The turbulent energy is an important quantity here. Vortex lines contribute two energy components to the system (see Fig. 1), the loss of condensation energy in the vortex core, where the superfluidity is suppressed (negligible in our case), and the kinetic energy in the flow fields surrounding the cores. In earlier experiments on quantum turbulence the quantity usually inferred from experiment is the average vortex-line density L (the vortex-line length per unit volume)2. Comparison with the velocity-field characterization of classical turbulence usually relies on a phenomenological model3 equating the line density to the mean square vorticity of classical turbulence: 〈ω2〉=κ2L2, with κ the circulation quantum. An effective kinematic viscosity ν is also introduced, enabling the dissipation per unit mass to be written ε=νκ2L2 in analogy with the classical turbulence dissipation ε=νω2〉, where ν is the classical kinematic viscosity2,3 (the true viscosity of a pure superfluid being identically zero). Although this model has not been directly tested, it agrees qualitatively with measurements of the vortex line density L(t) for decaying quantum turbulence4 in superfluid 3He–B and in superfluid 4He (refs 3, 5), allowing a determination of ν. In contrast, a direct measurement of the energy dissipated in quantum turbulence allows us to test the standard model of decaying turbulence independently of this phenomenological model.

Figure 1: The energy distribution in pure quantum turbulence.
figure 1

At the T=0 limit, a coherent condensate carrying turbulence has only two contributions to the energy; first, the loss of condensation energy in the vortex core (in yellow) arising from the suppression of the superfluidity along the linear singularity of the vortex, and second, the kinetic energy of the surrounding flow circulating the core, as indicated by the arrows.

In the experiment presented here, the energy dissipation is measured directly with a ‘black-body radiator’ (BBR; ref. 6) the main features of which are shown in Fig. 2. The BBR is a thin-walled box with a small orifice in one side, immersed in superfluid 3He–B, and cooled by nuclear refrigeration. Inside the radiator is a thermometer wire resonator, a heater wire resonator to inject power, and a very low-amplitude resonating grid, which generates the turbulence by the initial production of microscopic vortex rings, as discussed in the Methods section. The grid resonator consists of a 5×5 mm goalpost shaped Ta wire carrying a 5×3.5 mm Cu grid mesh of 197 lines cm−1.

Figure 2: The quasiparticle black-body radiator.
figure 2

The BBR is indeed a black-body radiator, but for quasiparticle excitations rather than conventionally for photons. Inside the ‘box’, one vibrating-wire resonator acts as a thermometer by probing the quasiparticle excitation flux. A second acts as a heater by injecting power directly into the superfluid when accelerated beyond the critical velocity for pair breaking. The grid oscillates at 1 kHz, but with negligible amplitude (of order hundreds of nanometres), to create the turbulence. The box acts a bolometer measuring the energy released by the decaying turbulence inside the enclosure.

The experiments are performed in the low-temperature ballistic regime below 200 μK, where the thermal quasiparticle excitation mean free path is greater than the container dimension. (All data presented here were taken at a pressure of 3.3 bar, where the superfluid transition temperature, TC, is 1.32 mK.) Heat entering the radiator from any source produces ballistic quasiparticles, which thermalize by scattering off the walls, finally emerging as a beam of excitations from the orifice. At steady state, the power emitted in the beam balances the power entering the radiator. From simple kinetic theory, the power emitted6 is given by , where 〈n vg〉 is the thermal flux of excitations of mean energy , Δ is the superfluid energy gap, and A the effective area of the orifice. The damping on the thermometer wire is dominated by quasiparticle scattering, which being well understood7 provides very sensitive thermometry8. The thermal flux of excitations can therefore be inferred from the measured damping on the thermometer wire9, and from this quantity we obtain the power leaving the BBR (as described in the Methods section below). When turbulence is present, the excess power leaving the BBR provides a direct measure of the energy being released by the freely decaying turbulence, and this allows us to reconstruct the energy content of the original turbulence.

Figure 3 shows the excess power leaving the BBR as a function of time after switching off the drive to the grid. This data was taken for an initial grid velocity of 4.51 mm s−1. While driven, the grid generates both vortices and large numbers of quasiparticle excitations10. In the absence of vortices, we expect the excess power to decay exponentially with the BBR time constant, δQ̇=Q̇0exp(−t/τi). This is estimated from simple kinetic theory as τi4V/(〈vgA), with V the volume of the BBR and 〈vg〉 the mean excitation group velocity8. The BBR dimensions were designed to give a time constant of order τi0.5 s. A series of similar measurements were made for a range of initial grid velocities. All the data presented here were taken at 3.3 bar.

Figure 3: The excess power leaving the BBR as a function of time after stopping the grid.
figure 3

While the grid is oscillating to produce turbulence, simultaneous pair breaking also produces a large excess of quasiparticle excitations. These are rapidly emitted through the orifice into the bulk superfluid outside, at a rate governed by the BBR time constant τ (see text). The measured early-time behaviour is shown inset, marking an exponential recovery with time constant close to the design value for the BBR, as shown by the solid lines. Thus the temperature in the box falls to its very low, equilibrium temperature in a few seconds. Further heat emission then represents the decay of the quantum turbulence in the condensate, which is now cold again, as shown by the late-time behaviour in the main figure. The initial grid velocity was 4.5 mm s−1. The right-hand scale shows the equivalent temperature difference between the liquid in the BBR and the bulk.

The green lines in the figure show the initial exponential recovery with an intrinsic time constant of τi=0.54 s, in good agreement with the design value. At later times the excess power decays much more slowly, revealing a much longer-lived source of excitations in the BBR. This is the dissipation of the quantum turbulence produced by the grid. We can subtract the initial thermal recovery, δQ̇=Q̇0exp(−t/τ), to leave only the contribution from the slowly decaying vortices. This is shown in Fig. 4, where we plot the dissipation from the turbulence as a function of time for various initial grid velocities. We note that the thermal recovery is only significant during the first few seconds.

Figure 4: The power dissipated by decaying quantum turbulence produced by the grid.
figure 4

The figure shows the power dissipated for various initial grid velocities. (Below an initial grid velocity of 2 mm s−1 there is essentially no turbulence signal seen above the noise, and we take this as the critical velocity for turbulence production.) The solid green lines show the expected behaviour for classical homogeneous isotropic turbulence having a Kolmogorov spectrum with a Kolmogorov constant C of 2.0.

From earlier measurements11, confirmed by computer simulations12, we know that at low grid velocities the generated vortex rings travel ballistically. The rings have a self-propagation speed13 of around 10 mm s−1, corresponding to diameters of 5 μm. In the current experiment, such rings would collide with the BBR walls after 0.5 s. As we expect these rings to dissipate rapidly after colliding with a wall, they will not contribute to the later-time dissipation shown in Fig. 4. Above some critical grid velocity the ring density becomes sufficient for collisions, and reconnecting rings then rapidly evolve into quantum turbulence11,12. The critical velocity for this grid is found to be 2 mm s−1, because there is no measurable turbulence dissipation below this velocity.

We now compare the results with expectations from classical turbulence. According to the standard model1,2, the Richardson cascade transfers energy from large to small length scales (low to high wavenumbers, k) over an inertial range where viscosity plays no role. The energy is mostly contained in large eddies of length scale le, but is dissipated by viscous forces at the smallest scales. The energy per unit wavenumber per unit mass in the inertial range is predicted by the Kolmogorov spectrum Ek=C ε2/3k−5/3, with C a constant of order unity. Assuming that this spectrum spans the full range of length scales from the cell size (le=d10 mm) to the smallest (dissipative) scale a, the total turbulent energy is obtained by integrating , where V is the volume of the radiator, kd=2π/d and ka=2π/a. Assuming ad, the dissipation Q̇=ρ V ε is independent of the dissipative scale a and given by:

where E0 is the initial turbulent energy.

Thus the predicted late-time decay of the dissipation should be proportional to t−3, and independent of all fluid properties other than the density. Thus the only unknown parameter is the Kolmogorov constant C. Therefore a measurement of the dissipated energy gives a particularly direct test of this model.

The behaviour at earlier times depends also on the initial energy E0, which determines t0 and introduces a second free parameter. The solid lines in Fig. 4 show the behaviour predicted by equation (1) with a Kolmogorov constant of C=2.0. The data at higher grid velocities are seen to agree very well with the model, with initial energies of order 100 pJ. The Kolmogorov constant extracted from the fits, C2.0±0.4, falls within the range of typical values attributed to classical fluids14. The behaviour differs at the lower velocities (3.79 mm s−1 and below), where the late-time behaviour is consistent with a t−2 dependence, as expected for a random tangle with no large scale structure2. (Here the rings have reconnected to form a tangle, but at too low a density to develop a Kolmogorov spectrum). A similar crossover from a random tangle to classical-like turbulence has also been inferred from measurements of the line density in both superfluid 3He–B and superfluid 4He at low temperatures4,5,15.

Our data at high grid velocities agree well with the standard model of classical turbulence. However, there are caveats. First, in practice the initial energy spectrum will depend on the turbulence–production process. To explain towed-grid measurements in superfluid 4He, a more elaborate model3 has been developed incorporating an initial k2 energy spectrum at large length scales and a time-dependent energy-containing length scale le(t) that grows until it saturates at the container size d, after which the behaviour follows equation (1). This model does not yield substantially improved fits to our data, suggesting that the initial conditions do not critically influence the late-time behaviour. Second, the initial turbulence is clearly not isotropic or homogeneous16. Nevertheless, we observe very similar behaviour with wire-generated turbulence, even though the initial turbulence distributions must differ substantially, which suggests that the initial homogeneity does not affect the later-time behaviour. Third, one should question whether the vortex-line density in the BBR is large enough to provide an inertial range sufficient to support a Kolmogorov spectrum, but we note that measurements of the line density in other experiments4,17 suggest that there are no large deviations from the simple model at very low densities, even when the spatial extent d corresponds to just a few line spacings. Finally, as the initial temperature in the BBR, whilst the turbulence is being generated, is quite high (0.25TC or 330 μK for the highest velocity data), mutual friction may influence the production process and the initial behaviour18. Fortunately, this effect rapidly becomes negligible, as the radiator promptly cools in the first few seconds.


The mechanism for generating quantum turbulence. From earlier work11, we know that when we oscillate an object in the superfluid, the flow fields lead initially to the generation of microscopic vortex rings. This process gives rise to a cloud of similar-sized rings leaving the moving object, with the rate of creation increasing with the amplitude of the oscillatory flow. Under our conditions, the rings are a few micrometres in diameter and have velocities of 10 mm s−1, which is fast on the scale of the BBR dimensions. The subsequent history of this gas of rings depends on the density. At low densities the rings travel independently and rapidly leave the generation site. At some critical density, however, the ring density is high enough that they collide and reconnect to form larger, slower structures, trapping more rings, and turbulence rapidly develops.

The generation process by the grid is thus very different from that in a classical towed-grid scenario. In the present case, the grid is essentially stationary, as the amplitude of motion at the highest grid velocities is only a few hundreds of nanometres, thus generating no significant large-scale motion in the liquid. The specific geometry of the grid essentially plays no role, as the grid is only acting as a coarse surface for assisting the generation of the gas of rings. It is perhaps better to think of this device more as a stationary planar transducer (emitting vortex rings instead of sound) rather than as a moving grid as used in classical flow experiments. Nevertheless, this flat transducer provides us with the closest we can currently get to homogeneous quantum turbulence.

Operation and calibration of the black-body radiator. The thermal damping on the thermometer wire9 is given by Δf2T=B exp(−Δ/kBT). At our temperatures (200 μK), ΔkBT, the exponential is changing very rapidly, providing an extremely precise value of T (at 170 μK a change of only 1 μK gives rise to a change in the damping of nearly 10%). Δf2T is also proportional to 〈n vg〉. Thus we have the quantities needed to determine the beam power, . The beam power can thus be rewritten as Q̇beam=c WT, where (which we designate the ‘width parameter’ of the thermometer wire), T and have been obtained directly from Δf2T, as above, and c is a calibration constant to be determined6.

On injecting large powers, the bulk superfluid outside the BBR also warms slightly, yielding an influx of excitations back through the orifice from the outside. This is accounted for by subtracting the width parameter W measured by a similar thermometer wire located outside the BBR, close to the orifice, but shielded from the excitation beam by a paper screen. The net power leaving the radiator is thus given by5out=c δ W, where δ W=WTW. In the absence of additional sources of heating, the power leaving the radiator balances the steady-state heat leak into the BBR, Q̇leak=c δ W0.

The radiator is calibrated by measuring the increase in the width parameter ΔW=δ Wδ W0 while a measured power Q̇ap is injected into the radiator with the heater wire. A linear response is observed from which the calibration constant c=Q̇apW is deduced. Once calibrated, the width parameter directly measures the net power leaving the radiator. The background heat leak is found to be Q̇leak2 pW, resulting in a base temperature inside the BBR of 172 μK (0.13TC) whilst the surrounding superfluid cools to 145 μK. To determine the energy being dissipated by turbulence in the BBR, we correct for the heat leak and refer to the excess power, leaving the BBR, δQ̇=Q̇out−Q̇leak.