betaFeedback

WorksPhysics and Astronomy

Mathematical reformulation of the Kolmogorov-Richardson energy cascade in terms of vortex stretching

Abstract

In this paper, with the aid of direct numerical simulations (DNS) of forced turbulence in a periodic domain, we mathematically reformulate the Kolmogorov-Richardson energy cascade in terms of vortex stretching. By using the description, we prove that if the Navier-Stokes flow satisfies a new regularity criterion in terms of the enstrophy production rate, then the flow does not blow up. Our DNS results seem to support this regularity criterion. Next, we mathematically construct the hierarchy of tubular vortices, which is statistically self-similar in the inertial range. Under the assumptions of the scale-locally of the vortex stretching/compressing (i.e. energy cascade) process and the statistical independence between vortices that are not directly stretched or compressed, we can derive the −5/3-5/3 power law of the energy spectrum of statistically stationary turbulence without directly using the Kolmogorov hypotheses.

Introduction

Since the seminal study by Orszag and Patterson (1972) [20], the numerical integration (i.e. the direct numerical simulations, DNS) of the Navier-Stokes equation without any turbulence model has been playing important roles in turbulence researches; see Ref. [11] for a review of DNS of turbulence. One of the most important features unveiled by DNS of turbulence is that turbulence is not random but composed of coherent motions. In particular, recent DNS [7, 9, 17, 18] of turbulence at sufficiently high Reynolds numbers have revealed that there exists a hierarchy of coherent vortices in developed turbulence.

In the present study, we investigate the energy cascade picture [23, 5] in turbulence of an incompressible fluid with uniform density and kinematic viscosity ν\nu under periodic boundary conditions in three orthogonal directions; and, on the basis of the picture, we develop arguments on the global-in-time smooth solution of the Navier-Stokes equation. Since we need external force to drive flow in a periodic cube, we impose body force f\boldsymbol{f} which injects the kinetic energy at a large scale L\mathcal{L}. Here, we define the Reynolds number by

Re=ULν(1)Re = \frac{U\mathcal{L}}{\nu} \tag*{(1)}

with UU being the characteristic velocity directly driven by f\boldsymbol{f}. When ReRe is sufficiently high (more concretely, higher than O(104)O(10^4) [4]), turbulence is composed of vortices with various length scales. We reemphasize that these vortices are coherent in time and space. The hierarchy of coherent vortices and its sustaining mechanism of this statistically stationary turbulence in a periodic cube are rather simple [9]. If we visualize vortices by using the magnitude of vortices, we can only observe the forest of the smallest-scale vortices (Fig. 1 in Ref. [11]). This is because the vorticity is predominantly determined by the smallest-scale eddies. Therefore, we need a scale decomposition to capture the hierarchy of coherent vortices. More precisely, the band-pass filter (i.e. the Littlewood-Paley decomposition) of the Fourier components of the vorticity reveals a clear hierarchical structures of coherent tubular vortices (see Fig. 2 in Ref. [9]). The sustaining mechanism of the hierarchy of vortices is also simple. Tubular vortices at a given length (i.e. a given thickness) are stretched and amplified in straining motions around (pairs of) larger tubular vortices. Here, it is important that this generation process of smaller vortices by larger ones occurs locally in scale. More quantitatively, 2–8 times larger vortices stretch and generate vortices at a given scale (see § 2.3 for the present DNS results). In other words, this sustaining process of turbulence is consistent with the classical Kolmogorov theory (i.e. the Kolmogorov-Richardson energy cascade [5]). It is also important that the smallest length scale, namely the Kolmogorov length η\eta, is determined by the length for which the two time scales of the vortex stretching and viscous effects are balanced. Therefore, the hierarchy of vortices is sustained in the inertial range of the length scales between L\mathcal{L} and η\eta, but vortices cannot be amplified in the viscous dissipation range of the length scales smaller than η\eta.

The above-mentioned scale locality of the vortex stretching is also consistent with the fact that the strain-rate, U/LU/\mathcal{L}, induced by the largest scale, L\mathcal{L}, flow cannot overcome the viscous effects for the length scales smaller than the Taylor length λ\lambda.1 Note that λ≫η\lambda\gg\eta for high Reynolds numbers. It is also important that the ratio L/η\mathcal{L}/\eta increases with the Reynolds number as L/η∼Re3/4\mathcal{L}/\eta\sim Re^{3/4}. Hence, physically speaking:

Observation 1.1. The increase of ReRe with fixing ff (i.e. the decrease of ν\nu with fixing UU and L\mathcal{L}) simply adds the number of the levels of the hierarchy of vortices.

In other words, this physical picture may allow us to expect that a finite energy dissipation rate ϵ\epsilon can be sustained in the shear flow around thin coherent tubular vortices even in the limit ν→0\nu\to0.

Remark 1. Jeong and the first author [12] mathematically considered the balance between the viscous dissipation and the vortex stretching. More precisely, they prepared small-scale vortex blob and large-scale anti-parallel vortex tubes for the initial data (in other words, shear flows around thin coherent tubular vortices), and showed that the corresponding 3D Navier–Stokes flow creates instantaneous vortex-stretching. Using this stretching, they showed that the flows satisfy a modified version of the zeroth law (but very close to the actual one) in a uniform time interval which in particular implies enhanced dissipation. The zeroth law postulates that, under the normalization of the initial data (depending on ν\nu) ∥u0∥L2=1\lVert u_0\lVert_{L^2}=1, the mean

energy dissipation rate of the corresponding 3D Navier-Stokes flow uνu^\nu does not vanish as ν→0\nu\to0:

lim inf⁡ν→0ν⟨⟨∣∇uν∣2⟩⟩>0,\liminf_{\nu\to0}\nu\langle\langle\lvert\nabla u^\nu\rvert^2\rangle\rangle> 0,

where ⟨⟨⋅⟩⟩\langle\langle\cdot\rangle\rangle denotes some ensemble or long-time, space averages.

The main purpose of the present study is to mathematically describe the feature of solutions of the Navier-Stokes equation on the basis of these physical pictures of turbulence. More concretely, with the aid of the DNS of turbulence in a periodic cube (the details of which are given in § 2), first, we observe that the Navier-Stokes turbulence is mathematically smooth enough (§ 3). Second, we propose a mathematical description of the energy cascading process in developed turbulence (§ 4). Then, we derive the Kolmogorov −5/3-5/3 power law of the energy spectrum by employing this energy cascade picture, without directly using the Kolmogorov hypotheses. (§ 5).

Notation. We use A≲BA \lesssim B (equivalently, B≳AB \gtrsim A) if there is an absolute constant C>0C > 0 such that A≤CBA \leq CB. Then, we say A≈BA \approx B if A≲BA \lesssim B and B≲AB \lesssim A. Moreover, A≲γBA \lesssim_\gamma B means A≤CγBA \leq C_\gamma B with some constant CγC_\gamma depending on γ\gamma.

Numerical simulations

In this section, we summarize the numerical method and parameters, and we show the results of our DNS.

Numerical method

Numerical method. We numerically integrate the Navier-Stokes equation in T3:=(R/2πZ)3\mathbb{T}^3 := (\mathbb{R}/2\pi\mathbb{Z})^3:

{∂tu+u⋅∇u+∇p=νΔu+f,∇⋅u=0,u(t=0)=u0(2)\begin{cases} \partial_t u + u \cdot\nabla u + \nabla p = \nu\Delta u + f,\\ \nabla\cdot u = 0,\\ u(t=0)=u_0 \tag*{(2)} \end{cases}

with the external force,

f=(−sin⁡xcos⁡y,+cos⁡xsin⁡y,0).(3)f=(-\sin x\cos y,+\cos x\sin y,0). \tag*{(3)}

Here, u:[0,∞)×T3→R3u:[0,\infty)\times\mathbb{T}^3\to\mathbb{R}^3 and p:[0,∞)×T3→Rp:[0,\infty)\times\mathbb{T}^3\to\mathbb{R} denote the velocity and pressure of the fluid, respectively. We use the standard Fourier spectral method. Here, we numerically integrate, by the fourth order Runge-Kutta-Gill scheme, the vorticity equation in terms of the Fourier component ω^(K)\widehat{\omega}(K) of vorticity, which is defined by

ω(x)=∑K∈Z3ω^(K)eiK⋅xwith(Fu)(K)=u^(K):=1(2π)3∫T3u(x)e−iK⋅x dx.(4)\omega(x)=\sum_{K\in\mathbb{Z}^3}\widehat{\omega}(K)e^{iK\cdot x} \quad\text{with}\quad (\mathcal{F}u)(K)=\widehat{u}(K):=\frac{1}{(2\pi)^3}\int_{\mathbb{T}^3}u(x)e^{-iK\cdot x}\,dx. \tag*{(4)}

Note that we need to integrate only two components of ω^(K)\widehat{\omega}(K) because of the solenoidal condition K⋅ω^=0K\cdot\widehat{\omega}=0. For numerical efficiency, we use the fast Fourier transform to evaluate the nonlinear term in the vorticity equation in the real space. Here, we remove the aliasing errors by the phase shift method.

RunN3N^3ν\nukmax⁡⟨η⟩k_{\max}\langle\eta\rangle⟨Rλ⟩\langle R_\lambda\rangle
2-1K102431024^30.0023.0296
8-1K102431024^30.0089.3123
8-2K204832048^30.00820136

Table 1. Numerical parameters. N3N^3, the number of the Fourier modes; ν\nu, the kinematic viscosity; ⟨η⟩\langle\eta\rangle, the temporal average of the Kolmogorov length scale (kmax⁡=2N/3k_{\max}=\sqrt{2}N/3 denotes the maximum wavenumber); ⟨Rλ⟩\langle R_\lambda\rangle, the temporal average of the Taylor-length Reynolds number.

Numerical parameters

Since we fix the external force ff, we change the Reynolds number by changing ν\nu. We denote the spatial average of the energy dissipation rate by ϵ(t)\epsilon(t). Then, the Kolmogorov length scale is expressed by η(t)=ϵ(t)−1/4ν3/4\eta(t)=\epsilon(t)^{-1/4}\nu^{3/4}. We choose the number N3N^3 of the Fourier modes so that we can resolve the small-scale structures of size η(t)\eta(t). In the numerical simulations of turbulence, we use the indicator of the resolution kmax⁡⟨η⟩k_{\max}\langle\eta\rangle, where ⟨η⟩\langle\eta\rangle is the temporal average of η(t)\eta(t) and the maximum wavenumber kmax⁡=2N/3k_{\max}=\sqrt{2}N/3 is the present numerical scheme. Usually, the condition kmax⁡⟨η⟩>1k_{\max}\langle\eta\rangle>1 is recommended. In the present study, however, we adopt much larger value (say, O(10)O(10); see table 1) to resolve flow structures much smaller than ⟨η⟩\langle\eta\rangle. We also examine that the small-scale statistics are independent of artificial parameters by changing N3N^3 with a common value of ν\nu.

We choose the time increment of the numerical integration so that the CFL condition is satisfied. We set the integration time of each run to be longer than the turnover time of the largest eddies. We evaluate the development of turbulence by the Reynolds number

Rλ(t)=u′(t)λ(t)/ν(5)R_\lambda(t)=u'(t)\lambda(t)/\nu \tag*{(5)}

where λ\lambda is the Taylor length and u′(t)u'(t) (i.e. the standard deviation of a velocity component) denotes turbulence intensity. In homogeneous isotropic turbulence,

Rλ(t)=203νϵ(t) E(t)(6)R_\lambda(t)=\sqrt{\frac{20}{3\nu\epsilon(t)}}\,\mathcal{E}(t) \tag*{(6)}

where E(t)\mathcal{E}(t) is the kinetic energy per unit mass. In general, turbulence is fully developed when Rλ>140R_\lambda>140 [4]. Turbulence with lower RλR_\lambda, the scale separation between the largest scale, i.e. the forcing scale, and η\eta is insufficient.

In the following, we show our numerical results obtained by the simulation with the parameters given in table 1.

Energy transfer due to vortex stretching

Four wavenumber ranges

There are four distinct wavenumber ranges (or equivalently, scale ranges) in each of which the energy transfer due to vortex stretching exhibits different characteristics. To show this, we define the energy transfer due to vortex stretching (see the energy equation (28)) by

T(K,K′):=1K2∫T3ω(K)⋅∇u(K′)⋅ω(K) dx(7)T(K,K'):=\frac{1}{K^2}\int_{T_3}\boldsymbol{\omega}(K)\cdot\nabla\boldsymbol{u}(K')\cdot\boldsymbol{\omega}(K)\,dx \tag*{(7)}

where

ω(K):=F−1[χ[K/2,2K]F[ω]](8)\omega(K) := \mathcal{F}^{-1}\left[\chi_{[K/\sqrt{2},\sqrt{2}K]}\mathcal{F}[\omega]\right] \tag*{(8)}

and

u(K):=F−1[χ[K/2,2K)F[u]].(9)u(K) := \mathcal{F}^{-1}\left[\chi_{[K/\sqrt{2},\sqrt{2}K)}\mathcal{F}[u]\right]. \tag*{(9)}

Note that in the next section we will define ωk\omega^k by ω(2k)\omega(2^k). We plot T(K,K′,t)T(K,K',t) in turbulence at Rλ≃280R_\lambda\simeq280 (Run 2-1K) in Figure 1. This figure shows results at a single time. We have confirmed that the behaviors of TT are qualitatively independent of time, though the values of TT evolve in time.

Image not extracted

Figure 1. Energy transfer T(K,K′,t)T(K,K',t) to the Fourier modes at wavenumber KK from the Fourier modes at K′K' due to vortex stretching in the four different wavenumber ranges. In each plot, darker (thinner) curves correspond to larger KK. Red and blue vertical lines indicate K′=0.58K(=K/1.7)K' = 0.58K(=K/1.7) and K′=1.7KK' = 1.7K, respectively. Note that, from this observation, we assume α=1.7\alpha= 1.7 in § 4. (a) In the energy containing range K∈[1.2−34η−1,1.2−26η−1]=[0.0020η−1,0.0087η−1]K \in[1.2^{-34}\eta^{-1},1.2^{-26}\eta^{-1}] = [0.0020\eta^{-1},0.0087\eta^{-1}], where the Fourier modes (which correspond to the whiskers observed at the lower wavenumber) directly driven by external force. (b) In the inertial range (K∈[1.2−25η−1,1.2−20η−1]=[0.010η−1,0.026η−1])(K \in[1.2^{-25}\eta^{-1},1.2^{-20}\eta^{-1}] = [0.010\eta^{-1},0.026\eta^{-1}]), where we can observe a clear collapse of the curves for different KK, and the peaks are located at K′=0.58KK' = 0.58K and the valleys are at K′=1.7KK' = 1.7K. (c) In the dissipation range K∈[1.2−19η−1,1.2−10η−1]=[0.031η−1,0.16η−1]K \in[1.2^{-19}\eta^{-1},1.2^{-10}\eta^{-1}] = [0.031\eta^{-1},0.16\eta^{-1}], where we do not observe the self-similarity of the energy transfer. (d) In the far dissipation range K∈[1.2−9η−1,1.2−6η−1]=[0.19η−1,0.33η−1]K \in[1.2^{-9}\eta^{-1},1.2^{-6}\eta^{-1}] = [0.19\eta^{-1},0.33\eta^{-1}], where the energy is dissipated by viscosity and almost no energy is transferred to higher wavenumbers ∣T(K′>K)∣≪∣T(K>K′)∣|T(K' > K)| \ll|T(K > K')|. Results of Run 2-1K (Rλ≈280R_\lambda\approx280).

First, let us look at the results shown in Figure 1(a) for a low wavenumber range (K∈[0.0020η−1,0.0087η−1]K \in[0.0020\eta^{-1}, 0.0087\eta^{-1}]), where the effect of the external force is dominant. This figure shows the energy transfer T(K,K′,t)T(K,K',t) as a function of K′/KK'/K, where the energy transfers from wavenumber K′K' to KK. The different curves in the plot show the results for different KK; darker (thinner) curves correspond to larger KK. It is clear in this figure that the Fourier modes at KK acquire energy from smaller wavenumber modes (K′<KK' < K) and transfer the energy to larger wavenumber modes (K<K′K < K'). These characteristics are common in the four wavenumber ranges shown in Figure 1(a)–(Figure 1d). However, we observe in Figure 1(a) “whiskers” at the lowest wavenumber. This corresponds to the contribution from the Fourier modes which the external force ff directly drives. Recall that, in this turbulence, the external force is steady and therefore these modes are rather robust.

Secondly, we look at the inertial range. Figure 1(b) shows the energy transfer in the wavenumber range K∈[0.0105η−1,0.026η−1]K \in[0.0105\eta^{-1}, 0.026\eta^{-1}]. This figure also shows that the energy transfers from lower (K′<KK' < K) to higher K′>KK' > K wavenumber modes. In contrast to the lower wavenumber range [Figure 1(a)], the contribution from about the half (more precisely, K′=0.58KK' = 0.58K) wavenumber modes (i.e. 1.7 times larger scales) transfer the energy to KK than the forcing modes. Note that the peaks at K′=0.58KK' = 0.58K is higher than the whiskers. This is also consistent with the observation in the same figure [Figure 1(b)] that the energy at the wavenumber KK transfers most to a higher wavenumber mode at K′=1.7KK' = 1.7K (i.e. 0.58 times smaller scales). It is further important to observe that all the curves for different KK collapse almost perfectly. This means that the energy transfer due to vortex stretching occurs in a self-similar manner, and that the energy is conserved by this cascading process. In other words, the viscous effect is negligible in this wavenumber range (i.e. the inertial range). We may verify the self-similarity in Figure 2, where we show Kmax⁡′/KK'_{\max}/K and Kmin⁡′/KK'_{\min}/K as functions of KK. Here, Kmax⁡′K'_{\max} and Kmin⁡′K'_{\min} denote the wavenumbers which attain the maximum and minimum of the energy transfer:

Image not extracted

Figure 2. The wavenumbers Kmax⁡′K'_{\max} (gray symbols) and Kmin⁡′K'_{\min} (pink) giving the maximum and minimum of T(K,K′,t)T(K,K',t), respectively, as a function of KK. We plot the data at 20 different instants. The red and black dots show the temporal averages of Kmax⁡′K'_{\max} and Kmin⁡′K'_{\min}, respectively. The dotted and dashed lines indicate K′∼KK' \sim K and K′∼K1/2K' \sim K^{1/2}, respectively. The two blue vertical dashed lines indicate K=0.01η−1K=0.01\eta^{-1} and 0.03η−10.03\eta^{-1}. Results of Run 2-1K (Rλ≈280R_\lambda\approx280). We do not show Kmin⁡′K'_{\min} for Kη>0.19K\eta>0.19 because the numerical resolution is not sufficient to determine it.

Kmax⁡′(K,t):=argmax⁡K′ T(K,K′,t)andKmin⁡′(K,t):=argmin⁡K′ T(K,K′,t).(10)K'_{\max}(K,t) := \underset{K'}{\operatorname{argmax}}\,T(K,K',t) \quad\text{and}\quad K'_{\min}(K,t) := \underset{K'}{\operatorname{argmin}}\,T(K,K',t). \tag*{(10)}

We can see that Kmax⁡′∼KK'_{\max} \sim K and Kmin⁡′∼KK'_{\min} \sim K in the wavenumber range K∈[0.01η−1,0.04η−1]K \in[0.01\eta^{-1}, 0.04\eta^{-1}].

Thirdly, we look at the dissipation range. Figure 1(c) shows the energy transfer in a higher wavenumber range: K∈[0.031η−1,0.16η−1]K \in[0.031\eta^{-1}, 0.16\eta^{-1}]. The behaviors of the energy transfer are qualitatively different from those in the inertial range. (i) There is no collapse of the curves because the amount of energy transfer decreases with wavenumber, since the energy is dissipated due to viscosity. (ii) the valleys of the energy transfer located always at K′=1.7KK' = 1.7K, whereas the wavenumber K′K' giving the maximum behaves in a non-trivial manner (see Figure 2). This implies the disruption of the self-similarity, which is also reasonable because of the viscous effects.

Fourthly, we look at the far dissipation range. Figure 1(d) shows the energy transfer in K∈[0.19η−1,0.33η−1]K \in[0.19\eta^{-1},0.33\eta^{-1}]. The behaviors of the energy transfer in this high wavenumber range is further different from those in Figure 1(c). Though the Fourier modes acquire the energy form larger scales, almost no energy transfers to higher wavenumbers.

Thus, the energy transfer behaves differently in the four wavenumber ranges. The bounds of these ranges are located at K=0.01η−1K=0.01\eta^{-1}, 0.03η−10.03\eta^{-1} and 0.19η−10.19\eta^{-1}. These wavenumbers are shown in Figure 3, in which we plot the energy and its dissipation spectra:

Image not extracted

Figure 3. (a) Energy and (b) energy dissipation spectrum (Results of Run 2-1K; Rλ≈280R_\lambda\approx280). Red, blue and black dotted vertical lines indicate K=0.01η−1K = 0.01\eta^{-1} (the lower end of the inertial range) 0.03η−10.03\eta^{-1} (the lower end of the dissipation range) and 0.19η−10.19\eta^{-1} (the lower end of the far dissipation range). Black dashed lines indicate (a) −5/3-5/3 and (b) 1/31/3 power laws.

E(K,t):=∫K≤∣K′∣<K+1∣u^(K′,t)∣2 dΩandD(K,t):=νK2E(K,t)(11)E(K,t):=\int_{K\leq|K'|<K+1}|\hat{u}(K',t)|^2\,\mathrm{d}\Omega \quad\text{and}\quad D(K,t):=\nu K^2 E(K,t) \tag*{(11)}

The wavenumbers K=0.01η−1K=0.01\eta^{-1} and 0.03η−10.03\eta^{-1} correspond to the lower and higher ends of the −5/3-5/3 power law, respectively. We expect that the lower bound normalized by η−1\eta^{-1} depends on the Reynolds number, whereas the higher bound does not. Furthermore, the boundary K=0.19η−1K=0.19\eta^{-1} between the dissipation and far dissipation ranges is located at the peak of the energy dissipation spectrum. In summary, we can describe the scale-local energy transfer once we define it by using the enstrophy production rate as in (7).

Enstrophy production rate in the far dissipation range

The results in the previous subsection clarify the usefulness of the energy transfer T(K,K′,t)T(K,K',t) defined by the enstrophy production rate in the description of the energy cascading process. This encourages us to estimate the maximum enstrophy production rate in the far dissipation range in order to investigate the possibility of the blowup of solutions to the Navier-Stokes equation (see § 3).

For this purpose (i.e. to investigate the dynamics in the far dissipation range), we use the DNS results with larger ν\nu (Runs 8-1K and 8-2K). Recall that we have conducted DNS with a common value of ν\nu but with different resolutions N3N^3. We plot in Figure 4 the maximum

Image not extracted

Figure 4. Maximum PsupP_{\mathrm{sup}}, defined by (12), of the normalized enstrophy production rate. We show the results with two different numerical resolutions with a common value of the kinematic viscosity ν\nu; gray symbols, N3=10243N^3 = 1024^3, Run 8-1K, Rλ≈120R_\lambda\approx120; black symbols, N3=20483N^3 = 2048^3, Run 8-2K, Rλ≈130R_\lambda\approx130. Solid black line indicates β=1\beta= 1.

Psup(K,t):=sup⁡K′P(K,K′,t)(12)P_{\mathrm{sup}}(K,t) := \sup_{K'} P(K,K',t) \tag*{(12)}

of the normalized enstrophy production rate

P(K,K′,t):=∫T3ω(K,x,t)⋅∇u(K′,x,t)⋅ω(K,x,t) dx∥ω(K,t)∥L22∥ω(K′,t)∥L2.(13)P(K,K',t) := \frac{\int_{\mathbb{T}^3}\omega(K,x,t)\cdot\nabla u(K',x,t)\cdot\omega(K,x,t)\,\mathrm{d}x}{\lVert\omega(K,t)\rVert_{L^2}^2\lVert\omega(K',t)\rVert_{L^2}}. \tag*{(13)}

The different symbols in the figure denote the results with different resolutions. We may confirm that the results are statistically independent of the resolution N3N^3, although the instantaneous values are, of course, dependent on the runs. In particular, we observe that the temporal fluctuation of PsupP_{\mathrm{sup}} is quite large in the far dissipation range (K>0.1η−1K > 0.1\eta^{-1}), but the amplitude of the fluctuations is also independent of the resolution.

The solid straight line shown in this log-log plot indicates β=1\beta= 1 in the main theorem. The plotted function seems a concave function of KK in the far dissipation range, in particular, it seems to satisfy

Psup(K)≤CKβP_{\mathrm{sup}}(K) \le CK^\beta

for some β<1\beta< 1 and C>0C > 0 which is independent of KK. This result encourages us to develop the mathematics in terms of the energy transfer due to the vortex stretching. However, the estimation of PsupP_{\mathrm{sup}} in the further higher dissipation range is numerically challenging because of the finite digits of the numerical accuracy, and we must leave this for a future numerical study.

Global existence of smooth flow on the NS turbulence.

In this section, with the aid of concavity of the normalized enstrophy production rate PsupP_{\mathrm{sup}} (defined in (12)) in the far dissipation range, we show that the corresponding solution to the Navier-Stokes equation (2) is smooth enough. Let us define the inhomogeneous Sobolev spaces HsH^s as follows:

Hs(T3):={u=∑K∈Z3u^(K)eiK⋅x∣∥u∥Hs:=(∑K∈Z3(1+∣K∣2)s∣u^(K)∣2)1/2<∞}.H^s(\mathbb{T}^3) := \left\{u = \sum_{K\in\mathbb{Z}^3} \widehat{u}(K)e^{iK\cdot x} \mathrel{\Big|} \lVert u\rVert_{H^s} := \left(\sum_{K\in\mathbb{Z}^3}(1+\lvert K\rvert^2)^s\lvert\widehat{u}(K)\rvert^2\right)^{1/2}<\infty\right\}.

and let us define the band-pass filter (Littlewood-Paley decomposition) as follows:

ωk:=F−1[χ[2k2,22k]F[ω]].\omega^k := \mathcal{F}^{-1}\left[\chi_{\left[\frac{2^k}{\sqrt{2}},\sqrt{2}2^k\right]}\mathcal{F}[\omega]\right].

where χ\chi is the characteristic function. Then, if the vorticity ω\omega is mean zero, we can decompose it as follows:

ω=∑k∈Z≥−1ωk\omega=\sum_{k\in\mathbb{Z}_{\geq-1}}\omega^k

and we see

∥ω∥Hs≈(∑j∈Z≥−122sj∥ωj∥L22)1/2.\lVert\omega\rVert_{H^s}\approx\left(\sum_{j\in\mathbb{Z}_{\geq-1}}2^{2sj}\lVert\omega^j\rVert_{L^2}^2\right)^{1/2}.

Now we decompose the vortex stretching term as follows:

∫T3(ω⋅∇)u⋅ω=∑k∈Z≥−1∑m∈Z≥−1∑ℓ∈Z≥−1∫T3(ωℓ⋅∇)um⋅ωk=∑k∈Z≥−1∑m∈Z≥−1∑ℓ∈Z≥−1P~(2k,2m,2ℓ,t)∥ωℓ∥L2∥ωm∥L2∥ωk∥L2,(14)\begin{aligned} \int_{\mathbb{T}^3}(\omega\cdot\nabla)u\cdot\omega=\sum_{k\in\mathbb{Z}_{\geq-1}}\sum_{m\in\mathbb{Z}_{\geq-1}}\sum_{\ell\in\mathbb{Z}_{\geq-1}}\int_{\mathbb{T}^3}(\omega^\ell\cdot\nabla)u^m\cdot\omega^k \\ &=\sum_{k\in\mathbb{Z}_{\geq-1}}\sum_{m\in\mathbb{Z}_{\geq-1}}\sum_{\ell\in\mathbb{Z}_{\geq-1}}\widetilde{P}(2^k,2^m,2^\ell,t)\lVert\omega^\ell\rVert_{L^2}\lVert\omega^m\rVert_{L^2}\lVert\omega^k\rVert_{L^2}, \tag*{(14)} \end{aligned}

where P~\widetilde{P} is defined as follows:

P~(2k,2m,2ℓ,t):=∫T3ωℓ(x,t)⋅∇um(x,t)⋅ωk(x,t) dx∥ωℓ(t)∥L2∥ωm(t)∥L2∥ωk(t)∥L2.(15)\widetilde{P}(2^k,2^m,2^\ell,t):=\frac{\int_{\mathbb{T}^3}\omega^\ell(x,t)\cdot\nabla u^m(x,t)\cdot\omega^k(x,t)\,dx}{\lVert\omega^\ell(t)\rVert_{L^2}\lVert\omega^m(t)\rVert_{L^2}\lVert\omega^k(t)\rVert_{L^2}}. \tag*{(15)}

Remark 2. By DNS, we have already observed that

P~sup∼Psup,\widetilde{P}_{\mathrm{sup}}\sim P_{\mathrm{sup}},

where

P~sup(K,t)=sup⁡K′,K′′P~(K,K′,K′′,t).\widetilde{P}_{\mathrm{sup}}(K,t)=\sup_{K',K''}\widetilde{P}(K,K',K'',t).

Physically, the coincidence of PsupP_{\mathrm{sup}} and P~sup\widetilde{P}_{\mathrm{sup}} is an important issue, thus, as the sequence of this study, we will provide another research article elsewhere.

We assume there exist β<1\beta< 1 and a constant C>0C > 0 (independent of kk) such that

sup⁡t∈[0,t∗)sup⁡ℓ,mP~(2k,2m⋅2ℓ,t)≤C2kβ(16)\sup_{t \in[0,t_*)}\sup_{\ell,m}\widetilde{P}(2^k,2^m\cdot2^\ell,t) \le C2^{k\beta} \tag*{(16)}

as long as the solution u(t)u(t) exists in t∈[0,t∗)t \in[0,t_*). By the DNS result in Subsection 2.3.2, it seems that the NS turbulence satisfies (16). Then we obtain the following regularity criterion.

Theorem 1. Let f∈L∞(0,∞;H−1(T3))f \in L^\infty(0,\infty;H^{-1}(\mathbb{T}^3)) and u0∈H1(T3)u_0 \in H^1(\mathbb{T}^3) with ∫T3f=∫T3u0=0\int_{\mathbb{T}^3} f = \int_{\mathbb{T}^3} u_0 = 0 and ∇⋅f=∇⋅u0=0\nabla\cdot f = \nabla\cdot u_0 = 0. Suppose that u∈C([0,t∗):H1(T3))u \in C([0,t_*):H^1(\mathbb{T}^3)) is a local-in-time strong solution to the three-dimensional Navier–Stokes equation (2). If β<1\beta< 1 for t∈[0,t∗)t \in[0,t_*), then uu can be extended to the strong solution up to the time t∗∗t_{**} with t∗∗>t∗t_{**} > t_*.

Remark 3. By applying a bootstrapping argument for regularity (see the proof of Theorem 2 in [27] for example) and another bootstrapping argument for existence time (see the proof of Theorem 4 in [30] for example), combining with the a-priori bound (3.5), we can show that the solution exists in an arbitrary time interval, and is smooth enough in both space and time, if the external force ff is smooth enough.

Proof. The proof is rather elementary. Our calculation is simply based on [1 Theorem 3.2] and [14 Section 3.3]. They considered global existence of solutions to the incompressible 3D Navier–Stokes equation with the Coriolis force in a periodic domain. The key point is to construct an a-priori estimate in H1H^1. First, we estimate (3.1). Let us choose δ>0\delta> 0 such that β+δ<1\beta+ \delta< 1. By Bony’s paraproduct formula (see [2] for example), combining Hölder’s inequality for sums, we have

∫T3(ω⋅∇)u⋅ω≤C∑k∈Z≥−1∑(ℓ,m)∈D1(k)2βk∥ωℓ∥L2∥ωm∥L2∥ωk∥L2+C∑k∈Z≥−1∑(ℓ,m)∈D2(k)2βk∥ωℓ∥L2∥ωm∥L2∥ωk∥L2+C∑k∈Z≥−1∑(ℓ,m)∈D3(k)2βk∥ωℓ∥L2∥ωm∥L2∥ωk∥L2,\begin{aligned} \int_{\mathbb{T}^3}(\omega\cdot\nabla)u \cdot\omega \le{} & C\sum_{k \in\mathbb{Z}_{\ge-1}}\sum_{(\ell,m)\in D_1(k)}2^{\beta k}\|\omega^\ell\|_{L^2}\|\omega^m\|_{L^2}\|\omega^k\|_{L^2}\\ &+C\sum_{k \in\mathbb{Z}_{\ge-1}}\sum_{(\ell,m)\in D_2(k)}2^{\beta k}\|\omega^\ell\|_{L^2}\|\omega^m\|_{L^2}\|\omega^k\|_{L^2}\\ &+C\sum_{k \in\mathbb{Z}_{\ge-1}}\sum_{(\ell,m)\in D_3(k)}2^{\beta k}\|\omega^\ell\|_{L^2}\|\omega^m\|_{L^2}\|\omega^k\|_{L^2}, \end{aligned}

where (this is well-known Bony’s paraproduct formula)

D1(k):={(ℓ,m)∈(Z≥−1)2:k−1≤m≤k+1, ℓ≤m−1},D2(k):={(ℓ,m)∈(Z≥−1)2:k−1≤ℓ≤k+1, m≤ℓ−1},D3(k):={(ℓ,m)∈(Z≥−1)2:ℓ≥k−1, ℓ−1≤m≤ℓ+1}.(17)\begin{aligned} D_1(k) &:= \{(\ell,m)\in(\mathbb{Z}_{\ge-1})^2:k-1\le m\le k+1,\ \ell\le m-1\},\\ D_2(k) &:= \{(\ell,m)\in(\mathbb{Z}_{\ge-1})^2:k-1\le\ell\le k+1,\ m\le\ell-1\},\\ D_3(k) &:= \{(\ell,m)\in(\mathbb{Z}_{\ge-1})^2:\ell\ge k-1,\ \ell-1\le m\le\ell+1\}. \tag*{(17)} \end{aligned}

Then direct calculations yield

≲∥ω∥L2∑k∈Z≥−12βk∥ωk∥L2∑ℓ≤k∥ωℓ∥L2+∥ω∥L2∑k∈Z≥−12βk∥ωk∥L2∑m≤k∥ωm∥L2+∑k∈Z≥−12βk∥ωk∥L2∑ℓ≥k−1∥ωℓ∥L2(∥ωℓ−1∥L2+∥ωℓ∥L2+∥ωℓ+1∥L2)≲∥ω∥L2(∑k∈Z≥−122(β+δ)k∥ωk∥L22)1/2(∑k∈Z≥−12−2δkk)1/2+∥ω∥L2(∑k∈Z≥−122(β+δ)k∥ωk∥L22)1/2(∑k∈Z≥−12−2δk)1/2≲δ∥ω∥Hβ+δ∥ω∥L22.\begin{aligned} &\lesssim\|\omega\|_{L^2}\sum_{k\in\mathbb{Z}_{\geq-1}}2^{\beta k}\|\omega^k\|_{L^2}\sum_{\ell\leq k}\|\omega^\ell\|_{L^2}\\ &\quad+\|\omega\|_{L^2}\sum_{k\in\mathbb{Z}_{\geq-1}}2^{\beta k}\|\omega^k\|_{L^2}\sum_{m\leq k}\|\omega^m\|_{L^2}\\ &\quad+\sum_{k\in\mathbb{Z}_{\geq-1}}2^{\beta k}\|\omega^k\|_{L^2}\sum_{\ell\geq k-1}\|\omega^\ell\|_{L^2}\left(\|\omega^{\ell-1}\|_{L^2}+\|\omega^\ell\|_{L^2}+\|\omega^{\ell+1}\|_{L^2}\right)\\ &\lesssim\|\omega\|_{L^2}\left(\sum_{k\in\mathbb{Z}_{\geq-1}}2^{2(\beta+\delta)k}\|\omega^k\|_{L^2}^2\right)^{1/2}\left(\sum_{k\in\mathbb{Z}_{\geq-1}}2^{-2\delta k}k\right)^{1/2}\\ &\quad+\|\omega\|_{L^2}\left(\sum_{k\in\mathbb{Z}_{\geq-1}}2^{2(\beta+\delta)k}\|\omega^k\|_{L^2}^2\right)^{1/2}\left(\sum_{k\in\mathbb{Z}_{\geq-1}}2^{-2\delta k}\right)^{1/2}\\ &\lesssim_\delta\|\omega\|_{H^{\beta+\delta}}\|\omega\|_{L^2}^2. \end{aligned}

Here we used the fact that

∑ℓ≤kaℓ≤(∑ℓ≤kaℓ2)1/2(∑ℓ≤k1)1/2\sum_{\ell\leq k}a_\ell\leq\left(\sum_{\ell\leq k}a_\ell^2\right)^{1/2}\left(\sum_{\ell\leq k}1\right)^{1/2}

for aℓ≥0a_\ell\geq0.

We just proceed the H1H^1 energy estimate as (recall ∥ω∥L2≈∥u∥H1\|\omega\|_{L^2}\approx\|u\|_{H^1} with ∫T3u=0\int_{\mathbb{T}^3}u=0)

ddt∥u(t)∥H12+2ν∥u(t)∥H22≲∥u(t)∥H12∥u(t)∥H1+β+δ+∥f∥H1∥u(t)∥H1.\frac{d}{dt}\|u(t)\|_{H^1}^2+2\nu\|u(t)\|_{H^2}^2\lesssim\|u(t)\|_{H^1}^2\|u(t)\|_{H^{1+\beta+\delta}}+\|f\|_{H^1}\|u(t)\|_{H^1}.

By interpolation and Young’s inequality (note β+δ≤1\beta+\delta\leq1),

∥u∥H12∥u∥H1+β+δ≲∥u∥H12∥u∥H2≲ν−1∥u∥H14+ν∥u∥H22,\|u\|_{H^1}^2\|u\|_{H^{1+\beta+\delta}}\lesssim\|u\|_{H^1}^2\|u\|_{H^2}\lesssim\nu^{-1}\|u\|_{H^1}^4+\nu\|u\|_{H^2}^2,
∥f∥H1∥u∥H1≲∥f∥H12+∥u∥H12,\|f\|_{H^1}\|u\|_{H^1}\lesssim\|f\|_{H^1}^2+\|u\|_{H^1}^2,

and hence,

ddt∥u(t)∥H12+ν∥u(t)∥H22≲(1+ν−1∥u(t)∥H12)∥u(t)∥H12+∥f∥H12,t>0.\frac{d}{dt}\|u(t)\|_{H^1}^2+\nu\|u(t)\|_{H^2}^2\lesssim(1+\nu^{-1}\|u(t)\|_{H^1}^2)\|u(t)\|_{H^1}^2+\|f\|_{H^1}^2,\qquad t>0.

On the other hand, we immediately have the energy inequality:

ddt∥u(t)∥L22+2ν∥u(t)∥H12≲∥f(t)∥H−1∥u(t)∥H1.\frac{d}{dt}\|u(t)\|_{L^2}^2+2\nu\|u(t)\|_{H^1}^2\lesssim\|f(t)\|_{H^{-1}}\|u(t)\|_{H^1}.

Thus, by the absorbing argument, we have

∥u(t)∥L22+ν∫0t∥u(t)∥H12≲ν−1∫0t∥f(t)∥H−12+∥u(0)∥L22.\|u(t)\|_{L^2}^2+\nu\int_0^t\|u(t)\|_{H^1}^2\lesssim\nu^{-1}\int_0^t\|f(t)\|_{H^{-1}}^2+\|u(0)\|_{L^2}^2.

By the Gronwall inequality, we then have

∥u(t)∥H12≲∥u0∥H12e∫0t(1+ν−1∥u(s)∥H12) ds+∫0te∫τt(1+ν−1∥u(s)∥H12) ds∥f(τ)∥H12 dτ≲∥u0∥H12exp⁡(t+ν−2∫0t∥f(s)∥H−12 ds+ν−1∥u0∥L22)+exp⁡(t+ν−2∫0t∥f(s)∥H−12 ds+ν−1∥u0∥L22)∫0t∥f(τ)∥H12 dτ<∞(18)\begin{aligned} \|u(t)\|_{H^1}^2 &\lesssim\|u_0\|_{H^1}^2 e^{\int_0^t(1+\nu^{-1}\|u(s)\|_{H^1}^2)\,ds}+\int_0^t e^{\int_\tau^t(1+\nu^{-1}\|u(s)\|_{H^1}^2)\,ds}\|f(\tau)\|_{H^1}^2\,d\tau\\ &\lesssim\|u_0\|_{H^1}^2\exp\left(t+\nu^{-2}\int_0^t\|f(s)\|_{H^{-1}}^2\,ds+\nu^{-1}\|u_0\|_{L^2}^2\right)\\ &\quad+\exp\left(t+\nu^{-2}\int_0^t\|f(s)\|_{H^{-1}}^2\,ds+\nu^{-1}\|u_0\|_{L^2}^2\right)\int_0^t\|f(\tau)\|_{H^1}^2\,d\tau<\infty \tag*{(18)} \end{aligned}

for t∈[0,t∗)t \in[0,t_*). By this a-priori estimate combining local existence result (see [14 Theorem 1.2] for example), we can immediately prove the main theorem.

Reformulation of the Kolmogorov-Richardson energy cascade

In this section, with the aid of the DNS result in § 2, we describe the Kolmogorov-Richardson energy cascade in terms of vortex stretching. In what follows (including the next section), we employ R3\mathbb{R}^3 which is as an approximation domain of T3\mathbb{T}^3. Also we simply write ∫f:=∫R3f(x) dx\int f := \int_{\mathbb{R}^3} f(x)\,dx. Let uu be a smooth solution to the 3D-Navier-Stokes equation (2.1) (but replace T3\mathbb{T}^3 to R3\mathbb{R}^3), and ω\omega be the corresponding vorticity. Let

ωˉK(t,x):=Fξ−1[χAK(ξ)ω^(t,ξ)](x),(19)\bar{\omega}_K(t,x):=\mathcal{F}^{-1}_{\xi}\left[\chi_{A_K}(\xi)\widehat{\omega}(t,\xi)\right](x), \tag*{(19)}

where (coarse-graining in an annulus, Littlewood-Paley decomposition)

AK:={ξ∈R3:K/α≤∣ξ∣≤αK}A_K:=\left\{\xi\in\mathbb{R}^3:K/\sqrt{\alpha}\leq|\xi|\leq\sqrt{\alpha}K\right\}

and α>1\alpha>1 is a prescribed constant expressing the ratio of adjacent scales (see Figure 1). In what follows we take wavenumber KK from αZ\alpha^{\mathbb{Z}}. Also, in the following arguments, we assume that the spatially integrated value of any quantity is independent of time because the size of the domain is much larger than the correlation length (the integral length) of the flow, that is,

ddt∥ωˉK(t)∥L22=0for any K and t≥0.(20)\frac{d}{dt}\|\bar{\omega}_K(t)\|_{L^2}^2=0\quad\text{for any }K\text{ and }t\geq0. \tag*{(20)}

Our specific purpose in this section is to approximate the following vortex stretching term:

∫(ω⋅∇)u⋅ωˉK.\int(\omega\cdot\nabla)u\cdot\bar{\omega}_K.

Note that ω=∑K∈αZωˉK\omega=\sum_{K\in\alpha^{\mathbb{Z}}}\bar{\omega}_K and u=∑K∈αZuˉKu=\sum_{K\in\alpha^{\mathbb{Z}}}\bar{u}_K. First, we propose a new hypothesis. For the original version of Kolmogorov hypothesis, see [15], for a recasted version, see [6 Section 6].

Outline of our hypothesis. The crucial point of our hypothesis is that there exists a universal vortex stretching/compressing mechanism, independent of the energy input rate ϵ\epsilon (which will appear later). More precisely, a pair of tubular vortices Wj,KW_{j,K} are stretching several tubular vortices in the adjacent smaller scale, or, several tubular vortices are compressing a pair of tubular vortices in the adjacent larger scale. Also, supports of these tubular vortices (with compact supports) are disjoint. We conjecture that such vortex stretching/compressing in adjacent two scales are the dominant event, so, in our cascade picture, we exclude multiscale events. Note that, by Remark 6, due to the uncertainty principle for the Fourier transform, we expect that the three-wave interaction of this adjacent scale event is rather nonlocal (c.f. [19]).

Balance of vortex stretching/compressing. Vortex stretching and compressing are balanced if (20) holds. See also (29).

A pair of tubular vortices. Let Wj,K∈Cc∞(R3)W_{j,K} \in\mathcal{C}^{\infty}_{c}(\mathbb{R}^{3}) represents a pair of “normalized” tubular vortices, where jj and KK are important indexes, will appear later. First we reasonably assume

supp⁡Wj,K∩supp⁡Wj′,K=∅for j≠j′.(21)\operatorname{supp} W_{j,K} \cap\operatorname{supp} W_{j',K} = \varnothing\quad\text{for } j \ne j'. \tag*{(21)}

Next we define W~j,K(x)\widetilde{W}_{j,K}(x) as follows:

W~j,K(x):=F−1[χF[Wj,K]](x),\widetilde{W}_{j,K}(x) := \mathcal{F}^{-1}\left[\chi\mathcal{F}[W_{j,K}]\right](x),

where χ\chi is a characteristic function such that

χ(ξ)={11α≤∣ξ∣≤α,0otherwise.\chi(\xi) = \begin{cases} 1 & \dfrac{1}{\sqrt{\alpha}} \le|\xi| \le\sqrt{\alpha},\\ 0 & \text{otherwise}. \end{cases}

Note that supp⁡F[W~j,K]⊂A1\operatorname{supp}\mathcal{F}[\widetilde{W}_{j,K}] \subset A_{1} and W~j,K\widetilde{W}_{j,K} is decaying (not compactly supported) in the physical space, due to the uncertainty principle. We assume that

∫R3∣Wj,K(x)∣2dx=γ\int_{\mathbb{R}^{3}} |W_{j,K}(x)|^{2}dx = \gamma

for some prescribed (non-dimensional) constant γ\gamma which is independent of all parameters.

Remark 4. In the real turbulence, this γ\gamma is rather probability distribution.

By Parseval’s identity, we see that

∫R3∣W~j,K(x)∣2dx=∫R3∣Wj,K(x)∣2dx+∫R3∣W~j,K(x)−Wj,K(x)∣2dx.\int_{\mathbb{R}^{3}} |\widetilde{W}_{j,K}(x)|^{2}dx = \int_{\mathbb{R}^{3}} |W_{j,K}(x)|^{2}dx + \int_{\mathbb{R}^{3}} |\widetilde{W}_{j,K}(x)-W_{j,K}(x)|^{2}dx.

Also assume these W~j,K\widetilde{W}_{j,K} and Wj,KW_{j,K} are close to each other, that is,

∫R3∣Wj,K(x)−W~j,K(x)∣2dx∫R3∣Wj,K(x)∣2dx≤δ<1(22)\frac{\int_{\mathbb{R}^{3}} |W_{j,K}(x)-\widetilde{W}_{j,K}(x)|^{2}dx}{\int_{\mathbb{R}^{3}} |W_{j,K}(x)|^{2}dx} \le\delta< 1 \tag*{(22)}

for some small δ>0\delta> 0. In Appendix, we give a typical example of tubular vortices Wj,KW_{j,K}, and we estimate δ\delta by using it.

Decomposition of vorticity field. In the inertial range, we employ the following hypothesis. For any fixed tt, assume that ω‾K\overline{\omega}_{K} is mainly expressed by {Wj,K}j\{W_{j,K}\}_{j}, and is independent of ν\nu and δ\delta, as follows:

ω‾K(t,x)=gK−H+1(∑j=1cKDWj,K(Kx)),(23)\overline{\omega}_{K}(t,x) = gK^{-H+1}\left(\sum_{j=1}^{cK^{D}} W_{j,K}(Kx)\right), \tag*{(23)}

where c,g,H,D∈R≥0c,g,H,D \in\mathbb{R}_{\ge0}. Rigorously, there must exist a small perturbation (in order to recover divergence-free and compact support in the Fourier space) depending on ν\nu and δ\delta, but we neglect it. HH and DD express the Hölder exponent and the fractal dimension respectively. Rigorously, we need to require cKDcK^{D} to be integer. In this case we slightly approximate cc and DD. This DD is ideally determined by physical experiments. For example, in [24], we have investigated this exponent by numerical computations. In particular, we identified the tubular vortices in each scales by the low-pressure method, and we figured out that the hierarchy of stronger tubular vortices is indeed intermittent with the dimension DD smaller than 3. The above assumptions are supported by [24] in some extent.

Remark 5. (The specific aim of this paper.) The vorticity version of the Littlewood-Paley spectra ELP(K)E_{LP}(K) is given by

ELP(K):=K−3∥ω‾K∥L22=cγg2K−53K−13(3−D)K23(−3H−2+D)E_{LP}(K) := K^{-3}\lVert\overline{\omega}_K\rVert_{L^2}^{2} = c\gamma g^2 K^{-\frac{5}{3}} K^{-\frac{1}{3}(3-D)} K^{\frac{2}{3}(-3H-2+D)}

(c.f. Sec. 2 in [3] for the definition of the Littlewood-Paley spectra). Thus, our specific aim is to figure out the constants gg and HH. Note that this ELPE_{LP} dimensionally coincides with another EE in (2.10).

The external force. For fixed c>0c > 0, we assume that, for any energy input rate ϵ\epsilon and wavenumber KfK_f, we can construct an external force ff such that supp⁡f^⊂AKf\operatorname{supp}\widehat{f} \subset A_{K_f} and

1Kf2∫(∇×f)Kf⋅ω=ϵ,\frac{1}{K_f^2}\int(\nabla\times f)_{K_f}\cdot\omega=\epsilon,

where ω\omega is a solution to the Navier-Stokes (vorticity) equation.

Hierarchy of tubular vortices. Let K±=α∓1KK_{\pm}=\alpha^{\mp1}K and let us assume that a pair of tubular vortices WJ,K+W_{J,K_+} stretches (through Biot-Savart law) Wj,KW_{j,K} (j∈ΩJ,K+⊂{1,2,⋯ ,cKD}j\in\Omega^+_{J,K}\subset\{1,2,\cdots,cK^D\}) and assume disjointness ΩJ,K+∩ΩJ′,K+=∅\Omega^+_{J,K}\cap\Omega^+_{J',K}=\varnothing (J≠J′J\ne J'). At the same time, this Wj,KW_{j,K} compresses WJ,K+W_{J,K_+}. If WJ,K+W_{J,K_+} does not stretch anything, then we regard ΩJ,K+\Omega^+_{J,K} as an empty set. In the forcing scale, the external force ∇×f\nabla\times f stretches Wj,KfW_{j,K_f} (j∈ΩKf+⊂{1,2,⋯ ,c(Kf)D}j\in\Omega^+_{K_f}\subset\{1,2,\cdots,c(K_f)^D\}). Let

ΩK+=⋃J=1cKDΩJ,K+.\Omega^+_K=\bigcup_{J=1}^{cK^D}\Omega^+_{J,K}.

To the contrary, assume that, for any j∈(ΩK+)c:={1,2,⋯ ,cKD}∖ΩK+j\in(\Omega^+_K)^c:=\{1,2,\cdots,cK^D\}\setminus\Omega^+_K, Wj,KW_{j,K} is compressed by Wj′,K−W_{j',K_-} (j′∈Ωj,K−⊂Nj'\in\Omega^-_{j,K}\subset\mathbb{N}) through Biot-Savart law. At the same time, Wj′,K−W_{j',K_-} is stretched by this Wj,KW_{j,K}. Note that ΩJ,K+=ΩJ,K+−\Omega^+_{J,K}=\Omega^-_{J,K_+} and ∣ΩK+∣+∣(ΩK+)c∣=KD|\Omega^+_K|+|(\Omega^+_K)^c|=K^D. The construction of this scenario is inspired by Movie S9 in “Supplementary materials” in [16]. In this movie, we can observe forming an ordered array of counter-rotating secondary vortex filaments perpendicular to the primary cores. We can also observe secondary filaments interacting to form a new generation of perpendicular tertiary vortex filaments.

Clarification of α\alpha and ∣Ωj,K+∣|\Omega^+_{j,K}|. We crucially assume α\alpha and ∣Ωj,K+∣(=∣Ωj,K−∣)|\Omega^+_{j,K}|(=|\Omega^-_{j,K}|) are non-dimensional prescribed constants (ideally determined by physical experiments, see also Figure 1), that is, independent of KK and jj. Rigorously this ∣Ωj,K+∣|\Omega^+_{j,K}| is integer, but here we are implicitly spatially averaging it, to be a positive real number. We conjecture that this ∣Ωj,K+∣|\Omega^+_{j,K}| is determined by an unstable mode (which is independent of any type of disturbance) induced by the elliptical instability (see Subsection 2.2 in [13], see also [16]). More precisely, the elliptical instability originates from the parametric excitation of Kelvin modes in the vortex cores in each tubular vortices Wj,KW_{j,K}.

Ratios of the number of tubular vortices: being stretched and being compressed. Let τ+(K)\tau_{+}(K) and τ−(K)\tau_{-}(K) be ratios of the number of tubular vortices, being stretched and being compressed, respectively. Since in the statistically steady state, these ratios could be estimated by the ratio of turnover time in adjacent scales, it follows that

turnover time∼lengthvelocity∼g−1KH−1.\text{turnover time} \sim\frac{\text{length}}{\text{velocity}} \sim g^{-1}K^{H-1}.

Thus we can define τ±\tau_{\pm} (satisfying τ++τ−=1\tau_{+}+\tau_{-}=1) as follows:

τ+(K):=KH−1KH−1+KH−1=11+αH−1.\tau_{+}(K) := \frac{K^{H-1}}{K^{H-1}+K^{H-1}} = \frac{1}{1+\alpha^{H-1}}.
τ−(K):=KH−1KH−1+KH−1=1α−H+1+1.\tau_{-}(K) := \frac{K^{H-1}}{K^{H-1}+K^{H-1}} = \frac{1}{\alpha^{-H+1}+1}.

Consequently τ±\tau_{\pm} are independent of KK.

Vortex stretching/compressing relation vs spatial and statistical independence. For tubular vortices Wj1,K1W_{j_{1},K_{1}} and Wj2,K2W_{j_{2},K_{2}}, let us define vortex stretching/compressing relations as follows:

  • Vortex stretching relation:

V+:={(j1,K1,j2,K2):j1∈Ωj2,K1+ and K2=α−1K1}.\mathcal{V}^{+}:=\{(j_{1},K_{1},j_{2},K_{2}):j_{1}\in\Omega^{+}_{j_{2},K_{1}}\text{ and }K_{2}=\alpha^{-1}K_{1}\}.
  • Vortex compressing relation:

V−:={(j1,K1,j2,K2):j2∈Ωj1,K1− and K2=αK1}.\mathcal{V}^{-}:=\{(j_{1},K_{1},j_{2},K_{2}):j_{2}\in\Omega^{-}_{j_{1},K_{1}}\text{ and }K_{2}=\alpha K_{1}\}.

Definition 1. (Vortex stretching/compressing relation: disjointness) We assume the following: If Wj1,K1W_{j_{1},K_{1}} and Wj2,K2W_{j_{2},K_{2}} have vortex stretching/compressing relation, that is, if (j1,K1,j2,K2)∈V+∪V−(j_{1},K_{1},j_{2},K_{2})\in\mathcal{V}^{+}\cup\mathcal{V}^{-}, then

supp⁡Wj1,K1∩supp⁡Wj2,K2=∅.\operatorname{supp}W_{j_{1},K_{1}}\cap\operatorname{supp}W_{j_{2},K_{2}}=\varnothing.

Definition 2. (Spatial and statistical independence.) Let Uj,K:=−∇Δ−1×Wj,KU_{j,K}:=-\nabla\Delta^{-1}\times W_{j,K}. Then we significantly assume the following:

∑(j1,K1,j2,K2)∈(V+)c∩(V−)c∫(Wj1,K1⋅∇)Uj2,K2⋅Wj∗,K∗=0\sum_{(j_{1},K_{1},j_{2},K_{2})\in(\mathcal{V}^{+})^{c}\cap(\mathcal{V}^{-})^{c}}\int(W_{j_{1},K_{1}}\cdot\nabla)U_{j_{2},K_{2}}\cdot W_{j^{*},K^{*}}=0

and

∑(j1,K1,j2,K2)∈V+∪V−(j1,K1)≠(j∗,K∗) and (j2,K2)≠(j∗,K∗)∫(Wj1,K1⋅∇)Uj2,K2⋅Wj∗,K∗=0.\sum_{\substack{(j_{1},K_{1},j_{2},K_{2})\in\mathcal{V}^{+}\cup\mathcal{V}^{-}\\ (j_{1},K_{1})\ne(j^{*},K^{*})\text{ and }(j_{2},K_{2})\ne(j^{*},K^{*})}}\int(W_{j_{1},K_{1}}\cdot\nabla)U_{j_{2},K_{2}}\cdot W_{j^{*},K^{*}}=0.

for any fixed j∗j^{*} and K∗K^{*}.

By Definitions 1 and 2, we have that

∫(ω⋅∇)u⋅ωˉK=∫(ωˉK⋅∇)uˉK+⋅ωˉK+∫(ωˉK⋅∇)uˉK−⋅ωˉK.\int(\omega\cdot\nabla)u\cdot\bar{\omega}_{K} = \int(\bar{\omega}_{K}\cdot\nabla)\bar{u}_{K+}\cdot\bar{\omega}_{K} + \int(\bar{\omega}_{K}\cdot\nabla)\bar{u}_{K-}\cdot\bar{\omega}_{K}.

Normalization of vortex stretching/compressing. Let ωK∗(x):=ω‾K(x)/(gK−H+1)\omega_K^*(x):=\overline{\omega}_K(x)/(gK^{-H+1}) and let ωK∗∗(x):=ωK∗(K−1x)\omega_K^{**}(x):=\omega_K^*(K^{-1}x). Then we see that

∫(ω‾K⋅∇)u‾K+⋅ω‾K+∫(ω‾K⋅∇)u‾K−⋅ω‾K=g3K−3(H−1)(α+−H+1∫(ωK∗⋅∇)uK+∗⋅ωK∗+α−−H+1∫(ωK∗⋅∇)⋅uK−∗⋅ωK∗)=g3K−3(H−1)K−3(α+−H+1∫(ωK∗∗⋅∇)uK+∗∗⋅ωK∗∗+α−−H+1∫(ωK∗∗⋅∇)⋅uK−∗∗⋅ωK∗∗).\begin{aligned} \int(\overline{\omega}_K\cdot\nabla)\overline{u}_{K_+}\cdot\overline{\omega}_K+\int(\overline{\omega}_K\cdot\nabla)\overline{u}_{K_-}\cdot\overline{\omega}_K \\ &=g^3K^{-3(H-1)}\left(\alpha_+^{-H+1}\int(\omega_K^*\cdot\nabla)u_{K_+}^*\cdot\omega_K^*+\alpha_-^{-H+1}\int(\omega_K^*\cdot\nabla)\cdot u_{K_-}^*\cdot\omega_K^*\right) \\ &=g^3K^{-3(H-1)}K^{-3}\left(\alpha_+^{-H+1}\int(\omega_K^{**}\cdot\nabla)u_{K_+}^{**}\cdot\omega_K^{**}+\alpha_-^{-H+1}\int(\omega_K^{**}\cdot\nabla)\cdot u_{K_-}^{**}\cdot\omega_K^{**}\right). \end{aligned}

Note that these ωK∗∗\omega_K^{**}, uK+∗∗u_{K_+}^{**} and uK−∗∗u_{K_-}^{**} are precisely expressed as

ωK∗∗(t,x)=∑j=1cKDWj,K(x)\omega_K^{**}(t,x)=\sum_{j=1}^{cK^D}W_{j,K}(x)
uK+∗∗(t,x)=∑j=1cK+DUj,K+(α−1x)=:∑Uj,K+,u_{K_+}^{**}(t,x)=\sum_{j=1}^{cK^D_+}U_{j,K_+}(\alpha^{-1}x)=:\sum U_{j,K_+},
uK−∗∗(t,x)=∑j=1cK−DUj,K−(αx)=:∑Uj,K−,u_{K_-}^{**}(t,x)=\sum_{j=1}^{cK^D_-}U_{j,K_-}(\alpha x)=:\sum U_{j,K_-},

for each fixed tt. By (4.3), then we see that

∫(ωK∗∗⋅∇)uK+∗∗⋅ωK∗∗=∑J∑j∈ΩJ,K+∫(Wj,K⋅∇)UJ,K+⋅Wj,K,\int(\omega_K^{**}\cdot\nabla)u_{K_+}^{**}\cdot\omega_K^{**}=\sum_J\sum_{j\in\Omega^+_{J,K}}\int(W_{j,K}\cdot\nabla)U_{J,K_+}\cdot W_{j,K},
∫(ωK∗∗⋅∇)uK−∗∗⋅ωK∗∗=∑J∑j∈(ΩK+)c∫(Wj,K⋅∇)Uj′,K−⋅Wj,K,\int(\omega_K^{**}\cdot\nabla)u_{K_-}^{**}\cdot\omega_K^{**}=\sum_J\sum_{j\in(\Omega^+_{K})^c}\int(W_{j,K}\cdot\nabla)U_{j',K_-}\cdot W_{j,K},

where

∇UJ,K:=−∇(∇Δ−1×WJ,K).\nabla U_{J,K}:=-\nabla(\nabla\Delta^{-1}\times W_{J,K}).

Let S±S_\pm and P±P_\pm be the strain and rotation tensors of ∇uK±∗∗\nabla u_{K_\pm}^{**} respectively. Then we can rewrite the following vortex stretching/compressing terms:

∫(ωK∗∗⋅∇)uK±∗∗⋅ωK∗∗=∫(ωK∗∗)T(P±+S±)⋅ωK∗∗=∫(ωK∗∗)TS±⋅ωK∗∗,\int(\omega_K^{**}\cdot\nabla)u_{K_\pm}^{**}\cdot\omega_K^{**}=\int(\omega_K^{**})^T(P_\pm+S_\pm)\cdot\omega_K^{**}=\int(\omega_K^{**})^TS_\pm\cdot\omega_K^{**},

where (ωK∗∗)T(\omega_K^{**})^T is the row vector of ωK∗∗\omega_K^{**}. By the direct calculation, the rotation tensor part disappears. Let SJ,K±S_{J,K_\pm} be the rate-of-strain tensor of ∇UJ,K±\nabla U_{J,K_\pm}.

Clarification of a universality of vortex stretching/compressing. We now clarify the universal vortex stretching/compressing hypothesis. In the vortex stretching case, we assume

Λ+:=∫(Wj,K)TSJ,K+⋅Wj,K>0for j∈ΩJ,K+,(24)\Lambda_+:=\int(W_{j,K})^TS_{J,K_+}\cdot W_{j,K}>0\quad\text{for }j\in\Omega^+_{J,K}, \tag*{(24)}

where Λ+\Lambda_+ is the absolute constant, in particular, independent of ϵ\epsilon. On the other hand, in the vortex compressing case, we assume

−Λ−:=∫(Wj,K)T∑j′∈Ωj,K−Sj′,K−⋅Wj,K<0for j∈(ΩK+)c,(25)-\Lambda_-:=\int(W_{j,K})^T\sum_{j'\in\Omega^-_{j,K}}S_{j',K_-}\cdot W_{j,K}<0\quad\text{for }j\in(\Omega_K^+)^c, \tag*{(25)}

where Λ−>0\Lambda_{-}>0 is the absolute constant, in particular, independent of ϵ\epsilon.

Remark 6. Due to the uncertainty principle, vortex stretching in the adjacent scale

F[∑j∈ΩJ,K+(Wj,K)TSJ,K+⋅Wj,K]\mathcal{F}\left[\sum_{j\in\Omega^{+}_{J,K}}(W_{j,K})^{T}S_{J,K+}\cdot W_{j,K}\right]

has non-compact support in the Fourier space (by the same reason, the vortex compressing also has non-compact support). More precisely we expect that the following three wave interaction is nonzero:

F[∑j∈ΩJ,K+(F−1[F[Wj,K]χAK1])TF−1[F[SJ,K+]χAK2]⋅F−1[F[Wj,K]χAK3]]\mathcal{F}\left[\sum_{j\in\Omega^{+}_{J,K}}\left(\mathcal{F}^{-1}\left[\mathcal{F}[W_{j,K}]\chi_{A_{K_{1}}}\right]\right)^{T}\mathcal{F}^{-1}\left[\mathcal{F}[S_{J,K+}]\chi_{A_{K_{2}}}\right]\cdot\mathcal{F}^{-1}\left[\mathcal{F}[W_{j,K}]\chi_{A_{K_{3}}}\right]\right]

for the following nonlocal three wave combinations (c.f. (17)):

{K1≈K3,K2≪K1,K2≈K3,K1≪K2,K1≈K2,K3≪K1.\begin{cases} K_{1}\approx K_{3},\quad K_{2}\ll K_{1},\\ K_{2}\approx K_{3},\quad K_{1}\ll K_{2},\\ K_{1}\approx K_{2},\quad K_{3}\ll K_{1}. \end{cases}

Summarizing the above arguments, we have

∫(ωK∗∗)TSK±⋅ωK∗∗=±cKDΛ±τ±α∓(−H+1).\int(\omega_{K}^{**})^{T}S_{K_{\pm}}\cdot\omega_{K}^{**}=\pm cK^{D}\Lambda_{\pm}\tau_{\pm}\alpha^{\mp(-H+1)}.

Remark 7. We can apply the same calculation to the convection term, and clearly it becomes zero:

∫(u⋅∇)ω⋅ωˉK=∫(u⋅∇)ωˉK⋅ωˉK=0\int(u\cdot\nabla)\omega\cdot\bar{\omega}_{K}=\int(u\cdot\nabla)\bar{\omega}_{K}\cdot\bar{\omega}_{K}=0

due to skew-symmetry.

We now summarize the important definitions of “coherent vortices” as the following:

Coherent vortices: Let us define the approximate parameters:

C+:=cΛ+τ+α−(−H+1),C_{+}:=c\Lambda_{+}\tau_{+}\alpha^{-(-H+1)},
C−:=cΛ−τ−α−H+1.C_{-}:=c\Lambda_{-}\tau_{-}\alpha^{-H+1}.

Note that the constants cc, Λ±\Lambda_{\pm} and α\alpha are independent of ϵ\epsilon.

  • Inertial range (c.f. enstrophy production rate (7))

1K2∫(ω⋅∇)u⋅ωˉK=g3(C+−C−)K−3H−2+D\frac{1}{K^{2}}\int(\omega\cdot\nabla)u\cdot\bar{\omega}_{K}=g^{3}(C_{+}-C_{-})K^{-3H-2+D}

for KF<K≤KdisK_{F}<K\leq K_{\mathrm{dis}}. Also, from the physical observation 1.1, we mathematically assume Kdis→∞K_{\mathrm{dis}}\to\infty as ν→0\nu\to0.

  • Driving force scale (beginning of the coherent vorticity)

In the forcing scale, we assume that the effect of stretching is negligible:

∫(ωˉKf⋅∇)uˉ(Kf/α)ωˉKf=0(26)\int(\bar{\omega}_{K_{f}}\cdot\nabla)\bar{u}(K_{f}/\alpha)\bar{\omega}_{K_{f}}=0 \tag*{(26)}

In other words, initially, driving force ff creates the largest scale vortices. Thus

1Kf2∫(ω⋅∇)u⋅ωˉKf=−g3C−Kf−3H−2+D.\frac{1}{K_f^2}\int(\omega\cdot\nabla)u\cdot\bar{\omega}_{K_f}=-g^3C_-K_f^{-3H-2+D}.

Derivation of Kolmogorov's −5/3-5/3 law

In this section, we derive Kolmogorov’s −5/3-5/3 law. The main theorem in this section is the following.

Theorem 5.1. Assume (4.2) and “coherent vortices”. Taking ν→0\nu\to0, then we have Kolmogorov’s −5/3-5/3 law:

ELP(K):=K−3∥ωˉK∥L22=cγ2/3K−5/3K−13(3−D)(27)E_{\mathrm{LP}}(K):=K^{-3}\|\bar{\omega}_K\|_{L^2}^2=c\gamma^{2/3}K^{-5/3}K^{-\frac{1}{3}(3-D)} \tag*{(27)}

for K≥KfK\ge K_f.

Proof. We multiply by ωˉK\bar{\omega}_K to (2.1), integrate on both sides, divided by K2K^2, to obtain

12K2ddt∫R3∣ωˉK∣2+νK2∫R3∣∇ωˉK∣2+1K2∫R3(u⋅∇)ω⋅ωˉK−1K2∫R3(ω⋅∇)u⋅ωˉK=1K2∫R3(∇×f)⋅ωˉK(28)\begin{aligned} &\frac{1}{2K^2}\frac{d}{dt}\int_{\mathbb{R}^3}|\bar{\omega}_K|^2+\frac{\nu}{K^2}\int_{\mathbb{R}^3}|\nabla\bar{\omega}_K|^2\\ &\quad+\frac{1}{K^2}\int_{\mathbb{R}^3}(u\cdot\nabla)\omega\cdot\bar{\omega}_K-\frac{1}{K^2}\int_{\mathbb{R}^3}(\omega\cdot\nabla)u\cdot\bar{\omega}_K\\ &\quad=\frac{1}{K^2}\int_{\mathbb{R}^3}(\nabla\times f)\cdot\bar{\omega}_K \tag*{(28)} \end{aligned}

for K≥KfK\ge K_f. By the assumption: coherent vortices and (4.2), we have

νK2∫R3∣∇ωˉK(⋅,x)∣2dx−g3(C+−C−)K−3H−2+D=0\frac{\nu}{K^2}\int_{\mathbb{R}^3}|\nabla\bar{\omega}_K(\cdot,x)|^2dx-g^3(C_+-C_-)K^{-3H-2+D}=0

for Kf<K<KdisK_f<K<K_{dis} and

νKf2∫R3∣∇ωˉKf(⋅,x)∣2dx+g3C−Kf−3H−2+D=ϵ.\frac{\nu}{K_f^2}\int_{\mathbb{R}^3}|\nabla\bar{\omega}_{K_f}(\cdot,x)|^2dx+g^3C_-K_f^{-3H-2+D}=\epsilon.

On the other hand,

νK2∥∇ωˉK(⋅)∥L22≈ν∥ωˉK(⋅)∥L22=νcγg2K−2H−1+D.\frac{\nu}{K^2}\|\nabla\bar{\omega}_K(\cdot)\|_{L^2}^2\approx\nu\|\bar{\omega}_K(\cdot)\|_{L^2}^2=\nu c\gamma g^2K^{-2H-1+D}.

Thus

∣C+−C−∣≲νcγg−1KH+1|C_+-C_-|\lesssim\nu c\gamma g^{-1}K^{H+1}

for Kf<K<KdisK_f<K<K_{dis} and

∣ϵ−C−g3Kf−3H−2+D∣<νcγg2Kf−2H−1+D.|\epsilon-C_-g^3K_f^{-3H-2+D}|<\nu c\gamma g^2K_f^{-2H-1+D}.

Since C±C_\pm, MM and HH are independent of ν\nu and KK, we have

C+=C−andϵ=C−g3Kf−3H−2+D(29)C_+=C_-\quad\text{and}\quad\epsilon=C_-g^3K_f^{-3H-2+D} \tag*{(29)}

for taking ν\nu to zero. Then the rest argument is rather obvious. First, assume H≠(−2+D)/3H\ne(-2+D)/3, then it contradicts (5.3) if we take KfK_f sufficiently large. Thus, we have H=(−2+D)/3H=(-2+D)/3. Consequently, HH (and also τ±\tau_\pm) is independent of ϵ\epsilon. Second, assume g≠ϵ1/3g\ne\epsilon^{1/3}, then it contradicts (5.3) if we take ϵ\epsilon sufficiently large. Thus, g=ϵ1/3g=\epsilon^{1/3}. Consequently, C−=C+=1C_{-}=C_{+}=1 and then Λ±\Lambda_{\pm} are also determined. Therefore the Littlewood–Paley spectra ELP(K)E_{LP}(K) is given by

ELP(K):=K−3∥ωˉK∥L22=cγϵ2/3K−5/3K−13(3−D).E_{LP}(K):=K^{-3}\lVert\bar{\omega}_{K}\rVert_{L^{2}}^{2}=c\gamma\epsilon^{2/3}K^{-5/3}K^{-\frac{1}{3}(3-D)}.

This is Kolmogorov’s −5/3-5/3 power law.

Conclusion

We investigated the hierarchy of coherent vortices observed in the DNS of forced turbulence in a periodic domain. First, we proved that if the normalized enstrophy production rate Psup⁡P_{\sup} is concave in the far dissipation range (more precisely, β<1\beta<1), the corresponding solution to the Navier-Stokes equation is smooth enough. Since the DNS results shown in Figure 4 seem to support the new regularity criterion that Psup⁡P_{\sup} is concave in the far dissipation range, we may conclude that the turbulence driven by the deterministic force in a periodic domain is smooth enough. Second, we construct the hierarchy of coherent vortices which is statistically self-similar in the inertial range so that we can estimate the energy transfer rate in adjacent scales due to the vortex stretching/compressing process. The crucial assumption, which is clearly supported by DNS results (Figure 2), of this construction is that the process is self-similar and local in scale. Then, under this assumption, we may derive Kolmogorov’s −5/3-5/3 power law of the energy spectrum of statistically stationary turbulence. We emphasize that our derivation does not require Kolmogorov’s hypotheses, but instead it requires a universal mechanism of the vortex stretching and compressing. Though the present study lacks the discussion on the origin of this universality, the recent study by McKeown [16] may provide with it; they investigated how the elliptical instability led to the persistence of the turbulent energy cascade through the local interactions of vortices over a hierarchy of scales.

Appendix: a typical example of antiparallel tubular vortices

In this section we give a typical example of a pair of antiparallel tubular vortices, and estimate δ\delta in (22). Using Fourier transform of a special function (see [21, 26, 28, 29]) with translation, we can construct a typical example of tubular vorticity W~n,ζ\widetilde{W}^{n,\zeta} within a band-pass filter:

W~n,ζ(x):=(0,0,∏k=1n(sin⁡(2−kx1)2−kx1sin⁡(2−kx2)2−kx2sin⁡(2−kx3)2−kx3)sin⁡(ζx1))\widetilde{W}^{n,\zeta}(x):=\left(0,0,\prod_{k=1}^{n}\left(\frac{\sin(2^{-k}x_{1})}{2^{-k}x_{1}}\frac{\sin(2^{-k}x_{2})}{2^{-k}x_{2}}\frac{\sin(2^{-k}x_{3})}{2^{-k}x_{3}}\right)\sin(\zeta x_{1})\right)

for some ζ>m:=∑k=1n2−k\zeta>m:=\sum_{k=1}^{n}2^{-k} (1/2≤m<11/2\le m<1). We will fix these parameters n∈Z≥1n\in\mathbb{Z}_{\ge1} and ζ∈R>0\zeta\in\mathbb{R}_{>0} later. Then we see that F[W~n,ζ]\mathcal{F}[\widetilde{W}^{n,\zeta}] is smooth (F[W~n,ζ]∈Cn−2(R3)\mathcal{F}[\widetilde{W}^{n,\zeta}]\in C^{n-2}(\mathbb{R}^{3})) and compactly supported. More precisely,

supp⁡F[Wn,ζ]⊂{ξ∈R3:ζ−mζ+mK<min⁡{∣ξ1∣,∣ξ2∣,∣ξ3∣} and max⁡{∣ξ1∣,∣ξ2∣,∣ξ3∣}≤ζ+mζ−mK}.\operatorname{supp}\mathcal{F}[W^{n,\zeta}]\subset \left\{\xi\in\mathbb{R}^{3}:\sqrt{\frac{\zeta-m}{\zeta+m}}K<\min\{|\xi_{1}|,|\xi_{2}|,|\xi_{3}|\}\ \text{and}\ \max\{|\xi_{1}|,|\xi_{2}|,|\xi_{3}|\}\le\sqrt{\frac{\zeta+m}{\zeta-m}}K\right\}.

for K=ζ+mζ−m(ζ−m)K=\sqrt{\frac{\zeta+m}{\zeta-m}}(\zeta-m) (c.f. (8) and (19)). Then we can define a typical example of antiparallel tubular vortices Wn,ζW^{n,\zeta} as follows:

Wn,ζ(x):=W~n,ζ(x)χ[−πζ,πζ)×[−2π,2π)2(x1,x2,x3).W^{n,\zeta}(x):=\widetilde{W}^{n,\zeta}(x)\chi_{[-\frac{\pi}{\zeta},\frac{\pi}{\zeta})\times[-2\pi,2\pi)^{2}}(x_{1},x_{2},x_{3}).

In this calculable setting, we cut off in the real space (not in the Fourier space). So it is rather natural to estimate

∫∣Wn,ζ−W~n,ζ∣2∫∣W~n,ζ∣2\frac{\int|W^{n,\zeta}-\widetilde{W}^{n,\zeta}|^2}{\int|\widetilde{W}^{n,\zeta}|^2}

instead of

∫∣Wn,ζ−W~n,ζ∣2∫∣Wn,ζ∣2.\frac{\int|W^{n,\zeta}-\widetilde{W}^{n,\zeta}|^2}{\int|W^{n,\zeta}|^2}.

In this paper, for the sake of simplicity, we have used the characteristic function type of band-pass filter. But, rigorously, using the overlapping smoothed Littlewood–Paley decomposition may be more effective to capture antiparallel tubular vortices. Thus it may be better to choose ζ\zeta and mm satisfying (ζ+m)/(ζ−m)>1.7\sqrt{(\zeta+m)/(\zeta-m)}>\sqrt{1.7}. In this appendix, we examine the case n=2n=2 and ζ=5/4\zeta=5/4 (in this case m=3/4m=3/4 and (ζ+m)/(ζ−m)=2\sqrt{(\zeta+m)/(\zeta-m)}=2). With the aid of the numerical computation, we can estimate as the following:

∫∣Wn,ζ−W~n,ζ∣2∫∣W~n,ζ∣2≃1−0.80×(0.994)2≃0.21.\frac{\int|W^{n,\zeta}-\widetilde{W}^{n,\zeta}|^2}{\int|\widetilde{W}^{n,\zeta}|^2}\simeq1-0.80\times(0.994)^2\simeq0.21.

From this estimate, we can naturally imagine that there exist various antiparallel tubular vortices for small δ\delta.

Acknowledgments. Research of TY was partly supported by the JSPS Grants-in-Aid for Scientific Research 17H02860, 18H01136, 18H01135 and 20H01819. SG was partly supported by the JSPS Grants-in-Aid for Scientific Research 20H02068 and 20K20973. This work was done while TY was an associate professor at the University of Tokyo, Japan.

Graduate School of Economics, Hitotsubashi University, 2-1 Naka, Kunitachi, Tokyo 186-8601, Japan

Email address: [email protected]

Graduate School of Engineering Science, Osaka University, 1-3 Machikaneyama, Toyonaka, Osaka 560-8531, Japan

Email address: [email protected]

Graduate School of Mathematical Sciences, University of Tokyo, Komaba 3-8-1 Meguro, Tokyo 153-8914, Japan

Email address: [email protected]

References

  1. [1]A. Babin, A. Mahalov and B. Nicolaenko, Global regularity of the 3D Rotating Navier-Stokes Equations for resonant domains. Indiana Univ. Math. J., 48, (1999), 1133-1176.
  2. [2]J.-M. Bony, Calcul symbolique et propagation des singularités pour les équations aux dérivées partielles non linéaires. Ann. Sci. École Norm. Sup., 14, (1981), 209-246.DOI
  3. [3]P. Constantin, The Littlewood-Paley Spectrum in Two-Dimensional Turbulence, Theo. Comput. Fluid Dynam., 9, (1997), 183-189.DOI
  4. [4]P. E. Dimotakis, The mixing transition in turbulent flows, J. Fluid Mech., 409, (2000), 69-98.DOI
  5. [5]U. Frisch, From global scaling, a la kolmogorov, to local multifractal scaling in fully developed turbulence, Proc. Roy. Soc. London A, 434, (1991), 89-99.DOI
  6. [6]U. Frisch, Turbulence, Cambridge University Press, Cambridge, 1995.
  7. [7]S. Goto, A physical mechanism of the energy cascade in homogeneous isotropic turbulence, J. Fluid Mech., 605, (2008), 355–366.DOI
  8. [9]S. Goto, Y. Saito, and G. Kawahara, Hierarchy of antiparallel vortex tubes in spatially periodic turbulence at high reynolds numbers, Phys. Rev. Fluids, 2, (2017), 064603.DOI
  9. [10]S. Goto and J. C. Vassilicos, Local equilibrium hypothesis and taylor's dissipation law, Fluid Dyn. Res., 48, (2016), 021402.DOI
  10. [11]T. Ishihara, T. Gotoh, and Y. Kaneda, Study of high-reynolds number isotropic turbulence by direct numerical simulation, Ann. Rev. Fluid Mech., 41, (2009), 165-180.DOI
  11. [12]I.-J. Jeong and T. Yoneda, Vortex stretching and enhanced dissipation for the incompressible 3D Navier–Stokes equations, Math. Annal., 380, (2021), 2041–2072.DOI
  12. [13]R. R. Kerswell, Elliptical instability, Annu. Rev. Fluid Mech., 34, (2002), 83–113.
  13. [14]N. Kishimoto and T. Yoneda, Global solvability of the rotating Navier-Stokes equations with fractional Laplacian in a periodic domain, Math. Ann., 372, (2018), 743–779.arxiv.org/abs/1702.07443
  14. [15]A. Kolmogorov, Local structure of turbulence in an incompressible fluid at very high reynolds number, Sov. Phys. Usp., 10, (1968), 734.
  15. [16]R. McKeown, R. Ostilla-Mónico, A. Pumir, M. P. Brenner and S. M. Rubinstein, Turbulence generation through an iterative cascade of the elliptical instability, Sci. Adv., 6, (2020), eaaz2717.DOI
  16. [17]Y. Motoori and S. Goto, Generation mechanism of a hierarchy of vortices in a turbulent boundary layer, J. Fluid Mech., 865, (2019), 1085-1109.DOI
  17. [18]Y. Motoori and S. Goto, Hierarchy of coherent structures and real-space energy transfer in turbulent channel flow, J. Fluid Mech., 911, (2021), A27.DOI
  18. [19]K. Ohkitani and S. Kida, Triad interactions in a forced turbulence, Phys. Fluids A, 4, (1992), 794-802.
  19. [20]S. A. Orszag and G. S. Patterson, Numerical simulation of three-dimensional homogeneous isotropic turbulence, Phys. Rev. Lett., 28, (1972), 76-79.DOI
  20. [21]V. L. Rvačov and V. O. Rvačov, A certain finite function, Dopovīdi Akad. Nauk Ukraïn. RSR Ser. A, 764, (1971), 705-707.
  21. [22]G. I. Taylor, Statistical theory of turbulence, Proc. Roy. Soc. A, 151, (1935), 421-444.
  22. [23]H. Tennekes and J. L. Lumley, A First Course in Turbulence, M.I.T. Press, Cambridge, Mass., 1972.DOI
  23. [24]T. Tsuruhashi, S. Goto, S. Oka and T. Yoneda, Self-similar hierarchy of coherent tubular vortices in turbulence, to appear in Philosophical Trans. A.DOI
  24. [25]J. C. Vassilicos, Dissipation in turbulent flows, Ann. Rev. Fluid Mech., 47, (2015), 95-114.DOI
  25. [26]W. Volk, Evaluation of the solution of an integral-functional equation, J. Comput. Appl. Math., 174, (2005), 423-436.DOI
  26. [27]M. Yamada and T. Yoneda, Resonant interaction of Rossby waves in two-dimensional flow on a beta plane, Physica D, 245, (2013), 1-7.
  27. [28]T. Yoneda, On the functional-differential equation of advanced type f'(x) = af(2x) with f(0) = 0, J. Math. Anal. Appl., 317, (2006), 320-330.DOI
  28. [29]T. Yoneda, On the functional-differential equation of advanced type f'(x) = af(λx), λ > 1, with f(0) = 0, J. Math. Anal. Appl., 332, (2007), 487-496.DOI
  29. [30]T. Yoneda, Long-time solvability of the Navier-Stokes equations in a rotating frame with spatially almost periodic large data, Arch. Ration. Mech. Anal., 200, (2011), 225-237.DOI

Paper details

vlnc:W_spmsu2b4
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents