betaFeedback

WorksPhysics and Astronomy

The eddies are attached, but it is all right

Abstract

The behavior of velocity fluctuations near a wall has long fascinated the turbulence community, because the prevalent theoretical framework of an attached-eddy hierarchy appears to predict infinite intensities as the Reynolds number tends to infinity. Although an unbounded infinite limit is not a problem in itself, it raises the possibility of unfamiliar phenomena when the Reynolds number is large, and has motivated attempts to avoid it. We review the subject and point to possible pitfalls stemming from uncritical extrapolation from low Reynolds numbers, or from an over-simplification of the multiscale nature of turbulence. It is shown that large attached eddies dominate the high-Reynolds-number regime of the near-wall layer, and that they behave differently from smaller-scale ones. In that limit, the near-wall layer is controlled by the outer flow, the large-scale fluctuations reduce to a local modulation of the near-wall flow by a variable friction velocity, and the kinetic-energy peak is substituted by a deeper structure with a secondary outer maximum. The friction velocity is then not necessarily the best velocity scale. While the near-wall energy peak probably becomes unbounded in wall units, it almost surely tends to zero when expressed in terms of the outer driving velocity.

Introduction

While there is reasonable agreement that turbulence should become independent of viscosity in the limit of very large Reynolds numbers[1, 2], this is not true for wall-bounded turbulent flows, where even bulk quantities such as the friction factor slowly decay when the Reynolds number increases[3]. The reason is that there is a layer near the wall where viscosity is always needed to enforce the no-slip boundary boundary condition and, even if the relative thickness of this layer steadily decreases with the Reynolds number, the classical argument that leads to the logarithmic velocity profile[4] also implies that the velocity gradients grow without limit near the wall, and that their contribution to the production and dissipation of the turbulence fluctuations cannot automatically be neglected.

As a consequence of this singular behavior, much of the discussion about the high-Reynolds number limit of wall-bounded turbulence has centered on the intensity of the near-wall peak of the streamwise velocity fluctuations, up′u'_p, empirically located at a distance yp+≈15y_p^+ \approx15 from the wall, where the ‘+’ superindex denotes ‘wall’ normalization with the friction velocity uτu_\tau and with the kinematic viscosity ν\nu. In numerical simulations, its magnitude approximately increases logarithmically with the friction Reynolds number, Reτ=uτh/νRe_\tau= u_\tau h/\nu, where hh is the flow thickness[5]. The evidence from experiments is more mixed, but not incompatible with numerics when both are available[5, 6, 7], and a similar logarithmic behavior is found in the spanwise velocity and in the pressure, although not in the fluctuations of the wall-normal velocity[5]. The implication that the velocity fluctuations become infinitely strong in the limit of infinitely high Reynolds number has caused some unease, leading to repeated efforts to avoid it[8, 9, 10] and to restore the asymptotic ReτRe_\tau-independence of wall turbulence.

This paper briefly reviews those efforts. One of our first conclusions will be that the available range of experimental and numerical Reynolds numbers is too narrow for an unguided extrapolation to Re→∞Re \to\infty, and is likely to remain so for some time. The decision between different models should rather come from theoretical arguments, if possible, and most of the paper deals with how such theoretical models are constrained by the data. We denote the streamwise, wall-normal and spanwise directions by xx, yy and zz, respectively, and the corresponding velocity components by uu, vv and ww. Capital letters denote yy-dependent ensemble averages, ⟨⟩\langle\rangle, as in the mean velocity profile, U(y)U(y), and lower-case ones are fluctuations with respect to these averages. Primes refer to root-mean-square (rms) fluctuation intensities.

In principle, an infinitely strong near-wall intensity peak does not present insurmountable theoretical difficulties. If we assume an approximately constant tangential Reynolds stress throughout the logarithmic layer, ⟨uv⟩≃−uτ2\langle uv\rangle\simeq-u_\tau^2, and a logarithmic profile for the mean velocity[3],

U+=A+κ−1log⁡y+,(1)U^+ = A + \kappa^{-1}\log y^+, \tag*{(1)}

where A≈5A \approx5, and κ≈0.4\kappa\approx0.4 is the Kármán constant, the total energy production in the flow is[3]

−∫0h⟨uv⟩∂yU dy≈∫δinReτuτ3κy+ dy+∼uτ3κlog⁡Reτ,(2)-\int_0^h \langle uv\rangle\partial_y U\,\mathrm{d}y \approx\int_{\delta_{\mathrm{in}}}^{Re_\tau} \frac{u_\tau^3}{\kappa y^+}\,\mathrm{d}y^+ \sim\frac{u_\tau^3}{\kappa}\log Re_\tau, \tag*{(2)}

which grows without bound with ReτRe_\tau. The reason for the singularity is not the lower bound of the integral in (2), which can be regularized by a suitable viscous cut-off, but the divergence of the integral of the mean shear, ∂yU∼O(1/y)\partial_y U \sim O(1/y), in its upper limit, O(h+=Reτ)O(h^+ = Re_\tau). Unless the logarithmic profile is assumed to cover a vanishingly small fraction of the flow thickness as ReτRe_\tau increases, Eq. (2) implies an infinitely large production of energy, at least when expressed in wall units, that could conceivably leak towards the wall and result in infinitely strong fluctuations in its neighborhood. Note, in particular, that the bulk velocity from Eq. (1), which arguably drives the flow, also becomes infinite as Reτ→∞Re_\tau\to\infty.

However, it is probably true that such a situation would result in extremely strong fluctuations near the wall when ReτRe_\tau is large but finite. There is some evidence of strong fluctuations in the atmospheric boundary layer[11, 12] but, even in that case, their intensity, up′+≈4u_p^{\prime+} \approx4 at Reτ∼O(106)Re_\tau\sim O(10^6), is weak with respect to the mean velocity at the same distance from the wall U+(yp)≈10.5U^+(y_p) \approx10.5. In fact, even using the most unfavorable of the extrapolations discussed below for up′+u_p^{\prime+}, the intensity of the near-wall fluctuation peak only becomes comparable to the mean profile for Reτ≈1074Re_\tau\approx10^{74}, which is well beyond any reasonable extrapolation from experimental or observational Reynolds numbers. More damaging from the theoretical point of view is the assumption that the viscosity required to enforce the boundary conditions remains relevant for infinitely strong velocity structures, and that the associated flow features remain stable.

A recent survey and discussion of currently popular models can be found in Ref. 10, and will not be repeated here. They can broadly be classified in two groups. The first one are structural models that propose mechanisms for how the flow is organized. The best-known is the attached-eddy model, first proposed by Townsend[13], and structurally developed by many others[14, 15, 16, 17]. It is the main support for the unbounded logarithmic growth of the near-wall intensity, and will be discussed later in more detail. The models in the second group typically assume a finite intensity asymptote at Reτ→∞Re_\tau\to\infty, and search for the most mathematically and physically consistent form of the defect with respect to this asymptote. The two best-known proposals are a logarithm[9, 10, 18],

up′2(Reτ)=up′2(∞)−O(1/log⁡Reτ),(3)u_p^{\prime2}(Re_\tau) = u_p^{\prime2}(\infty) - O(1/\log Re_\tau), \tag*{(3)}

and a power law[8, 19],

up′2(Reτ)=up′2(∞)−O(Reτ−1/4),(4)u_p^{\prime2}(Re_\tau) = u_p^{\prime2}(\infty) - O(Re_\tau^{-1/4}), \tag*{(4)}

which are, unfortunately, difficult to distinguish within the available range of Reynolds number. In this paper, we will mostly restrict ourselves to exploring the consequences of the attached-eddy model, both from the point of view of the near-wall fluctuations and of its interplay with the viscous boundary condition.

We will do this by comparing theoretical arguments with a homogeneous set of data presented in §II. The attached-eddy model and its supporting evidence are discussed in §III, and its weak points are discussed in §IV, together with possible solutions. This first part of the paper deals mostly with the fluctuations of the streamwise velocity component, but §V extends the argument to the spanwise velocity. Conclusions are summarized in §VI.

AL,BLA_L, B_LAdP,BdPA_{dP}, B_{dP}AdL,BdLA_{dL}, B_{dL}AS,BSA_S, B_S
u′2u^{\prime2}3.75, 0.6311.5, -19.313.8, -40.04.17, 0.57
w′2w^{\prime2}-0.97, 0.443.9, -10.05.18, -20.9-0.18, 0.34

Table II. Coefficients for the line fits in Fig. 1.

The data set

We will mostly restrict ourselves to numerical doubly periodic pressure-driven turbulent channel flow between parallel plates separated by 2h2h, for which there is a reasonably homogeneous set of simulations spanning two orders of magnitude in Reynolds number. They are summarized in Table I. At the same time, these simulations include enough variety of computational parameters, mainly in the numerical method and in the size of the computational box, to provide some safeguard against overfitting our results to one particular technique. Although restricting us in this way limits our conclusions somewhat, it avoids the scatter due to variable resolution in experiments, while keeping a range of ReτRe_\tau comparable to what can reliably be measured experimentally. Moreover, since our discussion will lead us to consider the effect on the near-wall region of the largest flow scales, the restriction to channels avoids the known differences between their large scales and those in pipes[26] or in boundary layers[18, 27, 28]. Most data sets in table I include mean fluctuation profiles, one- and two-dimensional spectra, and energy balances, and are freely accessible from the web pages of the different groups.

ReferenceLx/hL_x/hLz/hL_z/hReτRe_\tauSymbol
del Álamo et al.[20]8π8\pi3π3\pi180–950○
Hoyas and Jiménez[21]8π8\pi3π3\pi2000○
Lozano-Durán and Jiménez[22]60π60\pi6π6\pi550□
Lozano-Durán and Jiménez[22]2π2\piπ\pi4200□
Bernardini et al.[23]6π6\pi2π2\pi180–4000△
Lee and Moser[24]8π8\pi3π3\pi550–5200◇
Hoyas et al.[25]2π2\piπ\pi10000○

Table I. Numerical data sets most often used in the paper. The box dimensions are Lx×2h×LzL_x \times2h \times L_z, and the symbol in the rightmost column are used consistently in the figures.

Spectra and cospectra are used in premultiplied form, Φab(kx,y)=kx⟨a^b^⟩\Phi_{ab}(k_x,y) = k_x\langle\hat{a}\hat{b}\rangle, where a^(kx)\hat{a}(k_x) is the Fourier coefficient of the expansion of a(x)a(x) in terms of the streamwise wavenumber kxk_x, with similar definitions for spanwise spectra in terms of kzk_z, and for two-dimensional ones in terms of both wavenumbers. They will usually be expressed as functions of the wavelengths, λj=2π/kj\lambda_j = 2\pi/k_j.

The fluctuation profiles and peak intensity of the two wall-parallel velocities are given in Figure 1, which shows that they agree reasonably well among different data sets at similar Reynolds numbers. The growing trend of the profiles is clear in Figure 1(a,Figure 1b), as is the trend of the peak position to slowly move away from the wall when the Reynolds number increases (for which there is also some experimental evidence[30]).

Figure 1. (a) Intensity of the streamwise velocity fluctuations near the wall. The symbols mark the position of the near-wall peak, as in table I. Numerical channels[24, 25, 29]. The Reynolds number increases from blue to red in Reτ=180Re_\tau= 180–10410^4. (b) As in (a), for the spanwise velocity. (c) Peak intensity of the streamwise velocity fluctuations near the wall as a function of the Reynolds number. Symbols as in table I, with standard deviations as vertical bars when available. (d) As in (c), for the spanwise velocity. Lines are different data fits: logarithmic approximation[13], up′2+=AL+BLlog⁡(Reτ)u_p^{\prime2+}=A_L+B_L\log(Re_\tau); power defect law[8], up′2+=AdP+BdPReτ−1/4u_p^{\prime2+}=A_{dP}+B_{dP}Re_\tau^{-1/4}; logarithmic defect law[10], up′2+=AdL+BdL/log⁡(Reτ)u_p^{\prime2+}=A_{dL}+B_{dL}/\log(Re_\tau); shifted logarithm, up′2+=AS+BSlog⁡(Reτ−150)u_p^{\prime2+}=A_S+B_S\log(Re_\tau-150). Coefficients are collected in table II.

The peak intensities are collected in Figure 1(c,Figure 1d), with line fits from the models discussed above, with coefficients either taken from the original publications or fitted numerically to the data. They separate in two groups. The straight line is the basic logarithm already discussed above for the attached-eddy model[5, 21], which predicts an infinite intensity as Reτ→∞Re_\tau\to\infty. The second group includes the dashed and chaindotted curves in each figure[8, 10], both of which assume a finite limit for the intensity, as in Eqs. (3) and (4). The defect formulations appear to represent the data better than the straight line of the logarithm, especially if we disregard the lowest Reynolds number, Reτ≈180Re_\tau\approx180, but the most striking observation is how similar to each other they are. Finally, the red dotted lines in the two figures are a shifted logarithm with a virtual origin for ReτRe_\tau, which predicts an unbounded peak at large ReτRe_\tau. It is not intended as a serious proposal, and I am not aware of any theoretical basis for it (although neither is it absurd to shift the Reynolds number by a transition threshold), but it shows that there are simple approximations that fit the available data as well as the defect laws, while predicting an infinite limiting value for the near-wall peak. It emphasizes that simple curve fitting cannot decide the issue.

It may be relevant at this point to estimate which would be the Reynolds number required to distinguish between the different fits in Figure 1(c,Figure 1d). An order of magnitude could be how far the logarithmic straight line has to be extended before it reaches the asymptotic value of any of the two defect laws. The details depend on the approximation and on the variable chosen, but it is in all cases of the order of Reτ≈105Re_\tau\approx10^5. This is at least 15–20 years in the future for numerical simulations, but it is worth remarking that experiments in pipes[30] up to Reτ≈4×104Re_\tau\approx4 \times10^4 have proved inconclusive for this purpose[10], and that data from the atmospheric surface layer at Reτ=O(106)Re_\tau= O(10^6), although not strictly a channel, appear to follow the logarithmic trend reasonably well[11].

The attached-eddy model

The key theoretical contribution to understanding the velocity fluctuations was made by Townsend[13], who noted that the usual argument that there is an overlap layer in which neither the viscous unit of length nor the flow thickness are relevant[4] should also apply to them. Although the resulting attached-eddy model has been reviewed often, we will recall it here to identify where the logarithmic prediction could go wrong.

The naive argument is that, if there is no scale for lengths, but uτu_\tau is a scale for the velocities, the functional dependence of any variable with dimensions of velocity on a variable with dimensions of length should be logarithmic (see appendix A in Ref. 5). We could thus expect that u′/uτ∼log⁡(y/h)u^\prime/u_\tau\sim\log(y/h), which, when particularized at a fixed inner viscous limit, yp+y_p^+, results in up′/uτ∼log⁡Reτu_p^\prime/u_\tau\sim\log Re_\tau. There is substantial experimental and numerical evidence for both approximate logarithmic behaviors[31], which, as mentioned above, extend to the fluctuations of the pressure and of the spanwise velocity[32] but, in the absence of a rigorous theory, it is always possible that higher Reynolds numbers than those currently available may lead to something different.

Moreover, there are logical flaws in the previous argument. Most obviously, an argument similar to the one used for u′u' applies to any power of the velocity, and some selection rule is needed to decide which power to use. In addition, some reason needs to be found for why the logarithmic behavior applies to u′u' and w′w' but, as mentioned above, not to v′v', and a justification is required for why uτu_\tau is the right scaling unit for the velocity fluctuations.

In general, it is unwise to use similarity arguments without a dynamical model, and Townsend[13] proposed that the logarithm is implemented by the superposition of a self-similar family of ‘attached’ Reynolds-stress-carrying eddies linking the wall to the interior of the flow. In the absence of a fixed length scale, their height is proportional to their wall-parallel size (λx\lambda_x or λz\lambda_z), while their intensity is O(uτ)O(u_\tau) because each eddy family is responsible for carrying the tangential Reynolds stress (−uv-uv) at one distance from the wall. This is supported by observations: the solid contours in Figure 2(a,Figure 2b) are the premultiplied cospectra, Φuv+(kx)\Phi_{uv}^{+}(k_x) and Φuv+(kz)\Phi_{uv}^{+}(k_z) of the tangential Reynolds stress, respectively drawn as functions of yy and of the streamwise or spanwise wavelengths. The contours are normalized with the friction velocity, and the figures show that a stress −uv∼O(uτ2)-uv \sim O(u_\tau^2) is concentrated along a spectral ridge in which λx≈25y\lambda_x \approx25y and λz≈5y\lambda_z \approx5y, and which extends from a minimum wavelength that scales in wall units to an outer one that scales with hh. Since it can be shown that the streamwise and wall-normal velocities are well correlated in the regions of the λ−y\lambda-y plane where the Reynolds stresses are substantial, this extends the role of uτu_\tau from a velocity scale for the overall intensity to one for individual spectral bands, although the argument may not apply to ’sterile’ scales without tangential stress.

Figure 2. (a) Solid contours are premultiplied cospectra of the tangential Reynolds stress, −Φuv+(λx,y)=[0.2,0.8]-\Phi_{uv}^{+}(\lambda_x,y)=[0.2,0.8] times the common maximum of the four cospectra. Dashed ones are premultiplied spectra of the spanwise vorticity, Φωzωz+(λx,y)=[0.05,0.4]\Phi_{\omega_z\omega_z}^{+}(\lambda_x,y)=[0.05,0.4] the common maximum. The dashed diagonal is λx=25y\lambda_x=25y. (b) As in (a), but as functions of λz\lambda_z. The dashed diagonal is λz=5y\lambda_z=5y. (c) Two-dimensional integrated spectra as functions of (λx,λz)(\lambda_x,\lambda_z): Solid contours are −∫Φuv+ dy+-\int\Phi_{uv}^{+}\,dy^{+} integrated over y∈[0.005,0.5]λxy\in[0.005,0.5]\lambda_x. Contours are [0.2,0.8][0.2,0.8] the common maximum. Dashed contours are premultiplied spectra of the spanwise vorticity, ∫Φωzωz+ dy+\int\Phi_{\omega_z\omega_z}^{+}\,dy^{+}, integrated over y+∈[0,20]y^{+}\in[0,20]. Contours are [0.05,0.4][0.05,0.4] the common maximum. The dashed diagonal is λx=5λz\lambda_x=5\lambda_z. Numerical channels[22, 24], with the Reynolds number increasing from blue to red: Reτ=550Re_\tau=550, 1000, 2000, 5200.

Because there is no fixed length scale, these bands are logarithmic (e.g. from λx\lambda_x to 2λx2\lambda_x), and the number of eddy families found at a given distance yy from the wall is the number of logarithmic bands required to cover the range of lengths from yy to the flow thickness, O(h)O(h), which increases logarithmically with h/yh/y. If we further assume that the velocity fluctuations of the different bands are uncorrelated, their variances add, resulting in a logarithmic behavior with the Reynolds number for u′2‾(y)∼log⁡(y/h)\overline{u'^{2}}(y)\sim\log(y/h) and for u′2‾(yp)\overline{u'^{2}}(y_p). Townsend[13] distinguishes between ‘inactive’ attached variables, such as u′u' and w′w', which are not sufficient to generate tangential Reynolds stress and are only damped by the wall in a thin viscous layer, and ‘active’ ones, like v′v', that contribute to the stress but are inhibited by impermeability at distances from the wall comparable to their size. Only the former should have logarithmic profiles.

The difference between the two types of variables can be seen in Figure 2(a,Figure 2b), where the dashed isolines are spectra of the spanwise vorticity. Bradshaw 33 argued that, in the absence of other terms of the momentum equation, the flow at long wavelengths, and at distances from the wall below the Reynolds-stress ridge, can only be driven by the pressure footprint of active eddies further from the wall and should therefore be essentially irrotational. It is indeed clear from 2(a,b) that wavelengths longer than λx+≈104\lambda_x^+ \approx10^4 are approximately irrotational above y+≈10y^+ \approx10, and Fig. 12(d) in Ref. 5 shows that the pressure is a deep variable, correlated from the wall to the Reynolds-stress ridge. But potential flow cannot satisfy the no-slip condition at the wall, and a viscous rotational layer appears at long wavelengths below y+≲10y^+ \lesssim10. Since ωz≈−∂yu\omega_z \approx-\partial_y u in that region, this implies that non-trivial fluctuations of uu reach the vicinity of the wall at those wavelengths.

The solid contours in Figure 2(c) are two-dimensional Reynolds-stress cospectra integrated over the active band of wall distances corresponding to each wavelength, and the dashed ones are spectra of the spanwise vorticity integrated over the viscous near-wall layer. The figure strongly supports that the latter are the effect of the detached Reynolds stresses, although not necessarily at the same distance from the wall for all wavelengths. The same is implied by the advection velocity of the long wavelengths near the wall, which is not the local mean flow velocity[5, 34], but the velocity at y/h≈0.3−0.4y/h \approx0.3 - 0.4. However, the conclusion that they are driven by pressure is less clear. Pressure correlations only attach to the wall for λx/h≲4\lambda_x/h \lesssim4, and cannot therefore drive longer wavelengths[28]. In fact, it will be seen below that those scales are better described as internal turbulent layers, presumably driven from above by turbulent diffusion[5].

Families of self-similar attached eddies, as well as the distinction between active and inactive motions, have been observed and characterised in some detail numerically[5, 20, 35] and experimentally[17, 36, 37, 38].

There are several debatable points in these arguments, most of which have been discussed in the literature and will not be repeated here, but some of which deserve closer attention. A well-known limitation of the attached-eddy model is that it does not include viscosity, whose effects are lumped in the rule that something happens below yp+=15y_p^+ = 15. This has occasionally been suggested as responsible for scaling failures[10]. We have seen that what viscosity does is to generate the thin vortex layers in Figure 2, but how they are maintained, their dimensions, and their effect are unclear, and we mentioned in the introduction that their stability is problematic if the fluctuations become too strong.

However, the overall conclusion from the previous discussion remains that the reason why velocity fluctuations do not scale well is that they contain a wide range of wavelengths, even when they are very close to the wall. This often-made point[14, 15, 21, 39] is clear from the spectra in Figure 3 of the near-wall streamwise velocity at yp+=15y_p^+ = 15. There is a universal ‘core’, centered at

Figure 3. Premultiplied energy spectrum of the streamwise velocity, Φuu+\Phi_{uu}^+ at y+=15y^+ = 15, versus the wall-parallel wavelengths. Numerical channels[22, 24], with the Reynolds number increasing from blue to red: Reτ=550Re_\tau= 550, 1000, 2000, 5200. Contours are [0.15,0.4,0.8][0.15, 0.4, 0.8] times the common maximum of the four spectra. The dashed diagonal is λx=12λz\lambda_x = 12\lambda_z, and symbols are λz=h\lambda_z = h. The translucent patch is used in Figure 5.

λx+×λz+≈1000×100\lambda_x^+ \times\lambda_z^+ \approx1000 \times100, which does not reach above y+≈100y^+ \approx100 (Figure 2) and collapses well across ReτRe_\tau. That its dimensions scale in wall units shows that it is controlled by viscosity[40, 41, 42], but it is accompanied by larger-scale spectral tails that extend to λ∼O(h)\lambda\sim O(h), and which are responsible for the extra energy of the overall fluctuations. The dynamics of these tails, which do not have to be the same as for the viscous core, has to be studied if the overall peak intensity is to be understood.

This wide range of scales is a problem for asymptotic models that seek to represent the flow in terms of an outer ‘regular’ turbulence and a small-scale object near the wall, since the latter does not exist as such in real flows. In fact, given that the total peak energy includes contributions from a wide range of wavelengths, it is unclear whether it makes sense to speak of a single velocity and length scale for it.

The large scales

Figures 4(a,c) are similar to the cospectra in Figure 2(a,Figure 2b), but drawn for the streamwise velocity component. To facilitate comparison, the dashed diagonals in the two figures are the same, and show that the upper limit of the kinetic energy and of the Reynolds stress are similar, but it is clear from Figure 2 and Figure 4 that the streamwise velocity spectrum extends to the wall, while the cospectrum stays away from it. Notice, in particular, the different behavior of the larger wavelengths.

Four panels showing premultiplied streamwise-velocity spectra and their maxima against streamwise and spanwise wavelengths

Figure 4. (a,c) Contours are the premultiplied spectrum of the streamwise velocity. The contours are [0.2, 0.8] times the overall maximum of all the Φuu+\Phi_{uu}^{+}. Dashed and chain-dotted lines are the location of the inner and outer maxima of Φuu\Phi_{uu} at each wavelength. Numerical channels22,24^{22,24}, with the Reynolds number increasing from blue to red: Reτ=550Re_{\tau} = 550, 1000, 2000, 5200. Symbols are boundary layers from Ref. 43: ∘\circ, Reτ=1014Re_{\tau} = 1014; △\triangle, 1844; or from Ref. 17, □\square, Reτ=14000Re_{\tau} = 14000. (b,d) Maximum of Φuu+\Phi_{uu}^{+} versus wavelength, with the same lines and symbols as above. (a,b) Against λx\lambda_x. (c,d) Against λz\lambda_z. The dashed magenta line in (a) and (c) are λx+=25y+\lambda_x^{+} = 25y^{+} and λz+=5y+\lambda_z^{+} = 5y^{+}, respectively. The ones in (b) and (d) are Φuu,max⁡+=0.315log⁡(λx+)−1.8\Phi_{uu,\max}^{+} = 0.315\log(\lambda_x^{+}) - 1.8 and Φuu,max⁡+=0.315log⁡(λz+)−0.5\Phi_{uu,\max}^{+} = 0.315\log(\lambda_z^{+}) - 0.5.

Figure 4 includes as thick dashed lines and symbols the wall-normal location of the maximum of Φuu\Phi_{uu} at each wavelength. We saw when discussing Figure 1 that the energy peak drifts slowly away from the wall with the Reynolds number, and Figure 4(a,Figure 4c) show a clearer dependence on the wavelengths. The symbols are from experimental boundary layers^{17,43} and, although sparser near the wall, are compatible with the numerics. The inner maximum at the shorter wavelengths marks the edge of the viscous layer, and is superseded beyond λx+≈104\lambda_x^{+} \approx10^4 by an outer maximum located at ymax⁡+≈λz+/5y_{\max}^{+} \approx\lambda_z^{+}/5 or λx+/25\lambda_x^{+}/25 (Figure 4a,Figure 4c). Note that this implies that the transition to an outer peak always takes place when ymax⁡+≈200y_{\max}^{+} \approx200, which can be interpreted as the thickness at which the near-wall viscous layers become unstable.

The amplitude of these maxima is shown in Figure 4(b,Figure 4d). It collapses well with ReτRe_{\tau} in the viscosity-dominated region, λx+≲104\lambda_x^{+} \lesssim10^4 or λz+≲103\lambda_z^{+} \lesssim10^3, but not at the larger wavelengths associated with the outer maximum. Their growth with λ\lambda can be interpreted as a wavelength-by-wavelength counterpart to the logarithmic growth of up′2u_p^{\prime2} with ReτRe_{\tau}, represented here by the range of wavelengths over which the one-dimensional spectrum is summed. We have added to Figure 4(b,Figure 4d) logarithmic approximations to this growth, using half the slope in Figure 1 for up′2u_p^{\prime2}, on the assumption that half of the peak growth is due to the wider range of λx\lambda_x and the other half to the wider range of λz\lambda_z. These logarithms match the data relatively well, but they should only be considered as aids to the eye, subject to the same caveats as the curve fits in Figure 1(a,Figure 1b).

Fortunately, more can be said about the large flow scales. The outer maximum of the longest structures in Figure 4(a) represents layers of dimensions (λx×y×λz)+=O(20000×500×4000)(\lambda_x \times y \times\lambda_z)^{+} = O(20000 \times500 \times4000), which are thin both with respect to the channel height (Reτ=5200Re_{\tau} = 5200) and to their own length. They are also deep enough to be fully turbulent. Internal turbulent layers are common when wall-bounded flows cross boundaries between different types of wall or are otherwise perturbed, and have been extensively studied in meteorology^{44,45} and for heterogeneous surfaces^{46}. Using a rough approximation^{45}, there is an internal equilibrium layer (EL) whose thickness after time Δt\Delta t grows to δEL∼uτΔt/4\delta_{\mathrm{EL}} \sim u_{\tau}\Delta t/4, and satisfies the universal velocity profile corresponding to the friction velocity that develops after the perturbation. Since the lifetime of the stress-carrying structures is^{47} Δt≈Δy/uτ≈λz/uτ\Delta t \approx\Delta y/u_{\tau} \approx\lambda_z/u_{\tau}, the thickness of the EL generated by the passage of perturbations of width λz\lambda_z is δEL≈λz/4\delta_{\mathrm{EL}} \approx\lambda_z/4, which approximately agrees with the location of the outer maximum in Figure 4(c).

The model is that large-scale fluctuations with wavelengths of the order of λx+≈5λz+≳104\lambda_x^{+} \approx5\lambda_z^{+} \gtrsim10^4 (Figure 2c) are equilibrium turbulent boundary layers satisfying the universal profile U=uτF(y+)U = u_{\tau}F(y^{+}) with a perturbed friction velocity, and that, if they are viewed as perturbations to the overall mean velocity, they can be modeled as weak perturbations of uτ→uτ+δuτu_{\tau} \to u_{\tau} + \delta u_{\tau}. Upon linearization,

(δU)=(δuτ)(F+y+dFdy+)≡(δuτ)G(y+),(5)(\delta U) = (\delta u_{\tau})\left(F + y^{+}\frac{\mathrm{d}F}{\mathrm{d}y^{+}}\right) \equiv(\delta u_{\tau})G(y^{+}), \tag*{(5)}

which links the rms intensity of the velocity fluctuations to the rms perturbation of the friction velocity, (δuτ)′=u∗(\delta u_\tau)' = u_*. In the viscous layer, F=y+F = y^+, and

(δU)′/u∗=2y+.(6)(\delta U)'/u_* = 2y^+. \tag*{(6)}

This is essentially the modulation described in [48], which was shown in [49] to reduce in the buffer layer to a modulation of uτu_\tau. We extend it here to the logarithmic layer, where FF satisfies (1), and

(δU)′/u∗=κ−1(1+Aκ+log⁡y+).(7)(\delta U)'/u_* = \kappa^{-1}(1+A\kappa+\log y^+). \tag*{(7)}

These equations are tested in figure 5(a) for spectral densities within the patch of large-scale tails in Figure 3. The perturbation intensity, u∗u_*, of the friction velocity has been adjusted for each wavelength to fit Eq. (5) to the spectral profile from the wall to the location of the spectral maximum, but the definition of FF has not been modified. The symbols in Figure 5(a) are G(y+)G(y^+) in Eq. (5), computed from the mean velocity F(y+)F(y^+) of a turbulent channel, and the agreement is excellent, supporting the assumption of turbulent equilibrium internal layers. Note that the lowest Reτ=550Re_\tau= 550 is not in the figure because it never develops an outer peak. This model was probably first used by Bradshaw [33] to explain the differences between boundary layers subject to different streamwise pressure gradients. Some time later, Spalart [50] tested it on simulations of a zero-pressure-gradient boundary layer at Reτ≈400Re_\tau\approx400, and concluded that its large scales were better represented by laminar oscillating Stokes layers, but the comparison with Figure 4 shows that his Reynolds number was too low, and his wavelength too short (λx+≈103\lambda_x^+ \approx10^3), falling in the viscous range of Figure 4(a).

Figure 5. (a) Spectral density of the streamwise velocity for various wavelengths, from λx=2×104\lambda_x = 2 \times10^{4} to λx=25h\lambda_x = 25h, from blue to red, integrated in λz\lambda_z over the diagonal band in Figure 3, defined as λz∈λx/[30,5]\lambda_z \in\lambda_x/[30, 5]. The symbols are GG in Eq. (5), computed from the mean velocity profile in [24]. (b) Friction velocity perturbation intensities obtained from fitting (a). Channels[24]: (—·—·— and □) Reτ=1000Re_\tau= 1000; (— — — and ○) Reτ=2000Re_\tau= 2000; (——— and △) Reτ=5200Re_\tau= 5200. (c) Lines are fluctuation profiles estimated from Eq. (8). From bottom to top: Reτ=1000Re_\tau= 1000, 10410^{4}, 10510^{5} and 2×1062 \times10^{6}. Symbols are channel simulations[24, 25] and the atmospheric surface layer[12] at those Reynolds numbers. The dashed line is us′2+u_s^{\prime2+}, estimated from Reτ=1000Re_\tau= 1000.

More recently, Pirozzoli [19] has applied the equilibrium layer model to pipes, with results similar to Figure 5. However, the amplitude of his perturbations depends on λ+\lambda^+, while Figure 5(b) shows that u∗u_* scales with λx/h\lambda_x/h rather than in wall units. The reason for this discrepancy is unknown, and deserves further investigation. The rms of the fluctuations of uτu_\tau tend to some non-zero value when λx→0\lambda_x \to0, so that the fluctuation velocity profile can be approximated as

u′2+(y+)=us′2+(y+)+G2+(y+)∫λ1/hλ2/hu∗2+(λ/h) dlog⁡λ,(8)u'^{2+}(y^+) = u_s'^{2+}(y^+) + G^{2+}(y^+) \int_{\lambda_1/h}^{\lambda_2/h} u_*^{2+}(\lambda/h)\,\mathrm{d}\log\lambda, \tag*{(8)}

where G(y+)G(y^+) is defined in Eq. (5), λ1=max⁡(λin,25y)\lambda_1 = \max(\lambda_{\mathrm{in}},25y), and us′2+u_s'^{2+} is the contribution of scales whose λx\lambda_x is shorter than the wavelength, λin\lambda_{\mathrm{in}}, at which the viscous sublayer becomes unstable and fluctuations have to be modeled as turbulent profiles. It is important to note that the lower limit of this integral scales in wall units (λin+≈2×104\lambda_{\mathrm{in}}^+ \approx2 \times10^4 in Figure 4), while the upper one either extends to infinity, or scales in outer units (λ2=25h\lambda_2 = 25h in our case, because of the length of our computational box). Assuming that the small-scale contribution to Eq. (8) is independent of ReτRe_\tau, and that the integral of u∗2u_*^2 stays bounded at λ→∞\lambda\to\infty, the dominant contribution to Eq. (8) at high ReτRe_\tau comes from the lower limit of the integral,

u′2+(y+)≈us′2+(y+)+G2+(y+)u∗2(0)log⁡(Reτ/λin+).(9)u'^{2+}(y^+) \approx u_s'^{2+}(y^+) + G^{2+}(y^+)u_*^2(0)\log(Re_\tau/\lambda_{\mathrm{in}}^+). \tag*{(9)}

It is important to realize that U+δUU+\delta U in Eq. (5) is the mean profile of a perturbed boundary layer. These faster- or slower-than-average local equilibrium layers have their own small-scale perturbations that are modulated by the outer Reynolds-stress structures, but these are second-order effects, negligible with respect to the mean profile. The maximum fluctuation intensity at the inner energy peak is up′2+≈9u_p'^{2+} \approx9 in our data base, while the mean velocity at that point is U+(yp)≈100U^{+}(y_p) \approx100. The fluctuation profiles are only relevant if the mean profile can be considered fixed, but any perturbation of the latter overwhelms the modulation of the small scales.

(9) is also a logarithm that diverges as Reτ→∞Re_\tau\to\infty, but its most interesting aspect is that the form of the large-scale correction is not a peak near the wall, but something similar to the mean velocity profile of a regular boundary layer, so that the fluctuation peak will be absorbed into something closer to a plateau at high ReτRe_\tau. Although estimating this behavior necessarily implies extrapolation from lower Reynolds numbers, Figure 5(c) plots Eq. (8) for several large ReτRe_\tau. The lowest curve in the figure, Reτ=1000Re_\tau= 1000 is used to estimate the viscous contribution, us′u_s', displayed in the figure as a dashed line, and is therefore automatically fitted. But the corrections for the rest of the curves are computed by estimating the integral in Eq. (8), using a lineal least-square approximation to u∗u_* in Figure 5(b). The agreement with our highest Reynolds number, Reτ=104Re_\tau= 10^4 is excellent, and it is intriguing that the uppermost curve in the figure, intended to approximate data from the atmospheric surface layer[12], also appears to fit well. The fluctuation profile in this case is already very different from that at lower Reynolds numbers, including a second outer maximum of u′u' that requires experimental confirmation. Although the implied Reτ=O(106−107)Re_\tau= O(10^6-10^7) are well in the future for laboratory or numerical flows, they are not out of range for geophysical ones[11, 12]. In fact, some measurements in the atmospheric surface layer show a clear double peak in the profile of u′u', but there is some uncertainty as to whether this may be partly due to damping of the near-wall peak by roughness, buoyancy, or instrumental effects[51].

The spanwise velocity

Up to now we have mostly dealt with the streamwise velocity fluctuations, but it is clear from Figure 1 that the spanwise velocity also has a potentially infinite limit. Figure 6 displays the two-dimensional spectrum of w′w', and shows that the reason is also the effect of large structures. Their maximum size also scales with hh rather than in wall units, but they are wide rather than long, as required by continuity[52].

Figure 6. Premultiplied energy spectrum of the spanwise velocity, Φww+\Phi^+_{ww} at y+=50y^+ = 50, versus the wall-parallel wavelengths. Numerical channels[22, 24] with the Reynolds number increasing from blue to red: Reτ=550Re_\tau= 550, 1000, 2000, 5200. Contours are [0.15, 0.4, 0.8] of the common maximum of the four spectra. The dashed diagonal is λx=6λz\lambda_x = 6\lambda_z and symbols are λz=h\lambda_z = h. The translucent patch is used in Figure 7(c,Figure 7d).

The perturbation expansion is also slightly different from the streamwise component. The first term in the expression for GG in Eq. (5) comes from the incremental boundary layer created by the perturbation of uτu_\tau, which in this case has to be oriented spanwise. The second term is the deformation of the existing boundary layer by the change in length scale due to the new uτu_\tau, and is missing from δW\delta W, for which no preexisting spanwise flow exists. The fluctuation equation becomes,

(δW)′=(δwτ)′F(y+)=w∗F(y+),(10)(\delta W)^\prime= (\delta w_\tau)^\prime F(y^+) = w_* F(y^+), \tag*{(10)}

where F(y)F(y) is, as before, the mean velocity profile of a regular channel, and the symbols wτw_\tau and w∗w_* have been introduced to represent the friction velocity of the spanwise perturbation flow.

Figure 7 summarizes, for the spanwise velocity, the same information as Figure 4 and Figure 5 do for uu. As in the previous case, the relevant result is that the transition from the inner to the outer peak scales in wall units, at approximately the same wavelength as for the streamwise velocity, λz+≈1000\lambda_z^+ \approx1000 (Figure 7a,Figure 7b). Figure 7(d) shows that the perturbation w∗w_* required to fit the profiles in Figure 7(c) to Eq. (10) scales well in outer units, so that the total energy also diverges logarithmically. Notice that the fit of Eq. (10) to the data in Figure 7(c) is as good as that in Figure 5, even if the two predicted profiles are fairly different.

Figure 7. (a) Premultiplied spectrum of the spanwise velocity versus λz\lambda_z. The contours are [0.2,0.8][0.2, 0.8] times the overall maximum of all the Φww+\Phi^+_{ww}. Dashed and chaindotted lines are the location of the inner and outer maxima of Φww\Phi_{ww} at each wavelength. Numerical channels[22, 24], with the Reynolds number increasing from blue to red: Reτ=550,1000,2000,5200Re_\tau= 550, 1000, 2000, 5200. The dashed magenta line is λz+=5y+\lambda_z^+ = 5y^+. (b) Maximum of Φww\Phi_{ww} versus wavelength, with the same lines as in (a). (c) Spectral density of the spanwise velocity for various wavelengths from λz+=2×103\lambda_z^+ = 2 \times10^3 to λz=10h\lambda_z = 10h, from blue to red, integrated in λx\lambda_x over the diagonal band in Figure 6, defined as λx∈[2,20]λz\lambda_x \in[2, 20]\lambda_z. The symbols are the perturbation equilibrium profile in Eq. (10), computed from the mean velocity profile in[24]. (d) Friction velocity perturbation intensities obtained from (c). (— · — and ▽\triangledown) Channel[22], Reτ=550Re_\tau= 550; the rest are channels[24]: (— and □\square) Reτ=1000Re_\tau= 1000; (— — — and ∘\circ) Reτ=2000Re_\tau= 2000; (— — — and △\triangle) Reτ=5200Re_\tau= 5200.

There are some differences between the two velocity components. The Φww\Phi_{ww} spectra in Figure 7(a) are consistently wider and taller than Φuu\Phi_{uu} in Figure 4(c). The outer peak of Φww\Phi_{ww} follows λz≈10y\lambda_z \approx10y instead of λz≈5y\lambda_z \approx5y, and the distinction between inner and outer intensities in Figure 7(b) is much less clear than in Figure 4(d). While the inner and outer peaks of Φuu\Phi_{uu} are often two distinct maxima separated in yy, those of Φww\Phi_{ww} are a single maximum whose location depends on λ\lambda. There are essentially no bimodal intensity profiles in Φww\Phi_{ww}.

The distribution of perturbation friction velocities is also different in the two cases. Although both scale in outer units, the distribution in Figure 5(b) is fairly smooth, with what appears to be a definite limit at λx=0\lambda_x = 0, but Figure 7(d) could have a more singular limit at that point.

Discussion and conclusions

In summary, we have seen that, as most things in turbulence, the near-wall region is a multi-scale flow involving wide ranges of length and width. In consequence, the near-wall peak of the stream- and spanwise velocity fluctuation intensities cannot be modeled as an elemental object. There is a viscosity-dominated spectral core, centered at λx+×λz+≈1000×100\lambda_x^+ \times\lambda_z^+ \approx1000 \times100, which embodies the classical turbulence cycle and scales well across Reynolds numbers, and a large-scale component that behaves very differently. The high-Reynolds-number limit of the energy depends on those large scales, which we have explored using spectra. It turns out that they behave near the wall as internal equilibrium boundary layers, driven by the outer large eddies[5, 33]. Although this means that they are not essentially different from the mean flow, and share with it the scaled mean profile and physical parameters, they mimic oscillating Stokes layers when they are expressed as fluctuations in a Fourier expansion, and become part of what is known at low Reynolds numbers as the near-wall fluctuation peak.

The intensity of these large-scale fluctuations is predicted to increase logarithmically with ReτRe_\tau when expressed in wall units, at least within the Reynolds number range of available experiments, but not to remain concentrated near the wall. At extremely large Reynolds numbers, they should spread to a fraction of the channel thickness, of the order O(h/5)O(h/5). We have shown that their profile can be extracted at laboratory Reynolds numbers from the vertical structure of the spectrum at particular wavelengths. When extrapolated to geophysical Reynolds numbers, they result in a fairly different fluctuation profile, interestingly including a second maximum away from the wall.

In fact, the natural consequence of a model in which most of the near-wall kinetic energy at high Reynolds number is due to interactions with the outer flow is that even this near-wall region, and any part the boundary layer whose distance from the wall is fixed in wall units, should be considered as directly driven by the outer flow. Under those circumstances, and even if the velocity scale for the active turbulence motions responsible for the Reynolds stresses at a given distance from the wall continues to be the friction velocity, a more natural unit for the integrated inactive motions may be some measure of the driving velocity (U∞U_\infty, UbulkU_{bulk} or some velocity combination such as the mixed scaling in Ref. 6). It is interesting to note that, if we recall from Eq. (1) that U∞∼log⁡ReτU_\infty\sim\log Re_\tau, even a logarithmic growth of uτ′2u_\tau^{\prime2} implies that uτ′/U∞∼Reτ−1/2→0u_\tau^\prime/U_\infty\sim Re_\tau^{-1/2} \to0 as Reτ→∞Re_\tau\to\infty. The nature of the singularity is not that uτ′u_\tau^\prime tends to infinity, but that uτu_\tau tends to zero faster than uτ′u_\tau^\prime.

Acknowledgments

This work was supported by the European Research Council under the Caust grant ERC-AdG-101018287. I am grateful to R. Deshpande, G. Kunkel, M.K. Lee and I. Marusic for the use of their original data, and to S. Pirozzoli for his thoughtful critique of an early version of this manuscript.

References

  1. [1]A. N. Kolmogorov, “The local structure of turbulence in incompressible viscous fluid for very large Reynolds numbers,” Dokl. Akad. Nauk SSSR 30, 209–303 (1941).
  2. [2]L. Onsager, “Statistical hydrodynamics,” Nuovo Cimento Suppl. 6, 279–286 (1949).DOI
  3. [3]H. Tennekes and J. L. Lumley, A first course in turbulence (MIT Press, 1972).DOI
  4. [4]C. B. Millikan, “A critical discussion of turbulent flows in channels and circular tubes,” in Proc. 5th Intl. Conf. on Applied Mechanics (Wiley, 1938) pp. 386–392.
  5. [5]J. Jiménez, “Coherent structures in wall-bounded turbulence,” J. Fluid Mech. 842, P1 (2018).
  6. [6]D. B. deGraaff and J. K. Eaton, “Reynolds number scaling of the flat-plate turbulent boundary layer,” J. Fluid Mech. 422, 319–346 (2000).
  7. [7]I. Marusic, W. J. Baars, and N. Hutchins, “Scaling of the streamwise turbulence intensity in the context of inner-outer interactions in wall turbulence,” Phys. Rev. Fluids 2, 100502 (2017).DOI
  8. [8]X. Chen and K. R. Srnivasan, “Reynolds number scaling of the peak turbulence intensity in wall flows,” J. Fluid Mech. 908, R3 (2021).DOI
  9. [9]P. Monkewitz, “Asymptotics of streamwise Reynolds stress in wall turbulence,” J. Fluid Mech. 931, A18 (2022).DOI
  10. [10]Y. Hwang, “Near-wall streamwise turbulence intensity as Reτ → ∞,” Phys. Rev. Fluids 9, 044601 (2024).DOI
  11. [11]M. M. Metzger and J. C. Klewicki, “A comparative study of near-wall turbulence in high and low Reynolds number boundary layers,” Phys. Fluids 13, 692–701 (2001).DOI
  12. [12]M. M. Metzger, J. C. Klewicki, K. L. Bradshaw, and R. Sadr, “Scaling of near-wall axial turbulent stress in the zero pressure gradient boundary layer,” Phys. Fluids 13, 1819–1821 (2001).DOI
  13. [13]A. A. Townsend, “Equilibrium layers and wall turbulence,” J. Fluid Mech. 11, 97–120 (1961).DOI
  14. [14]A. E. Perry and M. S. Chong, “On the mechanism of wall turbulence,” J. Fluid Mech. 119, 173–217 (1982).DOI
  15. [15]A. E. Perry, S. M. Henbest, and M. S. Chong, “A theoretical and experimental study of wall turbulence,” J. Fluid Mech. 165, 163–199 (1986).DOI
  16. [16]A. J. Smits, B. J. McKeon, and I. Marusic, “High-Reynolds number wall turbulence,” Ann. Rev. Fluid Mech. 43, 353–375 (2011).DOI
  17. [17]R. Deshpande, J. P. Monty, and I. Marusic, “Active and inactive components of the streamwise velocity in wall-bounded turbulence,” J. Fluid Mech. 914, A5 (2021).DOI
  18. [18]P. Monkewitz and H. M. Nagib, “Large-Reynolds-number asymptotics of the streamwise normal stress in zero-pressure-gradient turbulent boundary layers,” J. Fluid Mech. 783, 474–503 (2015).DOI
  19. [19]S. Pirozzoli, “On the streamwise velocity variance in the near-wall region of turbulent flows,” J. Fluid Mech. 989, A5 (2024).
  20. [20]J. C. del Álamo, J. Jiménez, P. Zandonade, and R. D. Moser, “Self-similar vortex clusters in the logarithmic region,” J. Fluid Mech. 561, 329–358 (2006).DOI
  21. [21]S. Hoyas and J. Jiménez, “Scaling of the velocity fluctuations in turbulent channels up to Reτ = 2003,” Phys. Fluids 18, 011702 (2006).DOI
  22. [22]A. Lozano-Durán and J. Jiménez, “Effect of the computational domain on direct simulations of turbulent channels up to Reτ = 4200,” Phys. Fluids 26, 011702 (2014).
  23. [23]M. Bernardini, S. Pirozzoli, and P. Orlandi, “Velocity statistics in turbulent channel flow up to Reτ = 4000,” J. Fluid Mech. 742, 171–191 (2014).
  24. [24]M. K. Lee and R. D. Moser, “Direct numerical simulation of turbulent channel flow up to Reτ ≈ 5200,” J. Fluid Mech. 774, 395–415 (2015).DOI
  25. [25]S. Hoyas, M. Oberlack, S. Kraheberger, F. Alcántara-Ávila, and J. Laux, “Wall turbulence at high friction Reynolds numbers,” Phys. Rev. Fluids 7, 014602 (2022).DOI
  26. [26]H. C. H. Ng, J. P. Monty, N. Hutchins, M. S. Chong, and I. Marusic, “Comparison of turbulent channel and pipe flows with varying Reynolds number,” Exp. in Fluids 51, 1261–1281 (2011).
  27. [27]J. Jiménez, S. Hoyas, M. P. Simens, and Y. Mizuno, “Turbulent boundary layers and channels at moderate Reynolds numbers,” J. Fluid Mech. 657, 335–360 (2010).DOI
  28. [28]J. A. Sillero, J. Jiménez, and R. D. Moser, “Two-point statistics for turbulent boundary layers and channels at Reynolds numbers up to δ+ ≈ 2000,” Phys. Fluids 26, 105109 (2014).DOI
  29. [29]J. C. del Álamo and J. Jiménez, “Spectra of very large anisotropic scales in turbulent channels,” Phys. Fluids 15, L41–L44 (2003).DOI
  30. [30]C. E. Willert, J. Soria, M. Stanislas, J. K. Kline, O. Amili, M. Elsfelder, C. Cuvier, G. Bellani, T. Fiorini, and A. Talamelli, “Near-wall statistics of a turbulent pipe flow at shear Reynolds numbers up to 40000,” J. Fluid Mech. 826, R5 (2017).
  31. [31]I. Marusic, J. P. Monty, M. Hultmark, and A. J. Smits, “On the logarithmic region in wall turbulence,” J. Fluid Mech. 716, R3 (2013).DOI
  32. [32]J. Jiménez and S. Hoyas, “Turbulent fluctuations above the buffer layer of wall-bounded flows,” J. Fluid Mech. 611, 215–236 (2008).DOI
  33. [33]P. Bradshaw, “Inactive motions and pressure fluctuations in turbulent boundary layers,” J. Fluid Mech. 30, 241–258 (1967).DOI
  34. [34]J. C. del Álamo and J. Jiménez, “Estimation of turbulent convection velocities and corrections to Taylor’s approximation,” J. Fluid Mech. 640, 5–26 (2009).DOI
  35. [35]A. Lozano-Durán, O. Flores, and J. Jiménez, “The three-dimensional structure of momentum transfer in turbulent channels,” J. Fluid Mech. 694, 100–130 (2012).
  36. [36]R. J. Adrian, C. D. Meinhart, and C. D. Tomkins, “Vortex organization in the outer region of the turbulent boundary layer,” J. Fluid Mech. 422, 1–54 (2000).DOI
  37. [37]C. D. Tomkins and R. J. Adrian, “Spanwise structure and scale growth in turbulent boundary layers,” J. Fluid Mech. 490, 37–74 (2003).DOI
  38. [38]R. J. Adrian, “Hairpin vortex organization in wall turbulence,” Phys. Fluids 19, 041301 (2007).DOI
  39. [39]I. Marusic and G. J. Kunkel, “Streamwise turbulence intensity formulation for flat-plate boundary layers,” Phys. Fluids 15, 2461–2464 (2003).DOI
  40. [40]J. Jiménez and P. Moin, “The minimal flow unit in near-wall turbulence,” J. Fluid Mech. 225, 213–240 (1991).
  41. [41]J. M. FmetzHamilton, J. Kim, and F. Waleffe, “Regeneration mechanisms of near-wall turbulence structures,” J. Fluid Mech. 287, 317–348 (1995).
  42. [42]F. Waleffe, “On a self-sustaining process in shear flows,” Phys. Fluids 9, 883–900 (1997).DOI
  43. [48]I. Marusic, R. Mathis, and N. Hutchins, “Predictive model for wall-bounded turbulent flow,” Science 329, 193–196 (2010).DOI
  44. [49]J. Jiménez, “Cascades in wall-bounded turbulence,” Ann. Rev. Fluid Mech. 44, 27–45 (2012).
  45. [50]P. R. Spalart, “Direct simulation of a turbulent boundary layer up to $Re_\theta = 1410,$ J. Fluid Mech. 187, 61–98 (1988).
  46. [51]M. M. Metzger, B. J. McKeon, and H. Holmes, “The near-neutral atmospheric surface layer: turbulence and non-stationarity,” Phil. Trans R. Soc. A 365, 859–876 (2007).pubmed.ncbi.nlm.nih.gov/17244589
  47. [52]G. K. Batchelor, The theory of homogeneous turbulence (Cambridge U. Press, 1953).

Paper details

vlnc:W_wna3bwxk
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