betaFeedback

WorksPhysics and Astronomy

Is it true that no mathematical relation exists between the Navier-Stokes equations and the multifractal model?

Abstract

Contrary to accepted turbulence folklore, which holds that no mathematical relation exists between the Navier-Stokes equations (NSEs) and the multifractal model (MFM) of Parisi and Frisch, we develop a theory that reconciles the MFM with Leray's weak solutions of Navier-Stokes analysis. From a combination of Euler invariant scaling and the NSEs set in a three-dimensional box of size LL, we also derive the Paladin-Vulpiani inverse scale ηh,pavη_{h,pav}, which is related to the Reynolds number Re\mathit{Re} by Lηh,pav−1=Re1/(1+h)Lη_{h,pav}^{-1} = \mathit{Re}^{1/(1+h)}, and which acts as a mediator between the two theories. This is achieved by considering L2mL^{2m}-norms of the velocity gradient to find a correspondence between mm and the local scaling exponent hh in the multifractal model. The parameter mm acts as if it were the sliding focus control on a telescope which allows us to zoom in and out on different structures. The range 1⩽m⩽∞1 \leqslant m \leqslant \infty is equivalent to −2/3⩽hmin⩽1/3-2/3 \leqslant h_{min} \leqslant 1/3, which lies precisely in the region where Bandak et al. (2022, 2024) have suggested that thermal noise makes the NSEs inadequate and generates spontaneous stochasticity. The implications of this are discussed.

Introduction

Historical perspective

The language of fractal physics has been particularly useful in understanding high Reynolds number flow-visualizations that show the simultaneous existence and dynamic evolution of vortical structures ranging from fully three-dimensional large vortices, down to quasi-one-dimensional filaments, and even broken structures whose dimensions are less than unity.

Many historical references can be found in [77, 71, 46]; see also [74, 44, 60, 76, 22, 10, 11, 25, 26, 6, 54, 56, 9, 47] (Sreenisvasan 1991). In their recent report on an extensive series of 32,768332,768^3 computations, Yeung et al [78] have pointed out that

Three-dimensional turbulence governed by the Navier–Stokes equations continues to be a grand challenge in computational science, even as leadership-class computing power has recently advanced past exascale.

Theoretical approaches involving fractal geometry and dynamical systems began to intersect in the 1970s and 1980s when conventional descriptions of fluid turbulence as merely quasiperiodic motion (Landau & Lifshitz [48]) began to be questioned : see the history described by Ruelle [69]. In their influential paper, Ruelle & Takens [64] suggested that as the Reynolds number increased, fluid motion would pass through a generic sequence of bifurcations, which included both periodic and quasi-periodic motion, but whose end-state was chaotic motion on a strange attractor. This end-state was interpreted as ‘turbulence’, although it is now recognized that this specific sequence is only applicable to tightly confined low-dimensional dynamical systems : see the review by Eckmann [24] and the paper by Libchaber & Maurer (1982). The applicability of these ideas to the Navier-Stokes equations (NSEs) lay open to question because the theory expounded by Ruelle [65] was valid for finite dimensional dissipative dynamical systems. The subsequent proof by Foias & Temam [35] that the 2D2D Navier–Stokes equations possess a finite dimensional global attractor A\mathcal{A} made the connection secure in this case. Indeed, the further work by Ruelle [68]; Constantin & Foias [16] and Constantin [15] on Lyapunov exponents of A\mathcal{A} laid the basis for the sharp (to within logarithms) estimate of the Lyapunov dimension of the 2D2D Navier-Stokes global attractor (Constantin, Foias & Temam [17]). It must be stressed however, that this process cannot be followed through for the 3D3D NSEs. The Millenium problem of the global regularity of solutions in three dimensions remains open and therefore the existence of a finite dimensional global attractor remains open. What is known rigorously is the existence of Leray-Hopf weak solutions (Leray [49]; Hopf [43]), and it is the correspondence between these and the multifractal model which is the topic of this paper.

The wide range of scales observed in turbulent flows has traditionally been interpreted in terms of a Richardson cascade which involves the transfer of energy from large to smaller scales of motion: see Monin & Yaglom [55] and also the review by Jiménez (2000) who have discussed the effect of intermittency on cascades. Mandelbrot [51] popularized the idea that this wide range of scales indicates an underlying fractal geometric structure, but it also became increasingly clear that the idea of a monofractal system with just a single dimension was not enough to provide an adequate description. Parisi & Frisch [61] then interpreted these cascade processes in multifractal terms where there is a continuous spectrum of dimensions. A parameter designated as hh, which appears naturally in the invariant scaling of the underlying Euler equations (see (3) below), was used as a local scaling exponent, with each value of hh belonging to a given fractal set of dimension D(h)D(h). This has become known as the multi-fractal model (MFM) of turbulence: expositions can be found in Meneveau & Sreenivasan [52]; Frisch [37]; Bohr et al [8]; Boffetta et al [7]; Benzi & Biferale [5] and Frisch [38], together with two series of lecture notes respectively by Eyink [29] and Benzi & Toschi [6]. In fact, multifractal scaling has a broad application in the physical sciences (Stanley & Meakin [72]). While hydrodynamic turbulence is perhaps its most prominent example, other turbulent systems must be included, such as passive scalar fields (Prasad, Meneveau & Sreenivasan [62]) and bacterial suspensions (Liu & Lin [53]).

A reconciliation between the MFM and the NSEs

It has become part of the folklore of turbulence that no mathematical relation exists between the MFM, which is applicable to statistically steady, homogeneous, isotropic turbulence (HIT), and the forced incompressible 3D3D NSEs

(∂t+u⋅∇)u+∇p=νΔu+f(x)div⁡u=0,(1)(\partial_t + \boldsymbol{u}\cdot\nabla)\boldsymbol{u} + \nabla p = \nu\Delta\boldsymbol{u} + \boldsymbol{f}(\boldsymbol{x}) \qquad\operatorname{div}\boldsymbol{u}=0, \tag*{(1)}

which are evolution equations requiring specific initial conditions. In (1) u\boldsymbol{u}, pp and ν\nu are respectively the velocity field, pressure and viscosity of the fluid, while f(x)\boldsymbol{f}(\boldsymbol{x}) is the body forcing. Building on piecemeal attempts in this direction (Dubrulle & Gibbon [23]; Gibbon [42]), this paper will demonstrate that a reconciliation between the two theories is possible. There are two ways of approaching this reconciliation, both of which involve a result of Paladin & Vulpiani [58] who used the MFM formalism to introduce a length scale into Navier-Stokes turbulence which we call the PaV-scale ηh,pav\eta_{h,pav} (to avoid confusion with potential vorticity). In terms of the Reynolds number ReRe the inverse PaV-scale is

ℓηh,pav−1=Re11+h,(2)\ell_{\eta_{h,pav}}^{-1}=Re^{\frac{1}{1+h}}, \tag*{(2)}

where the definition of the Reynolds number in Paladin & Vulpiani [58] was originally based on the energy dissipation rate and was derived using phenomenological arguments rather than directly from the NSEs. Moreover, the velocity field was assumed to be statistically stationary. To establish a mathematical relation between the MFM and the NSEs, it is critical that we use a definition of the PaV-scale that is not only based on the NSEs but which is also expressed in terms of a NS-based Reynolds number. Let us consider the invariant scaling of the Euler equations. Let η\eta be an inner length scale 0<η≤ℓ0 < \eta\le\ell and ℓ\ell an outer scale which, in the periodic case, would be the box size ℓ=L\ell= L. Consider the incompressible Euler equations in dimensional variables u(x,t)\mathbf{u}(\mathbf{x},t). If we introduce a typical velocity U0U_0, a frequency ϖ0=U0ℓ−1\varpi_0 = U_0\ell^{-1} and the dimensionless parameter λ=ℓη−1≥1\lambda= \ell\eta^{-1} \ge1, then the dimensionless set of primed variables u′(x′,t′)\mathbf{u}'(\mathbf{x}',t') defined by (h+1>0h+1>0)

x′=λℓ−1x;t′=λ1−hϖ0t;u=U0λ−hu′(3)\mathbf{x}' = \lambda\ell^{-1}\mathbf{x}; \qquad t' = \lambda^{1-h}\varpi_0t; \qquad\mathbf{u} = U_0\lambda^{-h}\mathbf{u}' \tag*{(3)}

transforms the Euler equations in dimensional variables into the dimensionless Euler equations in primed variables. While the Euler equations are invariant under the transformation in (3) for every value of λ≥1\lambda\ge1, the NSEs are not. Instead, direct application of (3) to the NSEs (1.1) shows that η=ηh,pav\eta= \eta_{h,pav} is that particular value of η\eta for which the NSEs in dimensional variables u(x,t)\mathbf{u}(\mathbf{x},t) and with corresponding Reynolds number Re=ℓU0ν−1Re = \ell U_0\nu^{-1}, are transformed into the NSEs in dimensionless variables u′(x′,t′)\mathbf{u}'(\mathbf{x}',t') with Reynolds number equal to unity

(∂t+u′⋅∇′)u′+∇′p′=Δ′u′+f′(x′)div⁡′u′=0.(4)(\partial_t+\mathbf{u}'\cdot\nabla')\mathbf{u}'+\nabla'p'=\Delta'\mathbf{u}'+\mathbf{f}'(\mathbf{x}') \qquad\operatorname{div}'\mathbf{u}'=0 . \tag*{(4)}

Thus η=ηh,pav\eta= \eta_{h,pav} occurs exactly where the inertial and dissipative terms balance. This interpretation of ηh,pav\eta_{h,pav} has been used, in particular, to obtain ‘fusion rules’ for the statistical correlations between the inertial-range increments and the derivatives of the velocity ([4] (Friedrich et al. 2018)). Depending on the allowed range of hh we see that the invariant scaling transformation of the Euler equations lies behind the onset of NS-turbulence as a multi-scale phenomenon. Kolmogorov’s theory ([37]), known as K41, restricts this to a single scale by insisting on h=13h = \frac{1}{3} but, as we see below, the MFM allows a wider spread of hh. Moreover, we shall also see that the PaV-scale is much more than just a bridge between the Euler and the NSEs because it also stitches together the MFM and Leray’s weak solution formalism of the NSEs ([49, 63]) which is manifest as sequences of bounded time averages ([40, 41]).

Our first approach, expounded in § 4.1, is to adapt the method of Dubrulle & Gibbon [23] for evaluating the higher norms of the velocity gradient by using the Euler scaling invariance in the probabilistic sense of the MFM at the wave-number kh,pav=ηh,pav−1k_{h,pav} = \eta_{h,pav}^{-1}. Selecting only this wavenumber in the spectrum with a variation in hh, and then averaging over hh, circumvents the difficulty with the Kolmogorov picture where the spectrum is divided into the inertial range and the dissipation range, a division which conventional Sobolev methods of NS-analysis do not recognize. Comparison with time-averaged results from the NSEs in § 4.2 shows that the results are consistent with the four-fifths law; namely that the multifractal spectrum (co-dimension) obeys the inequality C(h)≥1−3hC(h) \ge1 - 3h, with hh lying in the range −23≤h≤13-\frac{2}{3} \le h \le\frac{1}{3}. This corresponds to the dissipation range. The typical velocity field U0U_0 in (3) is chosen as the root-mean-square velocity field of the NSEs, as in (9), and the outer scale ℓ\ell is chosen as the box-scale LL of the NSEs. The outer scale ℓ\ell could equivalently be taken as the integral scale of the velocity or the characteristic length scale of the forcing. Such choices would yield an additional prefactor (L/ℓ)3(L/\ell)^3 in the Navier–Stokes estimates provided in § 3, but would not change the way the estimates scale with ReRe, which is the key element in our analysis.

The second idea also involves the PaV-scale but in the reverse order. In § 4.2 the parameter mm in the L2mL^{2m}-norms of the velocity gradient is treated as if it were the sliding focus control of a telescope through which one can zoom in and out on intermittent events. This allows us to mimic weak solution time averages as if they exist on a multifractal set FmF_m whose range of dimensions is Dm=3/m\mathcal{D}_m = 3/m. Under the assumption that the NS-flow has settled into a fully developed state, Dm\mathcal{D}_m can then be compared with D(h)D(h). This gives a simple relation between the scaling exponent hh and mm. The range 1≤m≤∞1 \le m \le\infty corresponds to −23≤hmin⁡≤13-\frac{2}{3} \le h_{\min} \le\frac{1}{3}, which gives a cut-off deep in the dissipation range. The PaV-scale ηh,pav\eta_{h,pav} then re-emerges as the best estimate over all mm for the natural length scale of the problem.

Connection with two open questions

The invariance of the Euler equations and their correspondence with the NSEs also touches recent work on Onsager’s conjecture [28, 30, 30] (Onsager 1949). De Lellis & Székelyhidi (2009, 2010) have shown that the incompressible 3D Euler equations possess ‘wild solutions’ whose kinetic energy e(t)e(t) does not have to be decreasing in time and may have compact support, thus suggesting a strong degree of irregularity. These solutions, found by the method of convex integration, imply that the fluid could suddenly burst into rapid motion, and then return to rest after a finite period of time. An equivalent result has been proved for the NSEs by Buckmaster & Vicol (2019, 2021). These solutions belong to the space W1,1W^{1,1} for all time, which means that both ∥u∥1\lVert\boldsymbol{u}\rVert_1 and ∥∇u∥1\lVert\nabla\boldsymbol{u}\rVert_1 remain finite. However, ∥∇u∥2\lVert\nabla\boldsymbol{u}\rVert_2 is not integrable in time so they do not satisfy a Leray-Hopf-type energy inequality. The suggestion that wild solutions may be the root cause of NS-turbulence is both speculative and controversial but it is nevertheless an interesting and important open question. The magazine article by Eyink et al (2025) covers a wide range of opinion by several authors over the issue of whether the Buckmaster-Vicol result is physically relevant to hydrodynamics. What is clear is that this issue has yet to be settled. A further argument is that the flaw lies in the Navier–Stokes equations themselves. The suggestion that thermal noise could make the deterministic version of the incompressible NSEs inadequate to describe the dissipation range of turbulence was first raised by Ruelle (1979, 1995). This is precisely the region covered by the PaV-scale. More recently, Bandak et al (2022, 2024) have gone further by arguing that in molecular fluids Eulerian spontaneous stochasticity [39, 73] is triggered by thermal noise in 3D Navier-Stokes turbulence at high Reynolds numbers and that this directly alters the turbulent dissipation range below the Kolmogorov scale. Bandak et al (2024) conclude that the NSEs should be replaced by a Landau–Lifshitz-type set of equations which is comprised of the incompressible NSEs plus an additive fluctuating stress term proportional to div⁡ξ\operatorname{div}\boldsymbol{\xi}, where ξ\boldsymbol{\xi} is modelled as a mean-zero Gaussian random field with a given ensemble average. The range of scales over which it is conjectured that these thermal effects take place not only includes the dissipation range but also reaches into the inertial range, which illustrates why it is important that this issue is resolved. Moreover, if this argument turns out to be correct then it will change the direction of analysis in fluid dynamics. The search for a regularity proof of the 3D deterministic NSEs, although unsuccessful, has been one of the major issues in rigorous fluid dynamics in the past generation [49, 36, 63, 40]. In contrast, the Landau–Lifshitz equations, having been designed as mesoscopic field equations with a high-wavenumber cutoff, have been proved to have strong, pathwise-unique solutions [33, 30].

The multifractal model of turbulence

The MFM is well-established in the literature, so we provide only the briefest of summaries [37, 8, 7, 29, 5, 6]. The ppth-order longitudinal velocity structure function SpS_p at a point xx in an HIT flow is defined as

Sp(r)=⟨{[u(x+r)−u(x)]⋅r^]p⟩stat.av.,(5)S_p(r) = \left\langle\left\{ \left[ \boldsymbol{u}(x+r) - \boldsymbol{u}(x) \right] \cdot\hat{\boldsymbol{r}} \right]^p \right\rangle_{\mathrm{stat.av.}}, \tag*{(5)}

where rr is the radius of the sphere centred at x\boldsymbol{x}. While the details of the statistical average are of little concern, the scaling properties of SpS_p in the infinite Reynolds number limit are crucial as they revolve around the invariance property of the incompressible Euler equations. The K41 formalism assumes that SpS_p depends only on rr and the energy dissipation rate ε=νL−3∫V∣∇u∣2dV\varepsilon= \nu L^{-3}\int_V \lvert\nabla\boldsymbol{u}\rvert^2dV. Since the product εr\varepsilon r has the same units as u3u^3, the only scaling consistent with physical units is Sp∼(εr)p/3S_p \sim(\varepsilon r)^{p/3}. According to Frisch (1995), the symbol ∼\sim means ‘scales like’. This agrees with the exact result S3=−45εrS_3=-\frac{4}{5}\varepsilon r, which is known as the four-fifths law. While the result is elegant and satisfying, experimental and numerical data indicate that SpS_p scales like Sp∼rζpS_p\sim r^{\zeta_p}, where the exponents ζp\zeta_p deviate from linear in pp by lying on a concave curve that, for p≥3p\geq3, stays below the line 13p\frac{1}{3}p, a deviation which is attributed to intermittency [37]. To explain this, Parisi & Frisch [61] adapted the K41 formalism by relaxing the requirement that, for small rr, the velocity increment ∣u(x+r)−u(x)∣\lvert\boldsymbol{u}(\boldsymbol{x}+\boldsymbol{r})-\boldsymbol{u}(\boldsymbol{x})\rvert scales as rhr^h with h=13h=\frac{1}{3}. Instead, they allowed the scaling exponent hh to fluctuate. They achieved this by writing down the probability of observing a given value of hh at the scale rr in the form

Pr(h)∼rC(h),(6)P_r(h)\sim r^{C(h)}, \tag*{(6)}

where each value of hh belongs to a given fractal set of dimension D(h)D(h). The co-dimension C(h)=3−D(h)C(h)=3-D(h) plays the role of the multifractal spectrum in the large deviation theory version expounded by Eyink [29]. Interpreting (5) as an average over the probability Pr(h)P_r(h), and by applying a steepest descent argument, one finds the following relation for the exponent ζp\zeta_p

ζp=lim⁡r→0(ln⁡Spln⁡(εr)),whereζp=inf⁡h[hp+C(h)].(7)\zeta_p=\lim_{r\to0}\left(\frac{\ln S_p}{\ln(\varepsilon r)}\right), \qquad\text{where}\qquad\zeta_p=\inf_h [hp+C(h)]. \tag*{(7)}

Thus, ζp\zeta_p and C(h)C(h) are connected through a Legendre transform. Note that when D(h)=3D(h)=3, we recover the standard K41 result that Sp∼(εr)hpS_p\sim(\varepsilon r)^{hp}. The four-fifths law requires ζ3=1\zeta_3=1, thus leading to inequalities for D(h)D(h) and C(h)C(h)

C(h)≥1−3h⇒D(h)≤3h+2.(8)C(h)\geq1-3h \quad\Rightarrow\quad D(h)\leq3h+2. \tag*{(8)}

An alternative definition of multifractality is based on the fluctuations of energy dissipation (Meneveau & Sreenivasan [52]; Aurell et al. [1]; Frisch [37]). It assumes the existence of a continuous range of scaling exponents aa such that the local average of the energy dissipation rate over a ball of radius rr scales like rar^a over a fractal set of dimension F(a)F(a). Kolmogorov’s refined similarity hypothesis allows a connection to be established between the two definitions of multifractality; in particular, h=a/3h=a/3 and D(h)=F(a)D(h)=F(a) (Aurell et al. [1]; Frisch [37]). Finally, it is important to note that the MFM describes the scaling properties of the velocity field and the local energy dissipation but does not provide any information on the geometrical structure of the flow. To see how the multifractal formalism fits with the NSEs requires an examination of the L2mL^{2m}-norms of the velocity gradient, which we do in the next section.

The L2mL^{2m}-norms of the Navier-Stokes velocity gradient

The behaviour of the pointwise in time energy dissipation rate ε=νL−3∫V∣∇u∣2dV\varepsilon=\nu L^{-3}\int_V\lvert\nabla\boldsymbol{u}\rvert^2dV has been one of the primary goals in turbulence research (Verma [75]). Based on the ideas of Leray [49] and Hopf [43], the energy inequality in a 3D3D periodic domain is

12∫V(∣u(x,t)∣2−∣u(x,0)∣2)dV+ν∫0t∫V∣∇u(x,τ)∣2dV dτ≤∫0t∫Vu⋅f dV dτ.\frac{1}{2}\int_V\left(\lvert\boldsymbol{u}(\boldsymbol{x},t)\rvert^2-\lvert\boldsymbol{u}(\boldsymbol{x},0)\rvert^2\right)dV+\nu\int_0^t\int_V\lvert\nabla\boldsymbol{u}(\boldsymbol{x},\tau)\rvert^2dV\,d\tau\leq\int_0^t\int_V\boldsymbol{u}\cdot\boldsymbol{f}\,dV\,d\tau.

Note that this is an inequality, not an equality, and is valid for weak solutions of the 3D3D NSEs: see also Caffarelli, Kohn & Nirenberg [14] for a wider definition. Behind it lies a sophisticated theory of weak convergence properties that can be found in standard textbooks such as Robinson et al. [63]. Doering & Foias [20] showed how to translate estimates based on the forcing f\mathbf{f} (whose L2L^2-norm feeds into the Grashof number GrGr) into estimates based on the Reynolds number Re=U0L/νRe = U_0L/\nu which is itself based on a space-time-averaged velocity field U0U_0, and the time-average ⟨⋅⟩T\langle\cdot\rangle_T

U02=L−3⟨∥u∥22⟩T,with⟨⋅⟩T=1T∫0T⋅ dt.(9)U_0^2 = L^{-3}\left\langle\|\mathbf{u}\|_2^2\right\rangle_T,\qquad\text{with}\qquad\langle\cdot\rangle_T = \frac{1}{T}\int_0^T\cdot\,\mathrm{d}t. \tag*{(9)}

The Reynolds number Re=U0L/νRe = U_0L/\nu has a solid basis because U0U_0 is bounded for all time. The time averaged energy dissipation rate for all smooth Navier–Stokes initial conditions and square integrable forcing is (Doering & Foias [20])

⟨ε⟩T=νL−3⟨∫V∣∇u∣2 dV⟩T≤L−4ν3Re3.(10)\langle\varepsilon\rangle_T = \nu L^{-3}\left\langle\int_V|\nabla\mathbf{u}|^2\,\mathrm{d}V\right\rangle_T \leq L^{-4}\nu^3Re^3. \tag*{(10)}

Inequality (10) leads to the standard inverse Kolmogorov length estimate Lλk−1≤Re3/4L\lambda_k^{-1}\leq Re^{3/4}, consistent with K41. However, it does not take account of strong intermittent events that are washed over by the spatial L2L^2-average in (10). To address this issue we require higher L2mL^{2m}-norms of the velocity gradient tensor.

∥∇u∥2m≡(∫V∣∇u∣2m dV)1/2m,1≤m≤∞.(11)\|\nabla\mathbf{u}\|_{2m}\equiv\left(\int_V|\nabla\mathbf{u}|^{2m}\,\mathrm{d}V\right)^{1/2m},\qquad1\leq m\leq\infty. \tag*{(11)}

For values of mm near unity the integral is insensitive to strong intermittent events: the near L2L^2 spatial average washes over the most intense regions of ∇u\nabla\mathbf{u} with relatively little effect – see Figure 1. However, as the value of mm is raised, regions of greater intensity begin to dominate, and eventually at m=∞m=\infty the most intense point of all completely dominates the integral.

Evolution for trefoil vortex knot

Figure 1. From the evolution of a trefoil vortex knot for ν=3.12×10−5\nu= 3.12 \times10^{-5}, evidence appears of intermittent events at t≈110t \approx110 for m≥5m \ge5; courtesy of R. M. Kerr [47]. The vertical axis is FmαmF_m^{\alpha_m} defined in (13) for d=3d = 3 which represent scaled L2mL^{2m}-norms of the velocity gradient.

m=1m=1; corresponds to an r.m.s. average

mm finite but large; corresponds to the more intense structures

m=∞m=\infty; corresponds to the most intense point

Thus, the sliding scale of m=1→∞m=1\to\infty acts as a zoom lens of a telescope. To find time-averages of ∥∇u∥2m\|\nabla\mathbf{u}\|_{2m} for m≥1m\geq1 first requires a result of Foias, Guillopé & Temam [36], which deserves to be better known, and which is based around nn-th order weak derivatives within the volume integrals Hn=∫V∣∇nu∣2 dVH_n=\int_V|\nabla^n\mathbf{u}|^2\,\mathrm{d}V. The Foias, Guillopé & Temam [36] result is that Leray’s weak solutions satisfy

⟨Hn12n−1⟩T≤cL−1ν22n−1Re3+O(T−1),(12)\left\langle H_n^{\frac{1}{2n-1}}\right\rangle_T\leq cL^{-1}\nu^{\frac{2}{2n-1}}Re^3+O\left(T^{-1}\right), \tag*{(12)}

in which the energy inequality is simply the case n=1n=1. Using the Doering & Foias [20] method, the estimate has been translated into the Reynolds number ReRe. Interpolation inequalities, together with (12), can then easily be used to find time-averages of certain powers of ∥∇nu∥2m\|\nabla^n\mathbf{u}\|_{2m}. Generalizing to other spatial dimensions d=1,2,3d=1,2,3, these results can be rolled into a single formula (Gibbon [40, 41, 42]). While n≥3n\geq3 spatial derivatives in (12) are necessary for the proof, once this has been achieved, all we need are estimates of time-averages of powers of the single derivative ∥∇u∥2m\|\nabla\mathbf{u}\|_{2m} written as two simple dimensionless formulas

Fm,d=ν−1L1/αm,d∥∇u∥2m,αm,d=2m4m−d,(13)F_{m,d}=\nu^{-1}L^{1/\alpha_{m,d}}\|\nabla\mathbf{u}\|_{2m},\qquad\alpha_{m,d}=\frac{2m}{4m-d}, \tag*{(13)}

where all the results rolled together can be summarized in the formula

⟨Fm,d(4−d)αm,d⟩T≤cmRe3.(14)\left\langle F_{m,d}^{(4-d)\alpha_{m,d}}\right\rangle_T\leq c_mRe^3. \tag*{(14)}

This, then, is the dimensionless generalization of (10) to values m≥1m \ge1 and dd dimensions. Note that when m=1m = 1, the factor of 4−d4-d cancels and we recover the result on the energy dissipation rate in (11). This result is the key not only to the correspondence between the MFM and the NSEs but also the PaV-scale. Donzis et al. [21] studied the evolution of ∥ω∥2m\lVert\boldsymbol{\omega}\rVert_{2m} numerically for 4 different data-sets. for values m=1,…,9m = 1,\ldots,9. Figure 1 is a plot of Fm,3αmF_{m,3}^{\alpha_m} from Kerr [47] (courtesy of R. M. Kerr) similar to its equivalent in Donzis et al. [21]. The subscript d=3d=3 has been suppressed. The bound in (14) is a rigorous Navier–Stokes result. In the next section, we will show how it can be used in connection with the phenomenological predictions of the MFM to reconcile the two approaches.

Two ways of reconciling the MFM and the NSEs

The PaV-scale as a mediator between the MFM and the NSEs

Having seen the properties of the Euler invariant scaling transformation in § 1, it is now time to apply this to the L2mL^{2m}-norms of the velocity gradient defined in (11). To find a way round the problem of how to match results from the MFM with those from the NSEs, we adapt the method outlined in Dubrulle & Gibbon [23] where only the wavenumber kpavk_{pav} out of all possible wavenumbers was chosen, with a subsequent integration over hh. Instead, by using the scaling transformation (3) on the energy dissipation rate, we keep all wavenumbers but select the PaV-scale kh,pav=ηh,pav−1k_{h,pav}=\eta_{h,pav}^{-1} as the only one with hh-dependence, where the probability of observing a given exponent hh at the scale η\eta is Pη(h)P_\eta(h). We adapt (6) to make it non-dimensional so that

Pη(h)∼[L−1η]C(h)(15)P_\eta(h) \sim\left[L^{-1}\eta\right]^{C(h)} \tag*{(15)}

where each value of hh belongs to a given fractal set of dimension D(h)=3−C(h)D(h)=3-C(h). Our strategy is to calculate the MFM equivalent of (14) and compare the Re-scaling in the two formulations. This requires the application of the rescaling in (3) and an average over hh to obtain

∫V∣∇u∣2m dV∫VdV=U02m[∫hmin⁡hmax⁡Pη(h)η−2m(Lη−1)−2mh dh]∫V′∣∇u′∣2m dV′∫V′dV′(16)\frac{\int_V \lvert\nabla\boldsymbol{u}\rvert^{2m}\,dV}{\int_V dV} = U_0^{2m} \left[ \int_{h_{\min}}^{h_{\max}} P_\eta(h)\eta^{-2m}(L\eta^{-1})^{-2mh}\,dh \right] \frac{\int_{V'}\lvert\nabla\boldsymbol{u}'\rvert^{2m}\,dV'}{\int_{V'}dV'} \tag*{(16)}

in which case the Fm,3F_{m,3} in (13) for d=3d=3 becomes

Fm,3=ν−1L1/αm,3∥∇u∥2m∼ν−1U0L[∫hmin⁡hmax⁡(L−1η)C(h)(Lη−1)2m(1−h) dh]1/2m(∫V′∣∇′u′∣2m dV′∫V′dV′)1/2m.(17)F_{m,3}=\nu^{-1}L^{1/\alpha_{m,3}}\lVert\nabla\mathbf{u}\rVert_{2m} \sim\nu^{-1}U_0L\left[\int_{h_{\min}}^{h_{\max}}\left(L^{-1}\eta\right)^{C(h)}\left(L\eta^{-1}\right)^{2m(1-h)}\,dh\right]^{1/2m}\left(\frac{\int_{V'}\lvert\nabla'\mathbf{u}'\rvert^{2m}\,dV'}{\int_{V'}dV'}\right)^{1/2m}. \tag*{(17)}

Now we use the PaV-scale Lηh,pav−1=Re1/(1+h)L\eta_{h,\mathrm{pav}}^{-1}=Re^{1/(1+h)} to obtain

Fm,3∼Re[∫hmin⁡hmax⁡Re2m(1−h)−C(h)1+h dh]1/2m(∫V′∣∇′u′∣2m dV′∫V′dV′)1/2m.(18)F_{m,3}\sim Re\left[\int_{h_{\min}}^{h_{\max}}Re^{\frac{2m(1-h)-C(h)}{1+h}}\,dh\right]^{1/2m}\left(\frac{\int_{V'}\lvert\nabla'\mathbf{u}'\rvert^{2m}\,dV'}{\int_{V'}dV'}\right)^{1/2m}. \tag*{(18)}

In the primed variables in (18), the equivalent Reynolds number is unity when kpav=ηh,pav−1k_{\mathrm{pav}}=\eta_{h,\mathrm{pav}}^{-1}, so this term is not a function of ReRe. Therefore, in the limit Re→∞Re\to\infty, by approximating the integral over hh by a steepest descent argument and matching the exponents of ReRe on the LHS with Re3Re^3 on the right from (14), we must have

αm,3[1+max⁡h(2m(1−h)−C(h)2m(1+h))]≤3,(19)\alpha_{m,3}\left[1+\max_h\left(\frac{2m(1-h)-C(h)}{2m(1+h)}\right)\right]\le3, \tag*{(19)}

in which case

min⁡hC(h)≥4m[1−3(1+h)]+9(1+h).(20)\min_h C(h)\ge4m\left[1-3(1+h)\right]+9(1+h). \tag*{(20)}

The first term on the RHS of (20) must be negative to avoid C→∞C\to\infty as m→∞m\to\infty, thus restricting hh to the range h≥−23h\ge-\frac{2}{3}. Under this restriction on hh we need to maximize the lower bound on C(h)C(h) with respect to mm, which occurs when m=1m=1. Therefore, uniform in mm, we find that

C(h)≥1−3handh≥−23.(21)C(h)\ge1-3h\qquad\text{and}\qquad h\ge-\frac{2}{3}. \tag*{(21)}

This is consistent with the results of the four-fifths law in § 2 and the range of hh in (29).

The formula in dd dimensions in (14) incorporates a factor 4−d4-d in the exponent of Fm,dF_{m,d} with αm,d=2m/(4m−d)\alpha_{m,d}=2m/(4m-d). Therefore (20) is replaced by

C(h)≥4m[1−3(1+h)4−d]+3d(1+h)4−d,(22)C(h)\ge4m\left[1-\frac{3(1+h)}{4-d}\right]+\frac{3d(1+h)}{4-d}, \tag*{(22)}

in which case we require h≥13(1−d)h\ge\frac{1}{3}(1-d) to prevent C→∞C\to\infty as m→∞m\to\infty. Again, we must have C(h)≥1−3hC(h)\ge1-3h.

A multifractal interpretation and the re-emergence of the PaV-scale

We have now reached the point where we need to find a correspondence between the NSEs and the MFM. This only makes physical sense if a Navier–Stokes flow has become fully developed at large enough TT in the same way that Eswaran & Pope [27] showed that the negative velocity derivative skewness must have reached a certain critical level.

As mentioned in § 2, the multifractal formalism based on dissipation postulates the existence of fractal sets in each of which the local energy dissipation rate is scale invariant. This fact suggests a multifractal interpretation of the results from NS-analysis. Consider αm,d\alpha_{m,d}, defined in (13) as αm,d=2m/(4m−d)\alpha_{m,d}=2m/(4m-d). If we simply divide through by mm we obtain a modified version of α\alpha, namely α1,Dm\alpha_{1,\mathcal{D}_m}

αm,d=2m4m−d=24−Dm=α1,Dm,whereDm=dm,(23)\alpha_{m,d}=\frac{2m}{4m-d}=\frac{2}{4-\mathcal{D}_m}=\alpha_{1,\mathcal{D}_m},\qquad\text{where}\qquad\mathcal{D}_m=\frac{d}{m}, \tag*{(23)}

where the domain dimension dd has been replaced by Dm\mathcal{D}_m. This means that working in the space L2m(V)L^{2m}(\mathcal{V}) in dd dimensions can be mimicked by working in the space L2(Fm)L^2(\mathcal{F}_m) in Dm\mathcal{D}_m dimensions. We can therefore think of Fm\mathcal{F}_m as a fractal set with a range of dimensions Dm\mathcal{D}_m. Moreover, we can rewrite the exponent (4−d)αm,d(4-d)\alpha_{m,d} in (14) with dd replaced by Dm\mathcal{D}_m

(4−Dm)α1,Dm=2.(24)(4-\mathcal{D}_m)\alpha_{1,\mathcal{D}_m}=2. \tag*{(24)}

This suggests that instead of writing down ε\varepsilon in L2L^2 for d=3d=3 as in (1) we define a set of dissipation rates {εm}\{\varepsilon_m\}

εm=νL−Dm⟨∫Fm∣∇u∣2 dV⟩T,Dm=3/m(25)\varepsilon_m=\nu L^{-\mathcal{D}_m}\left\langle\int_{\mathcal{F}_m}|\nabla\boldsymbol{u}|^2\,\mathrm{d}V\right\rangle_T,\qquad\mathcal{D}_m=3/m \tag*{(25)}

for each value of mm, on the fractal set Fm\mathcal{F}_m. This latter may be regarded as the set in which the energy dissipation rate averaged over a ball of radius rr scales like ramr^{a_m}, although the dependence of the scaling exponent ama_m on mm remains unknown. Moreover, to obtain the full dissipation rate requires a measure over which to sum mm, but so far this has eluded us. Noting that Fm⊂T3\mathcal{F}_m\subset\mathbb{T}^3, (25) can be re-written as

εm≲νL3−Dm⟨L−3∫T3∣∇u∣2 dV⟩T≲cL−1−Dmν3Re3.(26)\varepsilon_m\lesssim\nu L^{3-\mathcal{D}_m}\left\langle L^{-3}\int_{\mathbb{T}^3}|\nabla\boldsymbol{u}|^2\,\mathrm{d}V\right\rangle_T\lesssim cL^{-1-\mathcal{D}_m}\nu^3Re^3. \tag*{(26)}

Aurell et al [1] wrote down the formula (26) but with an arbitrary fractal dimension DD whereas, for each mm, here we have a fractal set Fm\mathcal{F}_m with a dimension Dm\mathcal{D}_m that is directly associated with the NSEs. Furthermore, εmν−3\varepsilon_m\nu^{-3} can be expressed in terms of a set of inverse lengths ηm−1\eta_m^{-1} defined such that

εmν−3:=ηm−1−Dm⟹Lηm−1≲cRe31+Dm.(27)\varepsilon_m\nu^{-3}:=\eta_m^{-1-\mathcal{D}_m}\quad\Longrightarrow\quad L\eta_m^{-1}\lesssim cRe^{\frac{3}{1+\mathcal{D}_m}}. \tag*{(27)}

Note that when D1=3\mathcal{D}_1=3 we recover the Kolmogorov estimate Re3/4Re^{3/4} with a continuum of values up to Re3Re^3 for D∞=0\mathcal{D}_\infty=0.

If Dm\mathcal{D}_m is interpreted as the dimension of the set Fm\mathcal{F}_m in which the local energy dissipation rate scales as ramr^{a_m}, it now seems natural to equate Dm\mathcal{D}_m in (25) to the fractal dimension F(am)F(a_m) and therefore to D(h)D(h). Inequality (8) leads to a simple inequality relating mm and hh

3m≤2+3h.(28)\frac{3}{m}\leq2+3h. \tag*{(28)}

We find h≥hmin⁡h\geq h_{\min} with hmin⁡=m−1−23h_{\min}=m^{-1}-\frac{2}{3}. The two limits m→1m\to1 and m→∞m\to\infty give

−23≤hmin⁡≤13.(29)-\frac{2}{3}\leq h_{\min}\leq\frac{1}{3}. \tag*{(29)}

Thus hh is bounded away from h=−1h=-1. Using the inequality relation between hh and mm we can bound the exponent 3/(1+Dm)3/(1+\mathcal{D}_m) in (27) by

11+h≤31+Dm.(30)\frac{1}{1+h}\leq\frac{3}{1+\mathcal{D}_m}. \tag*{(30)}

The PaV inverse scale ηh,pav−1\eta_{h,pav}^{-1} thus emerges as the minimizer of the RHS for all mm. The left hand side of (30) is derived directly from the MFM, while the right hand side comes from the NSEs. In (29), hmin⁡≥−23h_{\min}\geq-\frac{2}{3} bounds hh away from the dangerous value of h=−1h=-1, where a singularity can occur. The limit hmin⁡=−23h_{\min}=-\frac{2}{3} corresponds to Dm=0\mathcal{D}_m=0.

Conclusion: Is our understanding of the dissipation range correct?

The PaV length scale ηh,pav\eta_{h,pav} is that value of η\eta in the invariant transformation of the Euler equations which, when applied to the NSEs, gives unit Reynolds number, thus implying that the inertial and dissipation terms balance. Figure 2 shows the inter-relation between the Euler equations, the NSEs, the MFM and the central intermediary role of ηh,pav\eta_{h,pav}. One of the difficulties of making a comparison between the MFM, applicable to statistically steady HIT, and the dynamic NSEs is that the weak solution methods devised by Leray [49] do not readily recognize a distinction between the inertial and dissipation ranges. Using the PaV scale ηh,pav\eta_{h,pav} solves this difficulty by choosing k=kh,pavk = k_{h,pav} as the only value in the spectrum that is allowed to be hh-dependent. In effect, this process auto-selects the dissipation range only. Taking the probability of observing a given scaling exponent hh at the scale η\eta to be Pη(h)∼ηC(h)P_{\eta}(h) \sim\eta^{C(h)}, where each value of hh belongs to a given fractal set of dimension D(h)D(h), it is shown in § 4.1 that a consequence of Leray’s weak solution estimates leads to the inequality C(h)≥1−3hC(h) \ge1 - 3h, which is consistent with the four-fifths law. It is also shown in § 3 that the parameter mm in the L2mL^{2m}-norms of the velocity gradient acts as a zoom lens focus control which is related to hh.

Pictorial representation of how the PaV-scale appears as a mediator between the Euler and NSEs

Figure 2. Pictorial representation of how the PaV-scale appears as a mediator between the Euler and NSEs. Following the arrows : (i) Clockwise : application of the PaV-scale to the NSEs implies results consistent with the MFM; (ii) Anti-clockwise : the MFM and NSEs together imply the PaV-scale.

The standard picture from large-scale computational fluid dynamics is of the formation of vortical sheets, their roll-up into tubes and the breakdown of these into low-dimensional fractal segments. These sub-Kolmogorov vortical structures are usually identified with the dissipation range, in this case defined by Lkh,pav=Re1/(1+h)L k_{h,pav} = Re^{1/(1+h)} with −23≤h≤13-\frac{2}{3} \le h \le\frac{1}{3}. In fact, the −23-\frac{2}{3} lower bound on hh corresponds to a Re3Re^3 upper bound on the inverse length scale which is already close to, or deeper than, molecular scales even for modest values of ReRe. The inter-relation between the NSEs, the MFM and the central role of ηh,pav\eta_{h,pav} identified in this paper is delicately balanced around the invariance of the Euler equations, a balance which could easily be overturned by other processes. If the suggestion by Bandak et al [2, 3] is correct that the thermal noise generates spontaneous stochasticity, thereby overwhelming the dissipation range, then this points to a rethink of how high-ReRe small-scale turbulence ought to be considered. In turn it would also suggest a necessary revision of the standard computational fluid dynamics picture of low-dimensional structures associated with the dissipation range.

Declaration of Interests : The authors report no conflicts of interest.

Acknowledgements : Thanks are due to R. M. Kerr for Figure 1. The authors would also like to thank the Isaac Newton Institute for Mathematical Sciences, Cambridge, for support and hospitality during the programme “Anti-diffusive dynamics : from sub-cellular to astrophysical scales” where work on this paper was undertaken. This work was supported by EPSRC grant no EP/R014604/1.

References

  1. [1]Aurell, E., Frisch, U., Lutsko, J. & Vergassola, M. 1992 On the multifractal properties of the energy dissipation derived from turbulence data, J. Fluid Mech., 238, pp. 467–486.DOI
  2. [2]Bandak, D., Goldenfeld, N., Mailybaev, A. & Eyink, G. 2022, Dissipation-range fluid turbulence and thermal noise, Phys. Rev. E, 105, 065113.arxiv.org/abs/2107.13954
  3. [3]Bandak, D., Mailybaev, A., Eyink, G. & Goldenfeld, N. 2024, Spontaneous Stochasticity Amplifies Even Thermal Noise to the Largest Scales of Turbulence in a Few Eddy Turnover Times, Phys. Rev. Letts., 132, 104002.arxiv.org/abs/2401.13881
  4. [4]Benzi, R., Biferale, L. & Toschi, F. 1998 Multiscale velocity correlations in turbulence, Phys. Rev. Lett., 80, pp. 3244–3247.DOI
  5. [5]Benzi, R. & Biferale, L. 2009 Fully Developed Turbulence and the Multifractal Conjecture, J. Stat. Phys., 135, pp. 977–990.DOI
  6. [6]Benzi, R. & Toschi, F. 2023 Lectures on Turbulence, Physics Reports, 1021, pp. 1–106.DOI
  7. [7]Boffetta, G., Mazzino, A. & Vulpiani, A. 2008 Twenty-five years of multifractals in fully developed turbulence: a tribute to Giovanni Paladin, J. Phys. A: Math. Theor., 41, pp. 363001.DOI
  8. [8]Bohr, T., Jensen, M., Paladin, G. & Vulpiani, A. 1998 Dynamical systems approach to turbulence, Cambridge University Press, Cambridge.DOI
  9. [9]Brewer, W., Gopalakrishnan, M. M., Mauterth, M., Kashi, A., Choi, J. Y., Zhang, P., Nichols, S., Balin, R., Couchman, M., de Bruyn Kops, S., Yeung, P. K., Dotson, D., Uma-Vaideswaran, R., Oral, S. & Wang, F. 2025 Intelligent Sampling of Extreme-Scale Turbulence Datasets for Accurate and Efficient Spatiotemporal Model Training. In Proceedings of the SC ’25 Workshop of the International Conference for High Performance Computing, Networking, Storage and Analysis, pp. 1–10. Association for Computing Machinery.DOI
  10. [10]Buaria, D., Pumir, A., Bodenschatz, E. & Yeung, P. K. 2019 Extreme velocity gradients in turbulent flows, New J. Phys., 21, pp. 043004.DOI
  11. [11]Buaria, D., Bodenschatz, E. & Pumir, A. 2020 Vortex stretching and enstrophy production in high-Re turbulence, Phys. Rev. Fluids, 5, pp. 104602.DOI
  12. [14]Caffarelli, L., Kohn, R. & Nirenberg L., 1982 Partial regularity of suitable weak solutions of the Navier-Stokes equations, Comm. Pure Appl. Math., 35, pp. 771–831.DOI
  13. [15]Constantin, P. 1987 Collective L∞ estimates for families of functions with orthonormal derivatives. Indiana Univ. Math. J., 36, 603–616.
  14. [16]Constantin, P. & Foias, C. 1985 Global Lyapunov Exponents, Kaplan-Yorke Formulas and the Dimension of the Attractors for 2D Navier-Stokes Equations, Comm. Pure Appl. Math., 38, pp. 1–27.
  15. [17]Constantin, P., Foias, C. & Temam, R. 1988 On the dimension of the attractors in two-dimensional turbulence, Physica D, 30, pp. 284–296.DOI
  16. [20]Doering, C. R. & Foias, C. 2002 Energy dissipation in body-forced turbulence, J. Fluid Mech., 467, pp. 289–306.
  17. [21]Donzis, D. A., Gibbon, J. D., Gupta, A., Kerr, R. M., Pandit, R. & Vincenzi, D. 2014 Vorticity moments in four numerical simulations of the 3D Navier-Stokes equations, J. Fluid Mech., 732, 316–331.DOI
  18. [22]Dubrulle, B. 2019 Beyond Kolmogorov cascades, J. Fluid Mech., 867, pp. 1–63.DOI
  19. [23]Dubrulle, B. & Gibbon, J. D. 2022 A correspondence between the multifractal model of turbulence and the Navier-Stokes equations, Phil. Trans. R. Soc. A, 380, pp. 20210092.DOI
  20. [24]Eckmann, J. P. 1981 Roads to turbulence in dissipative dynamical systems, Rev. Mod. Phys., 53, pp. 643–654.
  21. [25]Elsinga, G. E., Ishihara, T. & Hunt, J. C. R. 2020 Extreme dissipation and intermittency in turbulence at very high Reynolds numbers, Proc. R. Soc. A, 476, pp. 20200591.DOI
  22. [26]Elsinga, G. E., Ishihara, T. & Hunt, J. C. R. 2023 Intermittency across Reynolds numbers – the influence of large-scale shear layers on the scaling of the enstrophy and dissipation in homogeneous isotropic turbulence. J. Fluid Mech., 974, pp. A17.DOI
  23. [27]Eswaran, V. & Pope, S. B. 1988 An examination of forcing in direct numerical simulations of turbulence, Comput. Fluids, 16, pp. 257–278.DOI
  24. [28]Eyink, G. 1995 Les subgrid spaces and the multifractal hypothesis, J. Stat. Phys., 78, pp. 353–375.DOI
  25. [29]Eyink, G. L. 2008 Turbulence Theory – unpublished lecture notes chapters 3C and 3D, Baltimore, MD : The Johns Hopkins University.
  26. [30]Eyink, G. L. 2024 Onsager’s ‘ideal turbulence’ theory, J. Fluid Mech., 988, P1.arxiv.org/abs/2404.10084
  27. [33]Flandoli, F. 2008 An introduction to 3D stochastic fluid dynamics. In : Stochastic Partial Differential Equations in Fluid Dynamics : Recent Progress and Prospects, Alberverio, S., Flandoli, F., Sinai, Y.G. (eds.), (CIME Summer School, Cetraro, Italy 2007). Lecture Notes in Math. 1942, pp. 51–150, Springer, Berlin ; Fondazione C.I.M.E., Florence.
  28. [35]Foias, C. & Temam, R. 1979 Some Analytic and Geometric Properties of the Solutions of the Evolution of the Navier-Stokes Equations, J. Math. Pures Appl., 58, pp. 339–368.
  29. [36]Foias, C., Guillopé, C. & Temam, R. 1981 New a priori estimates for Navier-Stokes equations in dimension 3, Comm. PDE, 6, pp. 329–359.
  30. [37]Frisch, U. 1995 Turbulence : The Legacy of A. N. Kolmogorov, Cambridge University Press (Cambridge).
  31. [38]Frisch, U. 2016 The collective birth of multifractals, J. Phys A : Math. Theor., 49, 451002.
  32. [39]Gawedzki, K. 2008 Soluble models of turbulent transport, in Non-Equilibrium Statistical Mechanics and Turbulence, eds. Nazarenko, S. and Zaboronski, O. V., London Mathematical Society Lecture Note Series, 355, Cambridge University Press (Cambridge).
  33. [40]Gibbon, J. D. 2019 Weak and strong solutions of the 3D Navier–Stokes equations and their relation to a chessboard of convergent inverse length scales, J. Nonlinear Sci., 29, pp. 215–228.
  34. [41]Gibbon, J. D. 2020 Intermittency, cascades and thin sets in three-dimensional Navier–Stokes turbulence, EPL, 131, 64001.
  35. [42]Gibbon, J. D. 2023 Identifying the multifractal set on which energy dissipates in a turbulent Navier–Stokes fluid, Physica D, 445, 133654.
  36. [43]Hopf, E. 1951 Über die Anfangswertaufgabe für die hydrodynamischen Grundgleichungen, Mathematischer Nachrichten, 4, pp. 213–231.
  37. [44]Ishihara, T., Gotoh, T. & Kaneda, Y. 2009 Study of high–Reynolds number isotropic turbulence by direct numerical simulation, Annu. Rev. Fluid Mech., 41, pp. 165–180.
  38. [46]Jiménez, J. 2025 Chaos, coherence and turbulence, Phys. Rev. Fluids, 10, 100504.DOI
  39. [47]Kerr, R. M. 2025 Navier-Stokes bounds and scaling for compact trefoils in large (2ℓπ)³ domains, J. Fluid Mech., 1025, A3. doi:10.1017/jfm.2025.10513
  40. [48]Landau, L. D. & Lifshitz, E. M. 1959 Fluid Mechanics, Vol. 6 of Course of Theoretical Physics, 1st edition, Pergamon Press, Oxford.
  41. [49]Leray, J. 1934 Sur le mouvement d’un liquide visqueux exprimant l’espace, Acta Math., 63, pp. 193–248.
  42. [51]Mandelbrot, B. 1977 Fractals : Form, Chance and Dimension, Freeman, San Francisco.
  43. [52]Meneveau, C. & Sreenivasan, K. R. 1991 The multifractal nature of turbulent energy dissipation, J. Fluid Mech., 224, pp. 429–484.
  44. [53]Liu, K.-A. & Lin, J. 2012 Multifractal dynamics of turbulent flows in swimming bacterial suspensions, Phys. Rev. E, 86, 011924.
  45. [54]McKeown, R., Pumir, A., Rubinstein, S., Brenner, M. P. & Ostilla-Mónico, R. 2023 Energy transfer and vortex structures : Visualizing the incompressible turbulent energy cascade, New J. Phys., 10, pp. 103029.
  46. [55]Monin, A. S. & Yaglom, A. M. 1975 Statistical Fluid Mechanics, Vol. I, Mechanics of Turbulence. MIT Press, Cambridge, MA.
  47. [56]Mukherjee, S., Murugan, S. D., Mukherjee, R. & Ray, S. S. 2024 Turbulent flows are not uniformly multifractal, Phys. Rev. Letts. 132, 184002.
  48. [58]Paladin, G. & Vulpiani, A. 1987 Degrees of freedom of turbulence, Phys. Rev. A, 35, pp. 1971–1973.
  49. [60]Pandit, R., Perlekar, P. & Ray, S. S. 2009, Statistical properties of turbulence : An overview, Pramana - J. Phys., 73, 157–191.
  50. [61]Parisi, G. & Frisch, U. 1985 Turbulence and Predictability in Geophysical Fluid Dynamics. Proc. Int. School of Physics “E. Fermi” 1983 Varenna, Italy, edited by M. Ghil, R. Benzi and G. Parisi, pp. 84–87 North-Holland, (Amsterdam).
  51. [62]Prasad, R. R., Meneveau, C. & Sreenivasan, K. 1988 Multifractal nature of the dissipation field of passive scalars in fully turbulent flows, Phys. Rev. Lett., 61, pp. 74–77.pubmed.ncbi.nlm.nih.gov/10038697
  52. [63]Robinson, J. C., Rodrigo, J. L. & Sadowski, W. 2016 The Navier-Stokes equations, Cambridge University Press (Cambridge).
  53. [64]Ruelle, D. & Takens, F. 1971 On the nature of turbulence, Commun. Math. Phys., 20, pp. 167–192.
  54. [65]Ruelle, D. 1978 Dynamical systems with turbulent behavior. In: Dell’Antonio, G., Doplicher, S. & Jona-Lasinio, G. (eds) Mathematical Problems in Theoretical Physics. Lecture Notes in Physics, 80. Springer, Berlin, Heidelberg. pp. 341–360.
  55. [68]Ruelle, D. 1982 Large Volume Limit of the Distribution of Characteristic Exponents in Turbulence, Commun. Math. Phys., 87, pp. 287–302.
  56. [69]Ruelle, D. 1995 Turbulence, Strange Attractors and Chaos, World Scientific, Singapore ISBN: 981-02-2310-2.
  57. [71]Sreenivasan, K. R. & Yakhot, V. 2021 Dynamics of three-dimensional turbulence from Navier-Stokes equations, Phys. Rev. Fluids, 6, 104604.arxiv.org/abs/2106.01293
  58. [72]Stanley, H. E. & Meakin, P. 1988 Multifractal phenomena in physics and chemistry, Nature, 335, pp. 405–409.DOI
  59. [73]Thalabard, S., Bec, J. & Mailybaev, A., 2020 From the butterfly effect to spontaneous stochasticity in singular shear flows, Commun. Phys., 3, pp. 122–130.DOI
  60. [74]Vassilicos, J. C. 2015 Dissipation in Turbulent Flows, Annu. Rev. Fluid Mech., 47, pp. 95–114 (2015).DOI
  61. [75]Verma, M. K. 2019 Energy transfer in fluid flows : Multiscale and Spectral Perspectives, Cambridge University Press (Cambridge).DOI
  62. [76]Yeung, P. K., Sreenivasan, K. R. & Pope, S. B. 2018 Effects of finite spatial and temporal resolution on extreme events in direct numerical simulations of incompressible isotropic turbulence, Phys. Rev. Fluids, 3, 064603.DOI
  63. [77]Yeung, P. K. & Ravikumar, K. 2020 Advancing understanding of turbulence through extreme-scale computation : Intermittency and simulations at large problem sizes, Phys. Rev. Fluids 5, 110517.DOI
  64. [78]Yeung, P. K., Ravikumar, K., Uma-Vaniveswaran, R., Dotson, D. L., Sreenivasan, K. R., Pope, S. B., Meneveau, C. & Nichols, S. 2025 Small-scale properties from exascale computations of turbulence on a 32,768³ periodic cube, J. Fluid Mech., 1019, R2.

Paper details

vlnc:W_28mffjzv
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