuchi had already found a 5/3 exponent in simulations and in liquid-crystal turbulence experiments, but with a power spectrum that lacked a lower cutoff, implying non-stationary signals; the paper's contribution is the stationary regime, the lower cutoff that maintains constant power, the dynamic scaling of the correlation function, and wide-sense stationarity. The essay does not mention the small-system restriction or the earlier 5/3 result.
The correlation function is non-exponential and vanishes after a correlation time that diverges with system size. This creates a lower-frequency cutoff in the power spectrum that maintains constant power — the system has a finite memory set by its physical extent. Above this cutoff, the 5/3 scaling holds.
Crucially, the fluctuations are wide-sense stationary: the mean and autocorrelation depend only on time differences, not absolute time. This validates the use of the Wiener-Khinchin theorem — the power spectrum is genuinely the Fourier transform of the autocorrelation, not an artifact of nonstationarity. The KPZ class, famous for its anomalous spatial scaling, produces equally precise temporal scaling. The roughness has a rhythm, and the rhythm has an exact exponent.