Extended Data Fig. 3: Power spectral density of UED time series.

Spectra are computed as the square modulus of the discrete Fourier transform (DFT) and normalized so that the maximum value is unity. No detrending is applied. (a) The DFT of the monolayer WSe₂ Bragg peak data (main-text Fig. 2(g)): the trend is monotonic decay with higher frequency, as predicted by a purely Debye-Waller (DW) thermal response. (b) The DFT of the 2° WSe₂/MoSe₂ satellite peak data (main-text Fig. 2(a)): the trend contains two peaks, the DW contribution appearing at zero frequency and a second contribution that we attribute to the twisting-motion appearing in the bin 540 ± 90 GHz, overlapping with the central frequency of the best-fit dynamical model 510 ± 90 GHz. (c) The DFT of the 57° WSe₂/MoSe₂ satellite peak data (main-text Fig. 2(d)): the DW contribution is peaked at 160 ± 80 GHz, which is a windowing artifact indicating that the average change in diffraction intensity over all points in the time-scan is zero, i.e., the positive change mostly cancels the negative change. A second peak is present in the bin 830 ± 80 GHz, whereas the dynamical model of the experiment best fits 510 ± 90 GHz (the model spectrum is shown in main-text Fig. 2(f)): further discussion can be found in the Supplementary Information.