3

ing. We can extract quantitative information from these 0

data by fitting the points in the spectral region around 10
2
the Brillouin peaks to the damped harmonic oscillator -1 Ω
model54 . This involves the parameters Ω(q) (related to 10 4
the sound velocity by c(q) = Ω(q)/q, see Ref.51 ), and -2 Ω
Γ(q), which encode the characteristic frequency and in- 10
verse life-time (or broadening, full width at half maxi- -3 a) T=0.001
mum) of the sound excitations, respectively. 10
0
10 4 2
Ω Ω
A. Sound damping -1
10 Ωco
In Fig. 2 we show the (total) Γ as a function of the 3/2
-2 Ω
corresponding Ω, at the indicated values of tempera- 10 Ωco2

Γ(Ω)
ture. The Ω-dependence of these data is very com- b) T=0.01
plex, and strongly depends on T . (We use symbols of
the same color to identify the investigated temperatures, 0
10
and solid lines of the same color for mechanisms that 2
are not modified at different T -values.) At the lowest Ω
T = 10−3 (a), a clear crossover occurs between the high- -1 3/2
10 Ω
frequency disorder-controlled behaviour ∝ Ω255 , and a
Rayleigh-like scattering contribution, ∝ Ω4 , at lower fre- c) T=0.1
quencies56 . As already noticed, the crossover frequency -2
Ωco ≃ 1, is below the calculated Boson peak frequency,
10
0
ΩBP ≃ ΩIR ≃ 215 . Note that even at this very low T an- 10 2
harmonic interactions are obviously present and, for in- Ω
stance, still contribute to the thermal conductivity. Their
-1
intensity, however, is very low compared to other con- 10 3/2
tributions, while non-negligible effects should be visible Ω
at frequencies smaller than our spectral range. By in- d) T=0.2
creasing T , in contrast, we expect the strength of anhar- -2
10 0
Ω 10
monicities to increase, eventually entering the frequency
window.
This is indeed the case in (b) for T = 10−2 , where FIG. 2. Transverse sound broadening, Γ(Ω) (circles), de-
we detect a second T -dependent crossover, at Ωco2 ≃ termined as discussed in the main text as a function of the
0.6, between the Rayleigh region and a remarkable low- corresponding Brillouin frequency, Ω, at the indicated val-
frequency ∝ Ω3/2 regime57 , as theoretically predicted ues of T . The solid lines indicate the power-law scaling valid
in Ref.58 and reminiscent of the fractional attenuation in the different frequency regions, with strongly T -dependent
of Refs.58,59 (see below). Note that the latter is obvi- patterns as discussed in the text. We use the same color code
ously strongly T -dependent, whereas the Rayleigh and to indicate power laws which arise from the same mechanism
disorder-controlled regimes are not modified even at in- and are not modified at different values of T .
termediate T , a feature that we will exploit below. Also,
by increasing T , we expect the two crossover frequencies
to eventually merge Ωco2 ≃ Ωco , when the strength of the growths faster than the Ω3/2 region. Indeed, the pre-
anharmonic couplings becomes comparable to that asso- factor of the quadratic term δ2 (T ), shown in the inset of
ciated to the effect of the disorder, and the two mech- Fig. 3, keeps a constant value for T < 0.1, before sub-
anisms bury the Raleigh scaling in the entire Ω-range. stantially increasing (possibly linearly) at higher T .
This is exactly what we observe at T = 10−1 in (c), The above scenario is similar to that reported recently
where the ∝ Ω3/2 regime at low frequency directly joins in the experimental work of60 for a network glass (sodium
to the quadratic T -independent contribution at high Ω. silicate), although in that case, a plain ∝ Ω2 Akhiezer
Finally, at the highest T = 2 × 10−1 ≃ Tg /2 in (a), regime is reported instead of the fractional behaviour at
we observe a unique envelope of all scattering mecha- low frequencies and high T . This extremely complex
nisms which now scales uniformly as Ω2 (Akhiezer-like) situation definitely points to non-trivial effects due to
in most of the Ω-range, while a vestige of the Ω3/2 regime temperature on the sound waves propagation, which su-
is still detected at low Ω. It is worth to note that, at perimpose to T -independent effects of a completely dif-
this stage, the width is fully T -dependent, and the sec- ferent nature. In Ref.34 we have related the latter to
ond crossover Ωco2 shift towards lower frequency on in- the existence of local elastic heterogeneities. To obtain
creasing T , indicating that the Akhiezer-like Ω2 -regime additional quantitative insight we need at this point to