Statistics of the turbulent kinetic energy dissipation rate and its surrogates in a square cylinder wake flow

A numerical simulation based on the lattice Boltzmann method is carried out in the wake of a square cylinder with the view to investigating possible surrogates for the instantaneous turbulent kinetic energy dissipation rate, (cid:15) , as well as its mean values, (cid:15) . Various surrogate approximations of (cid:15) , based on local isotropy ( (cid:15) iso ), local axisymmetry along x ( (cid:15) a,x ) and y ( (cid:15) a,y ), local homogeneity ( (cid:15) hom ), and homogeneity in the transverse plane, ( (cid:15) 4 x ), are assessed. All the approximations are in agreement with (cid:15) when the distance downstream of the obstacle is larger than about 40 diameters. Closer to the obstacle, the agreement remains reasonable only for (cid:15) a,x , (cid:15) hom and (cid:15) 4 x . The probability density functions (PDF) and joint PDFs of (cid:15) and its surrogates shows that (cid:15) 4 x correlates best with (cid:15) while (cid:15) iso shows the smallest correlation. The results indicate that (cid:15) 4 x is a very good surrogate for (cid:15) and can be used for correctly determining the behaviour of (cid:15) .


I. INTRODUCTION
The mean turbulent kinetic energy dissipation rate is a fundamental quantity in turbulent flow studies. is given by where s ij = 1/2 (u i,j + u j,i ) is the fluctuating strain rate, ν is the kinematic viscosity, u i is the fluctuation part of the velocity in the direction x (i = 1), y (i = 2), z (i = 3) and u i,j ≡ ∂u i /∂x j is the derivative of the velocity component u i along the direction j.Since the groundbreaking work of Kolmogorov 19,20 , it is widely recognized that plays a crucial role in the theoretical study of small-scale turbulence 29 , such as in spectral analysis and two-point spatial analysis.An accurate estimation of is further essential for example in the context of closure models for the turbulent kinetic energy equation.
The estimation of requires the knowledge of 12 terms.While all of these terms can be easily obtained from direct numerical simulation data, only a few of them can be measured from single-point time signals and using the Taylor hypothesis.Generally, it is the isotropic expression iso which is measured to approximate .It is a considerable simplification to Eq.
(1) and is given by iso = 15ν(u 1,1 ) 2 . (2) However, while Eq. ( 2) may be appropriate in many circumstances where local isotropy is approximately satisfied (e.g.homogeneous and isotropic turbulence or HIT, centreline of wakes and jet, at a sufficiently large distance from the wall in pipe, channel and turbulent flows), this is not the case in general when the anisotropy of the flow is not negligible.
For example, Browne et al. 6 , who measured nine components of Eq. ( 1) and inferred the remaining three, showed that iso is not satisfied in the far-field of a circular cylinder wake (CC); iso was about 45% smaller than .Their results are consistent with those of Fabris 14 and Aronson and Lofdahl 5 .
Another simplification can be made for estimating Eq. ( 1) if local homogeneity is satisfied.
Under this condition and using the relation (22) given by Taylor 32 , the locally homogeneous expression for the dissipation rate is where ω 2 is the enstrophy (≡ ω 2 1 + ω 2 2 + ω 2 3 ).In deriving Eq. ( 3) the hypothesis ∇ 2 p = 0 is used since Eq. ( 1) can be written as follows: At this stage, it seems clear that the difference between and hom is related to the influence of pressure fluctuations in the flow under consideration.In the near 25 and far 6 fields of the CC wake, the experimental data show that hom is an accurate surrogate for .From an experimental viewpoint, this is of great interest because only nine velocity gradients need to be measured to obtain ω 2 and thus hom .
Local axisymmetry, which is next to isotropic in order of simplicity but which occurs more commonly in practice, can also be invoked for estimating the energy dissipation rate.
George and Hussein 16 were first to write the expressions for a,x .If local axisymmetry along the axis x 1 is assumed then 17 , Local axisymmetry with respect to the direction y can also be assumed, which leads to 33 , Finally, a fourth surrogate for the turbulent kinetic energy dissipation rate has been proposed by Zhu and Antonia 3,36 .These authors used a vorticity probe comprising 4 Xprobes to measure all the velocity derivatives simultaneously, except (u 2,2 ) 2 and (u 3,3 ) 2 .The incompressibility condition u 1,1 + u 2,2 + u 3,3 = 0 led to the following relation Assuming homogeneity in the transverse plane (y − z), Using Eqs. ( 7) and ( 8) leads to the following expression The above -surrogate should strictly be valid if local homogeneity holds.However, both experimental 31,36 and numerical 10 data reveal its range of validity may be more general.
For example, in the intermediate field, Thiesset 33 showed that iso , a,y and hom are good candidates for estimating .In the far field of a CC wake, Hao et al. 35 showed that, near the wake centerline, iso was only about 10% smaller than , as measured by Browne et al. 6 .This contrasts with the measurements of Browne et al. 6 who, as previously mentioned, found that iso is about 45% smaller that .These inconsistent results highlight the difficulty of measuring reliably.Even though iso can be suited in many situations, this does not necessarily imply that local isotropy holds.Indeed, in grid turbulence, Antonia et al. 4 showed that the velocity derivatives ratios differ from isotropy.However they remarked that there are some compensation effects among the 12 components of which altogether allow the use of iso .Similar observations were in a wake of different obstacles 34 .
Moreover, the intermittent character of the velocity derivatives may play a role in the instantaneous distributions of surrogates for .Indeed, in the original K41 framework 19,20 , the universality of the inertial range was proposed by supposing the latter being uniquely determined by .However, Landau 22 was first to comment that the energy dissipation rate has to be considered as a fluctuating quantity, in the same manner as any other fluctuating quantity.Since the fluctuations of are likely to be affected by the non-homogeneous large scales, the concept of universality may break down even for small scales.This behaviour, usually designated as internal intermittency has led to an explosion in publications over the last fifty years.Attention has been paid mostly to the deviation of scaling exponents to the original K41 predictions.However, the effect of fluctuations in the energy dissipation rate due to the large scales has been given much less attention, even though it is the essence of Landau's original comment.Therefore, not only the mean values but also the instantaneous behaviour of the dissipation rate is a major importance for giving some insight into the intermittent nature of turbulence.
Whilst numerical simulation is the most unambiguous tool for assessing the dissipation rate, the universality of turbulence can be only confirmed or invalidate at very high Reynolds numbers.Such high values of Reynolds numbers remain beyond the actual computing capabilities and can only be approached using experiments with the shortcoming that only few velocity derivatives components are generally captured and that the probe spatial resolution might not be suited.One may thus need to explore the adequacy of the different surrogate of in a range of Reynolds number for which experiments and simulation crossover.The most suited surrogate as inferred from DNS might then be used in experiments at a much higher value of Reynolds number.
The present work aims at assessing the performance of the -surrogates (Eqs.2, 3, 5, 6, and 9) in the wake of a square cylinder (SqC) through direct numerical simulations.The emphasize is on the mean as well as instantaneous values of .The particular choice of the square cylinder is motivated by the fact that this geometry is easier to implement on numerical simulations.In addition, as shown by Thiesset et al. 34 , a SqC generates a more isotropic wake that a CC, a screen, a plate and a screen strip.Accordingly, the SqC wake can be considered as an appropriate test case for above expressions of the -surrogates.One can, for example, expect a better agreement between and iso for the SqC than for the other obstacles.
The lattice Boltzmann method for simulating the wake flow is first described in §II and calculations are validated in §III.In §IV and §V, a particular attention is paid on the mean and root mean square values of and compared to its surrogates for three downstream positions (x = 20, 60 and 100D).Surrogates of are studied in either the centerline and off centerline to appraise the effect of the mean shear.Then, the analysis is extended to the instantaneous values of in §VI, by means of comparative study of the probability density functions (PDF).To further assess the adequacy of using approximates for , joint probability density functions (JPDF) between and its surrogates are employed ( §VII) yielding to a more stringent test.Finally, conclusion are drawn in §VIII.

II. LATTICE BOLTZMANN METHOD
The direct numerical simulation of the SqC wake is carried out using the lattice Boltzmann method (LBM).Rather than solving the governing fluid equations (Navier-Stokes equations), the LBM solves the Boltzmann equation for the probability distribution function on a lattice.The basic idea of the LBM is to construct a simplified kinetic model that incorporates the essential physics of microscopic average properties, which obey the desired (macroscopic) Navier-Stokes equations 15 .With a sufficient amount of symmetry of the lattice, the LBM implicitly resolves these latter equations with second-order accuracy.The Boltzmann equation for f i ( − → x , t) is given by where − → x is the displacement vector, − → e is the particle velocity and Ω is a collision operator approximated by the Bhatnagar-Gross-Krook (LBGK) model where f eq is an equilibrium particle distribution function and τ is the relaxation time.The lattice Boltzmann equation, is then expressed as ∆t is the time step, e i (= ∆x/∆t) is the velocity of the particle i in the i-direction; f eq i given by where ω i is the corresponding weights for each i-direction and the macroscopic quantities ρ and u are the local fluid density and the local fluid velocity vector defined respectively by Using Eq. ( 14), Eq. ( 12) returns the Navier-Stokes equation with a second-order accuracy.
In the collision term Ω i (Eq.( 11)), τ is related to the kinematic viscosity ν by The LBE is discretized on a Cartesian lattice, ∆x = ∆y = ∆z = 1, and ∆t = 1.A detailed description of the LBM can be found in Chen and Doolen (1998) 8 and Succi (2000) 30 .The motivation for the choice of the LBM over the resolution of the Navier-Stokes equations can be briefly summarised as follows: (i) extreme ease of solid surfaces implementation, (ii) no need for solving the Poisson equation for the pressure and (iii), which follows in parts from (ii), the collision is local in nature, making the implementation of parallel computation extremely easy.The method was successfully used for simulating turbulent flows 7,[9][10][11]13 .

III. VALIDATION OF COMPUTING
The three-dimensional computational Cartesian domain (Fig. 1) has 110D × 20D × 10D mesh points in the streamwise (x), across the wake (y), transversal (z) directions and D = 12 (D is the size of one side of the square section of the obstacle).The mesh increments in the three directions are equal, regular with ∆x = ∆y = ∆z = 1.The center of the obstacle is located at 8D from the inlet and 10D from both top and bottom boundaries of the domain.
The downstream distance extends to x/D = 100 when the origin of x is taken at the center of the cylinder.
Periodic conditions are applied in the y and z directions.At the inlet, a uniform velocity (U int = 0.0833 and V int = W int = 0) is imposed whilst a convective boundary condition is applied at the outlet of the domain.A no-slip condition at the wall is simulated with a bounce-back scheme, which means that, at the wall, the particles bounce back along the same direction.The Reynolds number R D = U 0 D/ν is equal to 2000.This value is sufficiently small to allow a relatively good spatial resolution and sufficiently large to be comparable with previous experiments.
In order to avoid the occurrence of instabilities where the magnitude of the local strain rate could be large, mainly around the cylinder, a Large Eddy Simulation scheme (LES) based on the Smargorinsky model is used where ν τ is the turbulent viscosity where C is the Smagorinsky constant (=0.1), the cut-off separation ∆ = ∆x and S = 2S ij S ij is the magnitude of the strain rate tensor (further details can be found in Djenidi, 2006 9 and Djenidi et al., 2013 11 ).
An a posteriori check of the spatial resolution has been made and reveals that the computational mesh resolution which varies from about 3.5η at x/D 1 to 1η at x/D 100 (Fig. 2a) is comparable to many existing DNS 9,18,27 .Fig. 2b compares the present PDF of ( 4x − 4x ) with that obtained in a 3D periodic turbulence box simulation 3 at R λ = 45 using the same approximation for .The relatively good collapse of the two distributions provide confidence in the LBM simulations for capturing the small-scales.Note that this collapse supports the quasi-universality of the small-scale motion.
The flow visualisations (not presented here) showed the classical shedding of the Bénard To further assess the validity of the LBM simulation, we compare some of our results with existing experimental and numerical results obtained in the wake of a circular cylinder.
There are fewer data for a SqC than a CC.Consequently, we carried out a LBM simulation of a CC to perform such comparison.Fig. 4 shows the normalized streamwise distributions of the mean velocity U , the velocity fluctuations, u and v (the prime denotes the rms value) on the wake centerline; the normalisation is based on the length of the recirculation zone behind the cylinder, l f , and the incoming velocity U 0 .The LBM results are consistent with experimental 33 and LES 27 results at similar Reynolds numbers.For example, both the location and magnitude of the minimum value of U are well reproduced by the LBM.
Also, the present LBM values of u are in relatively good agreement with the measurement and LES data.However, there are some discrepancies between the LBM values of v and the experiments.While the LBM predicts correctly the location of the maximum of v , the magnitude is overpredicted.This is yet to be explained, but the differences may be in part attributed to some differences in the initial conditions of both experiments and simulations.
Fig. 5 shows the profiles of U , V , u and v at x/D = 0.54.The LBM data are in good agreement with published data 21,27,33 and capture well all features.
Altogether, the results indicate that the LBM simulation is adequate and reliable for investigating notably the dissipation rate behaviour in the wake of a cylinder.

IV. RESULTS FOR THE MEAN KINETIC ENERGY DISSIPATION RATE
The adequacy of the surrogates for estimating the kinetic energy dissipation rate is first investigated at the level of mean quantities.To this end, the distribution of and its Note first that by comparing these distribution at x = 60D and 100D, one observes that self-similarity is verified satisfactorily.
All distributions collapse well for x/D = 60 and 100 indicating that all the surrogates are adequate at these locations.However, this is no longer the case at x/D = 20 where the collapse excludes iso and a,y .In particular, iso and a,y deviate significantly from the other distributions in the region y/D ≤ 1.It is likely that the constraints imposed by these two expressions on the components of are not satisfied in this region of the wake, not even on the wake centerline.y/y 0.5 y/y 0.5 y/y 0.5 The departure of iso from at 20D is most probably caused by the anisotropy of the flow.To assess this issue, the velocity derivative ratios, K = 2u i,j /u k,l , are examined.Fig. 7, 8 and 9 show all the 12 ratios of the velocity derivatives involved in the expression of for x/D = 20, 60 and 100, respectively; these ratios should be equal to one if isotropy is satisfied.On the wake centerline, the ratios K 1 , K 2 , K 5 , K 10 and K 12 are larger than unity  y/y 0.5 y/y 0.5 2.5 3 y/y 0.5 It is still remarkable that iso approximates at x/D = 60, 100 considering that the flow is not isotropic, as illustrated by the departure of K 1 from one at all locations and across the wake width.Using three-component vorticity measurements, Antonia et al. 4 showed that there is a compensation effect in grid turbulence which explains the equality = iso .These compensations appear to be at play also in the SqC wake flow in agreement with Ref. 34   In Ref. 4, they also noted satisfactory agreement between and 4x , which they attributed to the fact that relation Eq. ( 8) was satisfied.A recent LBM simulation of grid turbulence (Djenidi and Antonia, 2013) confirms their argument.Unpublished DNS data in a turbulent channel flow 1,2 by Abe (Private communication) indicate that Eq. ( 8) is closely satisfied for y/h ≥ 0.1 (y is the distance to the wall and h is the half-width of the channel).y/y 0.5 y/y 0.5 2.5 3 y/y 0.5 shows that Eq. ( 8) is also valid in the present flow, in particular for x/D = 60 and 100.
Clearly, the good collapse observed between and its surrogates in the present flow shows that compensations are likely to be at play here, in particular away from the centerline where the departure from isotropy becomes important.
Although we cannot comment on the behaviour of and its surrogates in the far wake, it is worth briefly recalling previously published results.As stated in the introduction, Browne et al. 6 found that the measured iso departed significantly (about 45% on the wake centerline and about 80% near the wake edge) from obtained by measuring nine of its terms and assuming isotropy to infer the remaining three terms.They argued that this latter assumption is unlikely to be critical as they expected the magnitude of those terms to be small by comparison with the other nine.The reasonable closure of their turbulent kinetic energy budget suggests that they measured with adequate accuracy.However, Hao et al. 35 measured ap = ν(6u 2 ,x + 3u 2 ,y + 2v 2 ,x + 2u ,y v ,x ) (they assumed local isotropy and homogeneity, and the continuity equation) and observed that iso was only about 6% smaller than ap across the wake at x/D = 420).Unfortunately, these authors did not verify whether or not the measured ap closed the turbulent kinetic energy budget.Clearly, the discrepancy between the Browne et al. 6 and the Hao et al. 35 results requires further investigation, preferably via DNS.It should be noted however that whereas Browne et al. estimated the mean squared values of u i,j from 2-points correlations as r → 0, Hao et al. 35 applied spectral corrections based on local isotropy, which may have introduced a systematic bias in their results and led to the close agreement between iso and .

V. STANDARD DEVIATION OF THE ENERGY DISSIPATION RATE
As far as fluctuating quantities are concerned, one can for example investigate the distribution of the standard deviation of across the wake.These are reported in Fig. 11 for x = 20, 60 and 100D.Here again, use was made of appropriate combinations of the velocity defect U d and the half-width y 0.5 to normalize these quantities.
Noticeable is the fact that, independently of x and the surrogate under consideration, the standard deviation of is of same magnitude as its mean value.This indicates that Eq. (1) Eq. ( 2) Eq. ( 5) Eq. ( 6) Eq. ( 3) Eq. ( 9) the dissipation rate is a highly fluctuating quantity, in the same way -if not more -as any other fluctuating quantity, in agreement with Landau's comment.This remarks is of major importance in the context of refining the theoretical description of turbulence that may therefore account for the fluctuating nature of the dissipation rate.
Another remark concerns self-similarity of when normalized by an appropriate set of variables.Indeed, it appears that self-similarity of is attained at the level of the standard deviation at x = 60D, i.e. at the same distance as for the mean value.
A careful analysis of Fig. 11 further reveals that, irrespectively of the downstream position, is relatively well approximated by 4x .Then, the best candidates appear to be the two axisymmetric expressions Eqs. ( 5) and ( 6) followed by the homogeneous surrogate Eq. ( 3) which altogether reveal a maximum departure from the real values of about 40-50%.
Finally, the use of the isotropic expression Eq. ( 2) systematically overestimates the standard deviation of by almost a factor 2.
In summary, whilst the 4x appears to be the best candidate for capturing the fluctuations of , the use of the isotropic surrogate leads to some drastic errors.This is particularly important when internal intermittency corrections which account for the fluctuating nature of the dissipation rate have to be applied.This is a remarkably challenging task since 4x requires measurements using a probe consisting of at least 4 X-wires.
It is worth mentioning that the fact that the standard deviation of 4x is in close agreement with that of does not necessarily imply that their respective probability density functions shape similarly.In the next section, we turn our attention to this issue.

VI. PROBABILITY DENSITY FUNCTION
The results of the preceding section showed that the kinetic energy dissipation rate can be relatively closely approximated by some of its surrogates at the level of mean quantities.
However some departures from the real values of the dissipation rate were observed as regards to the standard deviation of this quantity.This is of significant importance from an experimental point of view, as one can be confident in estimating from a limited set of measurements of velocity derivatives, whilst requires more complex measurements.
However, the results do not give any indigent information on how the instantaneous values of the surrogates compare with .Some of this information can be obtained through the probability density function (PDF), shown in figure 12.The normalization of the PDF is such that ∞ −∞ P α dα = 1, where P α is the PDF of α, and α represents and its surrogates.The figure also reports the PDF of obtained from the DNS data of decaying three-dimensional box turbulence 3 .
All the PDFs present a similar positive tail with a nearly exponential form, where the constant β is estimated to be about 1.25.However, significant differences in the PDFs are visible for α ≤ 0 at all x/D positions.For example, all the surrogate distributions depart from the PDF of at x/D = 20 and 60.Similar results were observed in grid turbulence 4 and a turbulent plane wake 36 (Zhu and Antonia, 1997).This means that the (2) Eq. ( 5) Eq. ( 6) Eq. ( 3) Eq. ( 9) DNS error that one makes using surrogates of will be observed for the most probable values of and are therefore of critical importance.
Of all the surrogate PDFs for P , only P 4x closely follows P ; the agreement between the two is satisfactory at x/D = 100.This indicates that 4x captures well all instantaneous events ranging from weak to high intensity, and explains the good agreement observed earlier results not only, associated with their mean values (figure 6) but also their rms distributions (figure 11).This remark holds irespectively of the transverse distance y suggesting that there is no particular effect of the mean shear on the PDFs.

VII. JOINT PDFS
A more stringent test of the adequacy of the surrogates to estimate instantaneous values of can be further assessed using Joint PDF.Indeed, a joint PDF highlights the probability to observe a particular value of inferred from one particular surrogate knowing the real value of .They are thus likely to be more severe in assessing the adequacy of such surrogates.
The agreement between P and P 4x is well reflected in the JPDFs of and 4x shown in Figure 13, which also shows the JPDFs between and the other surrogates.These JPDFs are obtained on the centerline of the cylinder wake at x/D = 100.A perfect correlation between two variables α 1 and α 2 would be represented by a straight line α 1 = α 2 inclined at 45 0 .Note that although none of JPDFs display a perfect correlation, the JPDF isocontours of and 4x reveal the best alignment along a 45 0 direction highlighting the reasonable agreement between P and P 4x .
Moreover, there is some degree of correlation between and its other surrogates.This can be further assessed from the data of table I, which reports the correlation coefficients between and its surrogates (ρ ,α , α represents any of the surrogates) at x/D = 100 on the centerline and at half-width location.As expected, ρ , 4x is the largest.Antonia et al. 4 measured a coefficient ρ , hom of about 0.8 in grid turbulence.They argued that a larger value would have been expected, had they been able to correct the instantaneous fluctuations for spatial resolution.However, with a value of about 0.61, the present ρ , hom is not as large as that in grid turbulence.This indicates that the correlation between the energy dissipation rate and the enstrophy fields is weaker in the wake flow than in grid turbulence.In other words, this highlights a stronger effect of pressure fluctuations in the inhomogeneous wake flow by comparison to homogeneous grid turbulence.
Nevertheless, the fact that all the coefficients exceed 0.60 may indicate a non-negligible level of correlation between and its surrogates in general, regardless of the lateral position in the wake.
the lattice Boltzmann method.Various expressions for approximating , based on local isotropy ( iso ), local axisymmetry along x ( a,x ) and y ( a,y ), local homogeneity ( hom ), and homogeneity in the transverse plane, ( 4x ), were assessed as surrogates for .
All the approximations show a good collapse with when the distance downstream of the cylinder is larger than about 40 diameters.Even thought iso is suited for such distance, it was shown that compensation between the components of are at play and explains the latter agreement between iso and Closer to the cylinder, the collapse is reasonable across the wake, except for iso and a,y , which depart from in the region 0 ≤ y/D ≤ 1.
The analysis is then extended to the instantaneous values of using PDFs and JPDFs.
It is emphasized that 4x correlates best with whereas iso shows the smallest correlation with .
In summary, results indicate that 4x (Eq.( 9) is an adequate surrogate for and can be used for correctly determining the probabilistic behaviors of .This have a major importance for experimentalists since a reliable estimate of the instantaneous properties of the kinetic energy dissipation rate.

-
von Kármán vortices, with a Strouhal number St = f D/U 0 = 0.129 (f is the shedding frequency) as shown in Fig. 3a.This value of St is remarkably close to the experimental data of Okajima 26 who found St = 0.13 in a wake of SqC at Re = 2000.

FIG. 6 .
FIG.6.Comparison of the mean dissipation rate of the turbulent kinetic energy dissipation across the wake.
while the others are close to one for x/D = 20.At x/D = 60 and 100, except for K 1 all the ratios roughly agree with local isotropy on the centerline.Off the centerline, the ratios are closer to one for x/D = 60 and 100 than for x/D = 20.These results clearly indicate that anisotropy is stronger at x/D = 20 than at the other two locations and explain why iso differs markedly from at x/D = 20.

c) K 9 to K 12 FIG. 8 .
FIG. 8. Distributions across the wake of ratios K 1 to K 12 at x/D = 60 ,(a) K l to K 4 , ; (b) K 5 to K 9 ; (c) K 9 to K 12 .Red dashed line, Isotropic value of 1.

c) K 9 to K 12 FIG. 9 .
FIG. 9. Distributions across the wake of ratios K 1 to K 12 at x/D = 100, (a) K l to K 4 ; (b) K 5 to K 9 ; (c) K 9 to K 12 .Red dashed line, Isotropic value of one.

FIG. 11 .
FIG. 11.Comparison of the root mean square dissipation rate of the turbulent kinetic energy dissipation across the wake.

FIG. 12 .
FIG.12.Comparison of the probability density function of the turbulent kinetic energy dissipation at two different transverse positions y = 0 and y = 0.5(P ( α ) * 100).Black: ; red: iso ; blue: a,x ; light blue: a,y ; green: hom ; purple: 4x ; The dashed line is the PDF of from the DNS 3 of a decaying box turbulence and is shown for reference.