Analysis of the turbulent law of the wall through the finite scale Lyapunov theory
aa r X i v : . [ phy s i c s . f l u - dyn ] N ov Analysis of the turbulent law of the wall through thefinite scale Lyapunov theory
Nicola de Divitiis ”La Sapienza” University, Dipartimento di Ingegneria Meccanica e Aerospaziale, ViaEudossiana, 18, 00184 Rome, Italy
Abstract
This work analyzes the turbulent velocity distribution in proximity of thewall using the finite-scale Lyapunov theory just presented in previous works.This theory is here applied to the steady boundary layer under the hypothesisof moderate pressure gradient and fully developed flow along the streamwisedirection. The analysis gives an equation for the velocities correlation andidentifies the parameters of the expression of the average velocity throughthe statistical properties of the velocity correlation functions. In particular,the von K´arm´an constant, theoretically calculated, is about 0.4, and thedimensionless Prandtl’s length is in function of the Taylor-scale Reynoldsnumber. The study provides the average velocity distribution and gives alsothe variation laws of the other variables, such as Taylor scale and Reynoldsstress. The obtained results show that the finite-scale Lyapunov theory isadequate for studying the turbulence in the proximity of the wall.
Keywords:
Finite-scale Lyapunov theory, von K´arm´an constant, Law of the wall.
1. Introduction
The average velocity law near the wall is well known from many years(von K´arm´an (1930)) for what concerns the flow in smooth pipes (Nikuradse(1932), Reichardt (1940)) and in the cases of turbulent boundary layers withmoderate pressure gradient (Klebanoff (1952), Klebanoff (1955), Smith & Walker(1958)). This law, usually expressed as U + = y + , y + ≤ , (1) Preprint submitted to Elsevier July 18, 2018 + ≃ A ln y + + B, < y + < , (2) U + = 1 k ln y + + C, y + > , (3)gives the average velocity in different regions near the wall, where y + = yU T /ν and U + = U/U T are the dimensionless normal coordinate and averagevelocity, and U T = p ν ( ∂U/∂y ) is the friction velocity.These expressions and the values of A , B , C and k , seem to be univer-sal properties of the flow, which do not depend on the Reynolds number.With reference to Fig. 1, these equations, obtained through considerationsof dimensional analysis and of self-similarity, hold under the hypothesis offully developed parallel flow along the streamwise direction x (von K´arm´an(1930)). Specifically, Eq. (1), being the direct consequence of the wall bound-ary condition and of the definition of U T , expresses the velocity distributionin the laminar sub-layer LL, a domain adjacent to the wall where the effects ofthe viscosity are dominant. Equation (3) holds in the turbulent region TR, azone of non-homogeneous turbulence, where the Taylor scale Reynolds num-ber is high and variable with y . Into Eq. (3), k is the von K´arm´an constantwhich according to von K´arm´an (1930) and in line with Eq. (3), can also beexpressed as k = − lim y + → y + e d U + dy +2 (cid:18) dU + dy + (cid:19) (4)where y + e defines the lower limit of the turbulent region. As far as Eq. (2)is concerned, this describes the velocity variations in the buffer layer BL, anintermediate zone between TR and LL in which viscous and inertia forcesare comparable as order of magnitude. k and A are directly related to the velocity variations along y , whereas C and B express the order of magnitude of U with respect to the friction ve-locity in case of smooth wall. These constants are dimensionless free param-eters which can not be theoretically calculated (Landau & Lifshitz (1959)),therefore several experiments dealing with their determination were carriedout (Fernholz & Finleyt (1996), Zagarola & Smits (1998)), and scaling lawsimilarities were proposed to justify Eq. (3) (Barenblatt (1993)). These2 igure 1: Schematic of the boundary layer. constants can be also identified through the elaboration of the results of di-rect numerical simulations of the Navier-Stokes equations (Spalart (1988),Fernholz & Finleyt (1996) and references therein). From the various sourcesof the literature, k = 0 . ÷ . C = 4 . ÷ . A ≃ ÷ B ≃ − ÷ − h u x u y i− h uv i + = g ( y + ) , h uv i + = h u x u y i /U T (5)This behaves like g ≈ y +3 very near the wall, and at about y + = 5, exhibitsan inflection point whose coordinate provides the order of magnitude of thelaminar sub-layer tickness. The value −h uv i + = 0.5, achieved in the bufferlayer at about y + = 10 ÷
13, corresponds to the maximum turbulent energyproduction due to the mean flow, whereas far from the wall g ≈ k included).The knowledge of the statistical properties of the spanwise correlation func-tion leads to the estimation of the variables at y + e and thus to identify k and the other parameters. Therefore, the von K´arm´an is here theoreticallyobtained, resulting to be about 0.4, in line with experiments and numericalsimulations, and the other results, in agreement with the several data fromthe literature, show that the finite scale Lyapunov theory can be an adequatetool for studying the wall turbulence.
2. Resume
This section summarizes the main results of de Divitiis (IJES 2011) whichregards the steady homogeneous turbulence in the presence of an average ve-locity gradient ∇ x U . There, the author, applying the Finite-scale Lyapunovtheory (de Divitiis (2010)) and the Liouville theorem, proposes the followingevolution equation of the pair distribution function F (2) of fluid velocities in4ase of arbitrary flow ∂F (2) ∂t + ∇ x F (2) · v + ∇ x ′ F (2) · v ′ = λ ( r ) (cid:16) F (2)0 − F (2) (cid:17) − J D (6)where F (2)0 is the pair distribution function of the isotropic turbulence whichexhibits the same momentum and kinetic energy of F (2) , whereas − J D repre-sents the rate of F (2) caused by the rate of the turbulent kinetic energy. λ ( r ) isthe finite-scale Lyapunov exponent associated to the finite-scale r = | x ′ − x | ,and v and v ′ are the fluid velocities calculated at x and x ′ respectively.In case of homogeneous turbulence, de Divitiis (IJES 2011) shows that thesteady distribution function reasonably tends to a quantity which dependsupon F (2)0 , ∇ x U and λF (2) = F (2)0 + 1 λ ( r ) ∂F (2)0 ∂v j ∂U j ∂x p v p + ∂F (2)0 ∂v ′ j ∂U j ∂x p v ′ p ! (7)The velocity correlation tensor R ki = h u i u ′ j i is then calculated, by definition R ki = Z v Z v ′ F (2) u k u ′ i d u d u ′ = R ki − λ (cid:18) ∂U k ∂x p R pi + ∂U i ∂x q R kq (cid:19) (8)where R ki is the second order velocity correlation tensor associated to theisotropic turbulence (von K´arm´an & Howarth (1938), Batchelor (1953)) R ki ( r ) = u (cid:16) ( f − g ) r k r i r + gδ ki (cid:17) (9)being f and g = f + 1 / r ∂f /∂r longitudinal and lateral velocity correlationfunctions, respectively, and u = v − U is the fluctuating velocity. For r = 0,Eq. (8) provides the expression of the Reynolds stresses in function of ∇ x U h u k u i i = u (cid:18) δ ki − (cid:18) ∂U k ∂x i + ∂U i ∂x k (cid:19)(cid:19) (10)Equation (10) is a Boussinesq closure of the Reynolds stress, being u = h u i u i i / h . i indicates the average calculated on the ensamble of the fluidvelocity, and Λ = λ (0) is the maximal Lyapunov exponent. The Lyapunovtheory presented in de Divitiis (2010) shows that Λ = u/λ T , being λ T theTaylor scale. 5ext, the condition of steady flow leads to the following ordinary differ-ential equation for f de Divitiis (IJES 2011) r − f dfd ˆ r + 2 R T (cid:18) d fd ˆ r + 4ˆ r dfd ˆ r (cid:19) + 10 R T f ˆ r p − f ) = 0 (11)whose boundary conditions can be reduced to the following conditions of f in the origin r = 0 f (0) = 1 , df (0) d ˆ r = 0 (12)where ˆ r = r/λ T , d f /d ˆ r (0) ≡ −
1, and R T = uλ T /ν is the Taylor scaleReynolds number. Thus, Eqs. (11) and (12) express an initial conditionproblem whose initial condition is given by Eqs. (12). The same steady con-dition gives the relationship between ∇ x U , Λ and R T , which represents theequation of the kinetic energy in the case of steady homogeneous turbulence S Λ = 15 R T (13)where S is related to ∇ x U S = ∂U i ∂x k (cid:18) ∂U k ∂x i + ∂U i ∂x k (cid:19) (14)The tensor R ik is then calculated with Eq. (8) and (9), by means of f .de Divitiis (IJES 2011) studies some of the properties of Eqs. (11)-(12),and shows that their solutions behave like f − ≈ r in a given interval of r .Accordingly, the statistical moments of velocity difference are h ∆ u n i ≈ r n/ for n <
5, and the energy spectrum E ( κ ) ≈ k − in the inertial subrange.Equation (6) was obtained for an arbitrary flow, whereas the other equa-tions were derived under the assumption of steady homogeneous turbulencewith a given average velocity gradient. The successive sections show thatthese equations can be applied also in the case of non-homogeneous turbu-lence where λ T and u vary with the space coordinates. Thus, the method ishere applied to the turbulent region of the fully developed boundary layer,where the several quantities vary with the wall normal coordinate.6 . Analysis An evolution equation for the velocity correlation is determined, in thecase of non-homogeneous turbulence with a nonzero average velocity gradi-ent.The fluid velocity, measured in the reference frame ℜ , is v = U + u , where U ≡ ( U x , U y , U z ) and u ≡ ( u x , u y , u z ) are, average and fluctuating velocity,respectively. The velocity correlation tensor is R ij = h u i u ′ j i , where u i and u ′ j are the velocity components of u calculated at x and x ′ = x + r , and r isthe separation distance. As the analysis is finalized to the description of theturbulent region of the developed boundary layer, and since there the effectsof the spatial variations of ∇ x U are orders of magnitude much smaller thanthose caused by ∇ x U (von K´arm´an (1930), Nikuradse (1932)), the velocitygradient is assumed to be a function of x alone, being ∇ x U = ∇ x U ′ .In order to determine the evolution equation of R ij , the Navier-Stokesequations are written for the fluctuating velocity in the points x and x ′ .The evolution equation of R ij is determined by multiplying first and secondequation by u ′ j and u i , respectively, summing the so obtained equations, andcalculating the average on the statistical ensemble (von K´arm´an & Howarth(1938), Batchelor (1953)) ∂R ij ∂t = T ij + P ij + 2 ν ∇ R ij − ∂U i ∂x k R kj − ∂U j ∂x k R ik + ∂R ij ∂r k ( U k − U ′ k ) + Γ ij (15)being T ij ( x , r ) = ∂∂r k (cid:10) u i u ′ j ( u k − u ′ k ) (cid:11) , P ij ( x , r ) = 1 ρ ∂ h pu ′ j i ∂r i (16)and p is the fluctuating pressure. The quantities of Eq. (15), which in turndepend on r , due to non-homogeneity depend also on x , and Eq. (15) incudesan additional term with respect to the homogeneous turbulence, representedby Γ ij ( x , r ) (Oberlack (1997))Γ ij ( x , r ) = − ρ ∂ h pu ′ j i ∂x i − ∂∂x k h u i u k u ′ j i − ρ ∂ h p ′ u i i ∂r j + ν ∂ R ij ∂x k ∂x k − ν ∂ R ij ∂r k ∂x k (17)which provides the non-homogeneity of the different terms of correlation.7aking the trace of Eq. (15), we obtain the following scalar equation ∂R∂t = 12 H + 2 ν ∇ R − ∂U i ∂x k R Sik + ∂R∂r k ( U k − U ′ k ) + Γ (18)being R Sik is the symmetric part of R ik and R = R ii , Γ = Γ ii R ( x ,
0) gives the turbulent kinetic energy, H ≡ T ii provides the mecha-nism of energy cascade (von K´arm´an & Howarth (1938), Batchelor (1953)), P ii ≡ R ij , H and U are decomposed intoan even function of r ≡ | r | (here called spherical part), plus the remainingterm: R ij ( x , r ) = ˆ R ij ( x , r ) + ∆ R ij ( x , r ) H ( x , r ) = ˆ H ( x , r ) + ∆ H ( x , r ) U ′ − U = ˆ U ( x , r ) + ∆ U ( x , r ) (20)where ˆ F is the spherical part of the generic quantity F , defined asˆ F ( x , r ) = 16 ( F ( x , r, ,
0) + F ( x , , r,
0) + F ( x , , , r ))+ 16 ( F ( x , − r, ,
0) + F ( x , , − r,
0) + F ( x , , , − r )) (21)and ∆ R ij ( x , )=∆ H ( x , ) = 0. Therefore, the Fourier transform of ˆ R identi-fies the part of the energy spectrum depending upon κ . As the consequenceof this decomposition, R and ∆ R satisfy the equations ∂ ˆ R∂t = ˆ H ν ∂ ˆ R∂r + 2 r ∂ ˆ R∂r ! − ˆ G (22) ∂ ∆ R∂t = ∆ H ν ∇ ∆ R − ∆ G (23)8n which Eq. (23) is obtained as the difference between Eqs. (18) and (22),andˆ G = ∂U i ∂x k ˆ R Ski + ˆ G , ∆ G = ∂U i ∂x k ∆ R Ski − ∆ (cid:18) ∂R∂r k ( U k − U ′ k ) (cid:19) − ∆Γ (24)and ˆ G represents the spherical part of − ∂R/∂r k ( U k − U ′ k ) − Γ. It is worthto remark that Eq. (22) formally coincides with the equation obtainedin de Divitiis (IJES 2011), with the difference that here, because of non-homogeneity, the quantities appearing into Eq. (22) depend also on x .The turbulence is here studied using Eq. (22) alone, whereas R ij ( x , r ) willbe determined in function of ∇ x U , by means of a proper statistical analysisof the two-points velocity correlation.
4. Pair distribution function
This section analyses the non-homogeneous turbulence through the pairdistribution function, taking into account that λ T and h u i u j i vary with thespatial coordinates, whereas ∇ x U is considered to be an assigned quantity.To study this, consider now the pair distribution function of the fluid velocity F (2) ( v , v ′ ; x , x ′ ) = F (2)0 ( v , v ′ ; x , x ′ ) + φ (2) ( v , v ′ ; x , x ′ ) (25)where F (2) obeys to Eq. (6), and φ (2) , representing the deviation fromthe isotropic turbulence, satisfies, at each instant, the following equations(de Divitiis (IJES 2011)) Z v Z v ′ φ (2) du du ′ = 0 , Z v Z v ′ φ (2) u du du ′ = 0 , Z v Z v ′ φ (2) u · u du du ′ = 0 . (26)Equations (26) state that momentum and kinetic energy associated to F (2)0 and F (2) are equal each other, respectively. The analysis supposes also that9ll the dimensionless statistical moments of F (2)0 are constant in all the pointsof the fluid domain, therefore the functional form of F (2)0 is assumed to be F (2)0 ( v , v ′ ; x , x ′ ) = F (2)0 (cid:18) v − U ( x ) u ( x ) , v ′ − U ( x ′ ) u ( x ′ ) (cid:19) (27)and its gradients are calculated in functions of the spatial derivatives of U and u∂F (2)0 ∂x k = − ∂F (2)0 ∂v j (cid:18) ∂U j ∂x k + u j u ∂u∂x k (cid:19) , ∂F (2)0 ∂x ′ k = − ∂F (2)0 ∂v ′ j (cid:18) ∂U j ∂x k + u ′ j u ′ ∂u ′ ∂x ′ k (cid:19) (28)where, as before, ∇ x U = ∇ x ′ U ′ . Substituting Eqs.(25) and (28) into Eq.(6), we obtain the following relationship between F (2)0 and φ (2) λφ (2) = − J D − ∂F (2)0 ∂t + ∂φ (2) ∂t + ∂φ (2) ∂x p v p + ∂φ (2) ∂x ′ p v ′ p ! + ∂F (2)0 ∂v j v p u j u ∂u∂x k ! + ∂F (2)0 ∂v ′ j v ′ p u ′ j u ′ ∂u ′ ∂x ′ k ! + ∂F (2)0 ∂v j v p + ∂F (2)0 ∂v ′ j v ′ p ! ∂U j ∂x p (29)As the last term of Eq. (29) identically satisfies Eqs. (26), these latter arewritten in the form Z v Z v ′ uu · u ( ∂F (2) ∂t + ∂φ (2) ∂x p v p + ∂φ (2) ∂x ′ p v ′ p − ∂F (2)0 ∂v j v p u j u ∂u∂x k − ∂F (2)0 ∂v ′ j v ′ p u ′ j u ′ ∂u ′ ∂x ′ k + J D ) du du ′ ≡ ∂F (2) ∂t + ∂φ (2) ∂x p v p + ∂φ (2) ∂x ′ p v ′ p − ∂F (2)0 ∂v j v p u j u ∂u∂x k − ∂F (2)0 ∂v ′ j v ′ p u ′ j u ′ ∂u ′ ∂x ′ k + J D ≡ F (2) isgiven by Eq. (7) which does not depend on the gradient of u and u ′ , as inthe case of the homogeneous turbulence. Really, F (2) changes starting froman arbitrary initial condition, therefore Eq. (7) represents an approxima-tion which can be considered to be valid far from the initial condition. Asthe consequence, also the spherical part of the correlation tensor, ˆ R ij , hereobtained ˆ R ki = ˆ R Sik = u (cid:18) f + ∂f∂r r (cid:19) (cid:18) δ ki − λ (cid:18) ∂U k ∂x i + ∂U i ∂x k (cid:19)(cid:19) (32)coincides with the expression given in de Divitiis (IJES 2011). Thus, sub-stituting Eq. (32) and into Eq. (22) and following the analytical procedureof de Divitiis (IJES 2011), we found that f obeys to Eq. (11) also in thepresent case, and Λ is related to ∇ x U and R T through Eq. (13). The dif-ference with respect to de Divitiis (IJES 2011) is that, here the turbulenceis non-homogeneous, thus R T = R T ( x ), and f also depends on x , being f = f ( x , r ).
5. Analysis of the boundary layer
Here, the steady turbulent boundary layer with a moderate pressure gra-dient is analyzed, assuming that the flow is fully developed along the stream-wise direction. To this purpose, return to Fig. 1, and consider only the de-veloped region of the flow. In the figure, ℜ is the wall frame of reference, x and y are, respectively, the streamwise direction and the coordinate normalto the wall, whereas z is the spanwise coordinate.In the developed region, u , h u x u y i and λ T change with y , and in thelaminar sublayer U x and u are both about proportional to y . In LL, BL andat the beginning of TR, the correlation scale of velocity is proportional tothe distance from the wall, and vanishes for y = 0, being λ T = (cid:18) ∂λ T ∂y (cid:19) y + ... (33)where ( ∂λ T /∂y ) = O (1), whereas the velocity fluctuations follow the Navier-Stokes equations and satisfy the wall boundary conditions u x = ∂u x ∂y (0) y + ..., u y = 12 ∂ u y ∂y (0) y + ..., u z = ∂u z ∂y (0) y + ... (34)11n case of fully developed flow along x , that is, parallel flow assumption( ∂/∂y >>> ∂/∂x ), the continuity and momentum equations of the meanflow are (Schlichting (2004)) ∂U y ∂y = 0 , (35) ∂∂y h u x u y i + 1 ρ ∂P∂x − ν ∂ U x ∂y = 0 ,∂∂y h u y i + 1 ρ ∂P∂y = 0 (36)where P is the average pressure, whereas the equation of the turbulent kineticenergy reads as ∂∂y (cid:28) u y (cid:18) pρ + u j u j (cid:19)(cid:29) + h u x u y i ∂U x ∂y + ν (cid:28) ∂u j ∂x i ∂u j ∂x i (cid:29) = 0 (37)The non-homogeneity is responsible for the first term of Eq. (37), whereassecond and third terms represent, respectively, the energy production due tothe average motion, and the dissipation.Because of the parallel flow assumption, in all these equations U x , h u i u j i are functions of y alone. From Eq. (35) and taking into account the boundaryconditions, U y ≡
0, whereas Eqs. (36) give the following first integrals P ( x, y ) ρ = F x + H − h u y i (38) h u x u y i − ν dU x dy = − F y − U T (39)being F = 1 /ρ ∂P/∂x and H is a proper constant proportional to the averagepressure at x = 0. Into Eq. (39), the boundary layer approximation and thehypothesis of moderate average pressure gradient along x provide that U T >> | F y | in all the regions, therefore introducing the dimensionless variables U + = U x /U T and y + = yU T /ν , Eq. (39) reads as dU + dy + = 11 + R T (40)12quation (40) is assumed to describe the average flow with moderate pressuregradient in the three regions of the boundary layers, when the parallel flowhypothesis is verified. The Turbulent Region
This section studies the distribution of the different variables in the tur-bulent region TR, by means of the analysis seen in the sections 3 and 4.First, observe that, in case of fully developed parallel flow, the effectsof non-homogeneity in TR are much smaller than those related to energyproduction and dissipation (Tennekes & Lumley (1972)), thus the first termof Eq. (37) is here neglected with respect to the other ones, and Eq. (13)is recovered. This approximation, in agreement with the analysis seen insect. 4, states that the kinetic energy production is balanced only by thedissipation in TR, and allows to express the several variables in terms of thelocal value of R T (or dU x /dy ).In order to calculate u + = u/U T , dU x /dy is eliminated between Eqs. (39)and (13), where S = ( dU x /dy ) u + = R / T / √ R T (41)Being R T = u + λ + T , also λ + T is in terms of R T λ + T = (15 R T ) / p R T (42)As these equations arise from Eq. (13) which holds in the turbulent region,Eqs.(41) and (42) describe the variations of u and λ T only in TR.From the comparison between Eqs. (40) and (4), ( dR T /dy + ) e identifiesthe von K´arm´an constant, here expressed taking into account that R T ≡ λ + T u + k ≡ (cid:18) dR T dy + (cid:19) e = (cid:18) du + dy + (cid:19) e λ + T e + (cid:18) dλ + T dy + (cid:19) e u + e (43)where the subscript e denotes the values calculated at the edge of TR. Substi-tuting Eqs. (41) and (42) into Eq. (43), we obtain the von K´arm´an constantin function of the variables at y + e k = 415 / (cid:18) dλ + T dy + (cid:19) e √ R T e R T e R / T e (44)13t is worth to remark that this estimation of k requires the knowledge of R T e and of ( dλ + T /dy + ) e , whereas does not need the assumption that U + isrepresented by a logarithmic profile.Next, the dimensionless Prandtl’s mixing scale l + p is calculated from thedefinition of Prandtl’s mixing length l p − h u x u y i = (cid:12)(cid:12)(cid:12)(cid:12) ∂U x ∂y (cid:12)(cid:12)(cid:12)(cid:12) ∂U x ∂y l p (45)and Eqs. (10) and (13). This is l + p = p R T (1 + R T ) (46)being l + p ≃ R T for R T >>
1, and its derivative calculated at y + e is alsoexpressed in terms of R T e (cid:18) dl + p dy + (cid:19) e = 1 + 2 R T e p R T e (1 + R T e ) k (47)In view of Eq. (44), this derivative can be also expressed in function of dλ + T /dy (cid:18) dl + p dy + (cid:19) e = 2 1 + 2 R T e R T e (cid:18) R T e (cid:19) / (cid:18) dλ + T dy + (cid:19) e (48)This expression gives the link between the variations of the Taylor scale andof the Prandtl’s length at the border of the turbulent region.Following Eqs. (41) and (42), u + and λ + T are functions of y + throughthe local value of R T (or of dU + /dy + ), therefore, the distribution of suchquantities along y + require the knowledge of the function R T = R T ( y + ).This latter can be expressed as R T ( y + ) = R T e + (cid:18) dR T dy + (cid:19) e (cid:0) y + − y + e (cid:1) + O (cid:0) y + − y + e (cid:1) (49)Now, in a range of y + where O ( y + − y + e ) is negligible with respect to theother terms, dU + /dy + is dU + dy + = 11 + R T e + (cid:18) dR T dy + (cid:19) e (cid:0) y + − y + e (cid:1) (50)14 + exhibits there logarithmic law, obtained integrating Eq. (50) from y + e to y + U + = 1 k ln (cid:18) R T e + k ( y + − y + e )1 + R T e (cid:19) + U + e (51)being U + e = U + ( y + e ). This law is defined as soon as the parameters y + e , k and U + e are known. Equation (51) differs from the classical expression (3)and formally tends to Eq. (3) when y + → ∞ . Therefore, Eq. (51) can givevalues of U + sizably different from (3) for small y + . The comparison betweenthese equations, for y + → ∞ identifies C in terms of the variables at y e C = U + e + 1 k ln (cid:18) k R T e (cid:19) (52)As far as the Reynolds stress is concerned, it is expressed in function of R T through Eqs. (39) and (40) h uv i + ≡ h u x u y i U T = − R T R T (53)Observe that, the wall boundary conditions (34) state that h uv i + ≈ y +3 nearthe wall, whereas Eq. (53) gives h uv i + ≈ R T ≈ y +2 . This disagreementis due to the fact that the expression of h uv i + has not been derived fromthe correlation equation with r = 0, but arises from Eq. (39) eliminating dU x /dy . This implies that Eq. (53) holds only in TR and BL, whereas inthe laminar sub-layer a proper matching condition must be applied. Matching Turbulent region - buffer layer
With reference to Fig. 1, the domains LL and BL constitute SL, a zonebetween wall and turbulent region. There, due to the presence of the wall,the analysis of sections 4 and 3 can not be applied. Therefore, the meanvariables in SL are expressed in function of y + , taking into account theboundary conditions (34) and that λ T follows Eq. (33), where it is assumed( dλ + T /dy + ) e = ( dλ + T /dy + ) . As the consequence, R T , u + and λ + T are supposed15o vary in SL according to R T = u + λ + T ,u + = u + e y + + C u y +2 y + e + C u y + e , (0 < y + ≤ y + e ) λ + T = λ + T e y + y + e , (0 < y + ≤ y + e ) , (54)where C u is a constant given by C u = 1 y + e R T e − ky + e ky + e − R T e (55)Equations (54) and (55) provide the matching condition between SL and TR.Specifically, Eqs. (54) state that u + , R T and λ + T are continuous functions for y + = y + e , whereas Eq. (55) gives there the continuity of their derivatives. Theaverage velocity is then calculated by quadrature, substituting the expressionof R T given by Eqs. (54), into Eq. (40), integrating this latter from 0 to y + ,with U + (0) = 0 U + ( y + ) = Z y +
11 + R T ( η + ) dη + (56)where U + ( y + ) ≃ y + for small y + . The matching between SL and TR givesthe value of U + e . U + e = Z y + e
11 + R T ( η + ) dη + (57)Thus, Eq. (56) and (51) establish that both U + and dU + /dy + are continuousfor y + = y + e Matching buffer layer - laminar sub-layer
As previously seen, the expression (53) of the Reynolds stress, valid inBL and TR, can not be applied in the laminar region. Since h uv i + ≈ y +3 near the wall, the Reynolds stress in LL is approximated by h uv i + = h uv i + ∗ y +3 + C uv y +4 y + ∗ + C uv y + ∗ , (0 < y + < y + ∗ ) (58)16eing C uv a constant C uv = 1 y + ∗ − ( d ln h uv i + /dy + ) ∗ y + ∗ ( d ln h uv i + /dy + ) ∗ y + ∗ − y + = y + ∗ , where the subscript ∗ indicates the valuecalculated at y + ∗ . This latter, obtained as the inflection point of h uv i + (i.e. d h uv i + /dy +2 = 0) according to Eqs. (53) and (54), gives the dimensionlesstickness of the laminar sub-layer.
6. Identification of the velocity law free parameters
The definition of the velocity law, requires the knowledge of the parame-ters which appear into Eqs. (44) and (51). In particular, the determinationof k needs the values of R T e and ( dλ + T /dy + ) e . To identify these latter, wewill proceed as follow.First, observe that h u i u j i is a symmetric tensor which can be obtainedfrom the diagonal tensor h u u i = h u i h v i
00 0 h w i (60)through an opportune rotation around y ≡ y , being h u i , h v i and h w i thevelocity components standard deviations in the canonical frame, and ϑ is theangle of this rotation. Therefore, u and h u x u y i are in terms of the elementsof h u u i u = 13 ( h u i + h v i + h w i ) h u x u y i = ( h u i − h v i ) sin 2 ϑ u and h u x u y i are related each other in such a way that u ≥ |h u x u y i| (62)17ence, Eq. (13) implies that R T ≥
15 or dU x dy λ T ≥ u (63)in the turbulent region. This limitation identifies the minimum value of R T in TR, which is assumed to be R T e = 15 . (64)To estimate ( dλ + T /dy + ) e , consider first the spanwise correlation function ofthe streamwise velocity components (that is R ( r z ) = h u x ( x, y, z ) u x ( x, y, z + r z ) i ). This can be calculated with Eq. (8), once f is known through Eqs.(11). Since ∂U/∂y leads to the development of coherent structures in thefluid, similar to streaks and caused by the stretching of the vortex linesalong the streamwise direction (Kim et al. (1990)), we expect that R ( r z )intersects the horizontal axis and remains negative for r z → ∞ . This im-plies a wide distribution of spacings between the different streaky structures,whose mean value depends on R T (Kim et al. (1990)). Accordingly, thereexists a minimum spanwise distance r ∗ z such that, for r z ∈ [0 , r ∗ z ], R ( r z )gives the necessary informations to describe the statistical properties of thecorrelation (Ventsel (1973)). According to the theory (Ventsel (1973)), if R monotonically tends to zero as r z → ∞ and R < r ∗ z can be estimated asthe distance at which R is negative and exhibits an inflection point (secondinflection point of the curve, see Fig. 2) R ( r ∗ z ) < , ∂ R ∂r z ( r ∗ z ) = 0 , Condition 1 (65)Alternatively, r ∗ z can be estimated as the intersection between the osculatingparabola in the point ¯ r z where R = min, and the horizontal axis, that is R (¯ r z ) + 12 ∂ R ∂r z (¯ r z )( r ∗ z − ¯ r z ) = 0 , Condition 2 (66)Since R ( r z ) is expressed in function of f through Eq. (8), this is relatedto the correlation functions associated to the other directions, thus r ∗ z isrepresentative also for the other coordinates x and y . As the result, thedistance from the wall of a point of TR must be always greater than r ∗ z (i.e.18 igure 2: Schematic of the spanwise correlation function of u x at the edge of the turbulentregion and definition of Eqs. (65) and (66). r ∗ z < y ). Hence, it is reasonable to assume that, at the edge of the turbulentdomain y e = r ∗ z R T e = 15 (67)In order to calculate R , f is first obtained by solving Eqs. (11)-(12)which correspond to the following Cauchy’s initial condition problem (de Divitiis(IJES 2011)) dfd ˆ r = FdFd ˆ r = − f ˆ r p − f ) − r − f R T + 4ˆ r ! Ff (0) = 1 , F (0) = 0 (68)19ence, R ( r z ) is obtained with Eq. (8), and r ∗ z is calculated for both theconditions (65) and (66). As the result, y + e is given by y e λ T e = r ∗ z λ T e where R T = R T e ≡
15 (69)and ( dλ + T /dy + ) e is determined according to Eqs. (54) (cid:18) dλ + T dy + (cid:19) e = λ + T e y + e (70)
7. Results and Discussion
As the calculation of k and dλ + T /dy + needs the knowledge of the sta-tistical properties of the velocity correlation, R is calculated for differentvalues of R T . To this purpose, f was first determined by solving numericallyEqs. (68), by means of the fourth-order Runge-Kutta method. The calcula-tion was carried out for R T = 15, 20, 40, 60 and 80, where these Reynoldsnumbers correspond to several distances from the wall in the turbulent re-gion. To obtain a good accuracy of the solutions, the step of integrationis chosen to be equal to 1/40 of the estimated Kolmogorov scale l K , where l K /λ T = 1 / / / √ R T (Batchelor (1953)). The results are given in Fig. 3 (a)which shows f in terms of r , where the bold line represents the correlationfunction at the edge of TR ( R T = 15). This latter exhibits the ratio (integralscale)/(Taylor scale) quite similar to that of a gaussian centered in the origin,whereas in the other cases, this ratio increases with R T , in agreement withthe analysis of de Divitiis (IJES 2011). A more detailed analysis about thechanging of f and of the corresponding energy spectrum E ( κ ) with R T isreported in de Divitiis (IJES 2011).The spanwise correlation function R ( r z ) is then calculated with Eq. (8),and is represented in Fig. 3 (b) for the same values of R T . From these data,the edge of TR, y + e ≡ r ∗ z , is calculated for both the conditions 1 and 2.The von K´arm´an constant and all the others parameters are shown inthe table 1. The table reports also the values of the free parameters A , B and C associated to the classical wall laws (1)-(3), which are here calculatedthrough the identification with Eqs. (51) and (56). In particular, C is cal-culated with Eq. (52), whereas A , which pertains the velocity law in the20 igure 3: (a) Longitudinal correlation function associated to ˆ R ij . (b) spanwise correlationfunction of u x , for different Taylor scale Reynolds number. The bold lines are calculatedfor R T = 15. buffer layer, is identified as the slope dU + ( y + ) /d ln y + of Eq. (56) where d U + ( y + ) /d ln y +2 = 0, and B is consequentely determined. The values of k and C are compared with the results given by different authors.For both the conditions, y + e ≈
50 and U + e ≈
15 and this correspondsto a difference with respect to the classical data of Nikuradse (1932) andReichardt (1940), which is less than 2 %. As far as the Prandtl’s length isconcerned, according to Eq. (46), it varies quite similarly to R T , and itsderivative, calculated for y + = y + e , is slightly greater than k (see Eq. (47)) (cid:18) dl + p dy + (cid:19) e = 1 . ... (cid:18) dλ + T dy + (cid:19) e ≃ . k (71)Its specific values, shown in the table, are in excellent agreement with theclassical results. For what concerns A , B and C , their values agree quite wellthe experiments, expecially for what concerns C and A .Once the free parameters are identified, R T is calculated in function of y + through Eq. (49) with O ( y + − y + e ) → U + . Figure 5 shows U + ( y + ) calculated with Eq. (51) and (56).It is apparent that the results agree very well with the experimental dataof different authors (Nikuradse (1932), Klebanoff (1952), Reichardt (1940))and with the formulas proposed by Spalding (1961) and Musker (1979), withan error which does not never exceed 5 % in the interval of y + of the figure.21 igure 4: Taylor scale Reynolds number in terms of y + : SL dashed lines, TR continuouslines In particular, the condition 1 (Eq. (65)) provides a maximum difference be-tween present results and the data of Nikuradse and Reichardt, less than 3% in this interval.The other variables are represented in Fig. 6. The dimensionless Taylorscale (Fig. 6 a), linear in the laminar sub-layer and in the buffer region,remains a rising function of y + in the turbulent zone and exhibits, there, aslope which decreases slightly with y + , whereas the Lyapunov exponent (Fig.6 b), defined only in TR, is represented by a monotonically decreasing func-tion of y + . This latter, being proportional to the square root of the turbulentdissipation rate (Λ = u/λ T ), agrees with the data of the different experiments(Fernholz & Finleyt (1996)), at least where U + exhibits logarithmic profile.The square root of the kinetic energy and the Reynolds stress are shownin Fig. 6 (c) and (d). These vary in TR following Eqs. (41)-(53) and aremonotonically rising functions of y + . As seen, u ≈ y + and h uv i + ≈ y +3 in thelaminar sub-layer, whereas the Reynolds stress calculated with Eq. (53) inthe buffer region, shows an inflection point y + ∗ which represents the separationelement between LL and BL. For y + > y + ∗ , −h uv i + rises with y + until to reachthe turbulent region where is about constant and equal to the unity. More22 igure 5: Dimensionless average velocity profile: SL dashed lines, TR continuous lines in detail, Fig. 7 shows −h uv i + in an enlarged region which includes LL andpart of BL. The distance y + ∗ , represented by F and F is y + ∗ = 5.734 and6.097, in line with the order of magnitude of the laminar sub-layer tickness,and this is achieved at about −h uv i + = 0.24 in the two cases. Next, the value −h uv i + = 0.5 is obtained in the buffer layer for y + ≃ h uv i + , h uv i + dU + /dy + and of the corresponding y + , are in excellent agreement with the data of Tennekes & Lumley (1972)and Hinze (1975).It is worth to remark that, although the monotonic trend of u and h uv i + can contrast some experiments which can give non-monotonic variations ofthese variables, the values of such quantities and the corresponding y + ,are comparable with those of the several experiments (Fernholz & Finleyt(1996)). This discrepancy can be due the fact that, here, the effects of23 igure 6: Distribution of the dimensionless variables in the boundary layer: SL dashedlines, TR continuous lines. (a) Taylor scale. (b) Maximal Lyapunov exponent. (c) r.m.s.of fluctuating velocity. (d) Reynolds stress. non-homogeneity of h u y ( p/ρ + u j u j / i on the kinetic energy equation areneglected in TR, and that the average velocity is analyzed only in the loga-rithmic range.
8. Conclusions
This work analyzes the turbulent wall laws through the Lyapunov theoryof finite scale. The results, valid for fully developed flow with moderate pres-sure gradient, are subjected to the hypothesis that in the turbulent region,24 igure 7: Dimensionless Reynolds stress in the laminar sub-layer and in the buffer region the energy production due to the average flow is balanced only by the dissi-pation rate, whereas the non-homogeneity of h u y ( p/ρ + u j u j / i is neglected.The free parameters of the velocity law, here theoretically calculatedthrough the statistical properties of the velocity correlation functions, andthe wall laws, are in very good agreement with the literature. In particular: • The von K´arm´an constant, theoretically identified as k = ( dR T /dy + ) e ≈ . ÷ .
42, depends on the scale of the spanwise velocity correlation anddoes not requires the assumption of the logarithmic velocity profile. • The average velocity law and the distributions of the other dimen-sionless quantities such as kinetic energy and Reynolds stress in theboundary layer, agree -or at least are comparable- with experimentsand direct simulations.These results, which represent a further application of the analysis pre-sented in de Divitiis (IJES 2011), show that the finite scale Lyapunov anal-ysis can be an adequate theory to explain the wall turbulence.25 igure 8: Dimensionless rate of kinetic energy production in the laminar sub-layer and inthe buffer region
References von K´arm´an, T. , Mechanical similitude and turbulence, NACA-TM-611,1931 (1930).
Nikuradse, J. , Laws of Turbulent Flow in Smooth Pipes, NASA TT F-10,359, 1966 (1932).
Reichardt, H. , Heat Transfer Through Turbulent Friction Layers, NACA-TM-1047, 1943 (1940).
Klebanoff, P. S., Diehl, Z. W. , Some features of artificially thickenedfully developed turbulent boundary layers with zero pressure gradient,NACA-TR-1110, 1952.
Klebanoff, P. S. , Characteristics of turbulence in boundary layer withzero pressure gradient, NACA-TR-1247, 1955.
Smith, D. W., Walker, J. H. , Skin-friction measurements in incompress-ible flow, NACA-TN-4231, 1958. 26 palart, P. R. , Direct simulation of a turbulent boundary layer up to R ϑ = 1410, Journal of Fluid Mechanics , , pp. 61-98, 1988 Fernholz, H. H., Finleyt, P. J. , The incompressible zero-pressure-gradient turbulent boundary layer: An assessment of the data,
Progressin Aerospace Sciences , , pp. 245–311, 1996 DOI: 10.1016/0376-0421(95)00007-0 Landau, L. D., Lifshitz, M. , Fluid Mechanics . Pergamon London, Eng-land, 1959.
Barenblatt, G. I. , Scaling laws for fully developed turbulent shear flows.Part 1. Basic hypotheses and analysis.,
Journal of Fluid Mechanics , ,pp. 513–520, 1993 DOI: 10.1017/S0022112093000874 Zagarola, M. V., Smits, A. J. , Mean-flow scaling of turbulent pipe flow.,
Journal of Fluid Mechanics , , pp. 33–79, 1998 de Divitiis, N. , Lyapunov Analysis for Fully Developed HomogeneousIsotropic Turbulence, Theoretical and Computational Fluid Dynamics ,DOI: 10.1007/s00162-010-0211-9. de Divitiis, N. , Self-Similarity in Fully Developed Homogeneous IsotropicTurbulence Using the Lyapunov Analysis,
Theoretical and ComputationalFluid Dynamics , DOI: 10.1007/s00162-010-0213-7. de Divitiis, N. , Steady Homogeneous Turbulence in the Presence of anAverage Velocity Gradient,
International Journal of Engineering Science ,DOI: 10.1016/j.ijengsci.2011.10.013 von K´arm´an, T. & Howarth, L. , On the Statistical Theory of IsotropicTurbulence.,
Proc. Roy. Soc. A, , 14, 192, 1938. Batchelor, G. K. , The Theory of Homogeneous Turbulence . CambridgeUniversity Press, Cambridge, 1953.
Oberlack, M. , Non-isotropic dissipation in non-homogeneous turbulence,
Journal of Fluid Mechanics , pp. 351–374, 1997.
Schlichting, H., Gersten, K. , Boundary-Layer Theory . 8th ed. Springer-Verlag 2004. 27 ennekes, H., Lumley, J. L. , A first course in turbulence . MIT Press,1972
Hinze, J. O. , Turbulence . McGraw-Hill, 1975
Lee, M. J., Kim, J., and Moin, P. , Structure of turbulence at high shearrate,
J. Fluid Mech. , , pp. 561–583, 1990. Ventsel, E. S. , Theorie des probabilites . Ed. Mir, CCCP, Moskow, 1973.
Spalding, D. B. , A Single Formula for the Law of the Wall,
J. AppliedTransaction of the ASME , Series E , pp. 455, 1961 Musker, A. J. , Explicit Expression for the smooth wall velocity distributionin a turbulent boundary layer,
AIAA Journal , , pp. 665-657, 197928 able 1: Velocity law parameters and comparison of the results. P.R. as for ”PresentResult”. Parameter P. R. P. R.Condition 1 Condition 2 Nikuradse Zagarola Barenblatt Smith FernholzEq. (65) Eq. (66) k C A B -4.1226 -4.7166 r ∗ z λ T e (cid:18) dλ + T dy + (cid:19) e (cid:18) dl + p dy + (cid:19) e y + e u + e −h uv i + e U + ee