THE STRUCTURE OF TURBULENT SHEAR FLOW BY A, A. TOWNSEND, F.R.S. Reader in Experimental Fluid Mechanics University of Cambridge SECOND EDITION CAMBRIDGE UNIVERSITY PRESS CAMBRIDGE LONDON NEW YORK NEW ROCHELLE MELBOURNE SYDNEYPublished by the Press Syndicate of the University of Cambridge The Pitt Building, Trumpington Street, Cambridge CB2 {RP 32 East 57th Street, New York, NY 10022, USA 296 Beaconsfield Parade, Middle Park, Melbourne 3206, Australia © Cambridge University Press 1976 Library of Congress Catalogue Card Number: 74-14441 ISBN 0 521 20710 X hard covers ISBN 0 521 29819 9 paperback First published 1956 Second edition 1976 First paperback edition 1980 ‘Transferred to digital printing 19991a 1.2 1.3 14 1.5 16 17 18 21 2.2 23 24 25 2.6 27 3A 3.2 33 34 35 3.6 3.7 CONTENTS Preface page xi 1 THE STATISTICAL DESCRIPTION OF TURBULENT FLOW Introduction The development of a theory for turbulent flow The statistical description of turbulent flow Notation for turbulent flows Three-dimensional correlation and spectrum functions One-dimensional correlation and spectrum functions Correlations and spectra with time delay Homogeneity and symmetry of turbulent flows 2 THE EQUATIONS OF MOTION FOR TURBULENT FLOW Assumption of a continuous fluid The equations of fluid motion Approximate forms of the equations of motion Mean value equations for momentum, energy and heat Energy dissipation by viscosity Conductive dissipation of temperature fluctuations The relation between the pressure and velocity fields 3 HOMOGENEOUS TURBULENT FLOWS Introduction Eddy interactions in homogeneous turbulence Experimental approximations to homogeneous turbulence Isotropic turbulence: general Reynolds number similarity in isotropic turbulence Self-preserving development in isotropic turbulence Space-time correlations in isotropic turbulence Rawew- 28 32 33 36 38 al a3 43 45 49 st 53 59 62vi CONTENTS 3.8. The Taylor approximation of frozen flow 3.9 The tendency to isotropy of homogeneous turbulence 3.10 Uniform distortion of homogeneous turbulence 3.11 Irrotational distortion of grid turbulence 3.12 Unidirectional, plane shearing of homogencous turbulence 3.13 Local isotropy and equilibrium of small eddies 3.14 Measurement of spectrum and structure functions 3.15 Energy transfer in the inertial subrange 3.16 The equilibrium spectrum in the viscous subrange 3.17 Local isotropy in non-Newtonian fluids 4 INHOMOGENEOUS SHEAR FLOW 4.1 Large eddies and the main turbulent motion 4.2 Structural similarity of the main turbulent motion 4.3 Nature of the main turbulent motion 4.4 Generation and maintenance of the main motion 4.5 Flow inhomogeneity and the large eddies 4.6 The dependence of Reynolds stress on mean velocity 4.7 Statistical distributions of velocity fluctuations 5 TURBULENT FLOW IN PIPES AND CHANNELS 5.1. Introduction 5.2 Equations of motion for unidirectional mean flow 5.3. Reynolds number similarity in pipe and channel flow 5.4 Wall similarity in the region of constant stress 5.5. Flow over rough walls 5.6 Mean flow in the central region 5.7 The turbulent motion in constant-stress equilibrium layers 5.8 Eddy structure in equilibrium layers 5.9 Motion in the viscous layer next the wall 5.10 Fluctuations of pressure and shear stress on a wall 5.11 The magnitude of the Karmén constant 5.12 Turbulent flow and flow constants 66 m 4% 80 88 93 99 103 105 106 118 120 122 124 126 130 131 133 135 140 145 150 156 163 “165 168 169CONTENTS 5.13 Similarity flows in channels and pipes of varying widths 5.14 Equilibrium layers with variable stress 8.15 Equilibrium layers with linear distributions of stress 5.16 Equilibrium layers with surface transpiration 5.17 Equilibrium layers with variable direction of flow 6 FREE TURBULENT SHEAR FLOWS 6.1 General properties of free turbulence 6.2 Equations of motion: the boundary-layer approximation. 6.3 Integral constraints on free turbulent flows 6.4 Self-preserving development of free turbulent flows 6.5 The distributions of mean velocity and Reynolds stress 6.6 The balance of turbulent kinetic energy 6.7 The bounding surface of free turbulent flows 6.8 Distributions of turbulent intensity and Reynolds stress 6.9 Flow constants for self-preserving jets and wakes 6.10 The flow constants of plane mixing layers 6.11 The entrainment of ambient fluid 6.12 Basic entrainment processes 6.13 Entrainment eddies in plane wakes 6.14 Mechanism of the entrainment eddies 6.15 Control of the entrainment rate 6.16 Fluctuations outside the turbulent flow: sound radiation 6.17 Irrotational fluctuations in the near field 6.18 Development of nearly self-preserving flows 6.19 Development of a jet in a moving stream of constant velocity 7 BOUNDARY LAYERS AND WALL JETS 7.1. Wall layers in general 7.2. Self-preserving development of wall layers 7.3. General properties of self-preserving wall layers 7.4 Flow parameters of self-preserving wall layers 7.5 Development of self-preserving wall jets 7.6 Development of self-preserving boundary layers 472 176 180 184 186 188 188 193 195 201 205 214 220 227 230 232 241 243 247 248 251 252 255 259 262 263 266 268 2mviii CONTENTS 7.7 Boundary-layer development with zero wall stress 7.8 Wall layers with convergent flow 7.9. Almost self-preserving development 7.10 Layers with nearly uniform velocity in the free stream 7.11 Turbulent flow in self-preserving boundary layers 7.12 Development of boundary layers in arbitrary external conditions 7.13 Boundary-layer development after a sudden change of external conditions 7.14 Development in a region of strong adverse pressure gradient 7.15 Layer development after a sudden change of roughness 7.16 Boundary layers with three-dimensional mean flow 7.17 Three-dimensional flow with negligible Reynolds stresses 7.18 Homogencous three-dimensional flow — the Ekman layer 7.19 Secondary flow in a boundary layer with a free edge 7.20 Lateral variations of stress in boundary layers 7.21 Periodic structure of flow near the viscous layer 8 TURBULENT CONVECTION OF HEAT AND PASSIVE CONTAMINANTS 8.1 Governing equations and dimensional considerations 8.2. Diffusion by continuous movements: effect of molecular diffusive transport 8.3 Eulerian description of convective flows: mean valuc equations and correlation functions 8.4 Local forms of the Richardson number 8.5. Spectrum functions and local similarity 8.6 Scattering of light by density fluctuations in a turbulent flow 8.7 Self-preserving development of temperature fields in forced convection flows 8.8 Forced convection in wall flows 8.9 Rates of heat transfer in forced convection 8.10 Convection in a constant-stress layer after an abrupt change in wall flux or temperature 8.11 Longitudinal diffusion in pipe flow 276 280 283 287 289 294 298 301 307 312 316 318 323 328 331 334 336 338 MI 342 348 350 352 356 361 364CONTENTS ix 8.12 Natural convection and energy transfer 366 8.13 Buoyant plumes and thermals 366 8.14 The effect of buoyancy forces on turbulent motion 372 8.15 Horizontal wall layers with heat transport 375 8.16 Nature of turbulence in strongly stable flows 378 8.17 Transient behaviour of boundary layers with heat transfer 379 8.18 Convective turbulence 380 8.19 Heat convection between horizontal, parallel planes 381 8.20 Heat transfer in Benard convection 384 8.21 Similarity and structure of Benard convection 386 8.22 Natural convection in wall layers 390 9 TURBULENT FLOW WITH CURVATURE OF THE MEAN VELOCITY STREAMLINES 9.1 Mean value equations for curved flow: the analogy between the effects of flow curvature and density stratification 393 9.2. Couette flow between rotating cylinders 398 9.3 Flow with the outer cylinder stationary 400 9.4 Turbulent motion with the outer cylinder stationary 404 9.5. Flow with the outer cylinder rotating 407 References 413 Index 425PREFACE Since my original monograph was published, so much new material has appeared that I am amazed to read my statement that (around 1950) ‘experimental knowledge...was being accumulated very rapidly’. In the last ten years, papers have been appearing at sucha rate that I must extend apologies to all those whose work I have either not read, ignored, or used without realising the origin. In spite of a considerable increase in size and the inclusion of sections on convection, the same approach is followed - to develop a con- sistent view of the nature of turbulence from observations of simple flows and then to use it to interpret and predict the behaviour of a variety of flows of more general interest. Although my current views have been developed from those I held in 1956, they have undergone considerable change. Perhaps I should thank especially Dr H. L. Grant who, as my research student, began the process by demolishing a complete chapter before the ink was wholly dry. No doubt, I shall trouble some readers by my habit of omitting the density from most equations and by changing to the meteor- ological practice of using ‘z’ to denote displacement in the direction of shear. I offer them my sympathy but not my repentance. My thanks are offered to the Cambridge University Press for their incredible patience, to many friends whose cries of ‘When will it appear?’ have flattered me into continued activity, and to Pro- fessor G. K. Batchelor for a line of tactful harassment. At the moment, I am more pleased than even my wife to have completed the writing,CHAPTER THE STATISTICAL DESCRIPTION OF TURBULENT FLOW 1.1 Introduction One of the few phenomena in the field of fluid motion that find their way into physics text-books is the existence of a critical velocity for the flow of a viscous fluid through a circular pipe. The critical velocity separates a regime of steady laminar flow from a regime of highly irregular turbulent flow in which the flow resistance is considerably greater than is indicated by the Poiseuille equation. The difference between the two kinds of flow can be seen if a filament of dye is injected near the centre-line of the pipe. In laminar flow, the filament remains straight and coherent but, with the onset of turbulent flow, it meanders, winds itself up into tight coils and is diffused rapidly over the whole section of the pipe. Although the transition from laminar to turbulent flow is not as simple as this and similar de- scriptions make it appear, the phenomenon illustrates very well the fundamental differences in character between laminar and turbulent flow, particularly the ability of a turbulent flow to transmit larger shear stresses and to diffuse heat and matter more rapidly than the corresponding laminar flow. It is well known that the differences arise from an intricate and eddying motion of the fluid which con- vects momentum, heat and matter from one part of the flow to another, the direction of net transport being in general down the gradient of the quantity concerned, Formally, the overall effect is equivalent to increasing greatly the effective coefficients of viscosity, heat conductivity and diffusion, and it is natural to draw an analogy between the turbulent motion and the molecular motion that is responsible for transport phenomena in gases. A similarity does exist but the analogy is imperfect in two important respects. First, at any moment the motion of a gas molecule is affecting the motion of at most one other molecule and mixing on the microscopic scale takes place freely. Turbulent diffusive movements of fluid particles2 TURBULENT SHEAR FLOW are essentially part of the general motion of the fluid and the direct effect is to mingle rather than to mix parcels of fluid from different parts of the flow. Complete mixing depends on molecular diffusion which is intensified by increases of concentration gradient caused by mingling. Secondly, the turbulent motion is retarded by viscous stresses and requires a continuous supply of energy to maintain it, obtained from the working of the mean flow against the turbulent Reynolds stresses. The turbulent motion so depends for its kinetic energy on one of the quantities that it diffuses, the momentum of the mean flow, and the diffusing processes cannot be considered as small perturbations of an already existing motion as in the kinetic theory of gases. The necessary connection between the diffusion and the supply of energy to the turbulent motion is a fundamental characteristic of turbulent flow. As a consequence of the irregularity and complexity of the motion, it is only practicable to consider mean values of functions of the instantaneous and local values of the fluid velocities and pressures, and all theoretical and experimental work uses mean values. For many years after Osborne Reynolds’s formulation of the problem, measurements were confined to mean values of velocity and temperature and few measurements of the fluctuations were made. The theories developed at this time used speculative models of the fluid motion, sometimes derived by considering properties of the equations of fluid motion and sometimes by dimensional reasoning and analogies with the kinetic theory of gases. The most important were the various forms of the mixing-length theory developed by L. Prandtl and by G. I. Taylor, which served a purpose in providing a framework for current theoretical and experimental work, but they were admittedly incomplete and contained internal inconsistencies. About forty years ago, H. L. Dryden, A. M. Kuethe and others developed the hot-wire anemometer, making available for the first time a convenient means of studying the fluctuations of velocity, and its exploitation has provided detailed knowledge of the turbulent motion in a variety of flows. With so much information, it should now be possible to devise a physical theory of turbulence, based ondynamically consistentassumptions about the motionandcap- able of describing with fair accuracy the structure and properties of the simpler flows. The following account is an attempt in that direction.STATISTICAL DESCRIPTION 3 1.2 The development of a theory for turbulent flow Turbulent flow forms such a complicated mathematical problem that its solution must lean heavily on experimental data. One dis- advantage is that experimental measurements commonly concern those quantities that are easy to measure rather than those that have an easily understood significance, and the sheer volume and detail of the data may be a bar to the understanding of the physical processes involved. The description of turbulent flow that is the theme of this book has grown in an irregular and haphazard way that may be appropriate to the subject but is a little difficult to follow in the course of its development. Its basis is a number of general principles and hypotheses about turbulent motion and it is convenient to introduce them as the results of study of a group of simple turbulent flows which lack one or more of the usual character istics of ‘complete’ flows. The generalisations are of two kinds, those postulating the existence of various kinds of similarity and those dealing with the nature and mechanism of turbulent flow. Most of the first group are in no way new, having been at least implicit in most work for more than sixty years, but, without detailed assumptions about the nature of the motion, they lead to a number of important results which have been claimed to verify particular views about turbulent flow. Generalisations of the second kind depend on particular views of the structure and dynamics of tur- bulent motion, and their justification depends mostly on the success with which they may be used to predict the quantitative behaviour of turbulent flows. It is important to know which predictions about a particular flow may be derived solely from the similarity hy- potheses before applying those assuming a particular kind of turbulent motion. It is also important not to apply the second kind of generalisation to the wrong kind of flow, a consideration that leads to a classification of turbulent flows by the restrictions placed on energy transfer from the mean flow to the fluctuations by the boundary conditions of the whole flow. 1,3 The statistical description of turbulent flow The methods of statistical mechanics are used in the deseription of turbulent motion to the extent that use is made only of statistical4 TURBULENT SHEAR FLOW mean values of the flow variables, i.e. of particle position, particle velocity, pressure and so on, but there are important differences. Unlike the molecular motion of gases, the motion at any point in a turbulent flow affects the motion at other distant points through the pressure field, and an adequate description cannot be obtained by considering only mean values associated with single fluid particles. This might be put by saying that turbulent motion is less random and more organised than molecular motion, and that to describe the organisation of the flow requires mean values of functions of the flow variables for two or more particles or at two or more positions. Even in the simplest (statistically) of turbulent flows - isotropic turbulence - the number of these functions necessary in the theory is large and, for normal turbulent flows whose asymmetry imposes still more organisation, an even larger number seems to be neces- sary. All this is true ‘ut, if the development of the theory is to be guided by experimental measurements, use must be made of prac- ticable specifications that are very incomplete by the standards of the unaided theory. A flow may be specified either in the Lagrangian way by X(Xo, fo; #), the position at time ¢ of a particle which was at position Xo at time fo or in the Eulerian way by u(x, /), the velocity of the particle which is at position x at time t. No two realisations of a turbulent flow are identical and the complete statistical description is contained in the distribution function for the flow specification X(Xo, fo; 1), which is the density in function-space of the points representing the realisations of the flow. In practice, measurements of Lagrangian flow quantities are extremely difficult and Eulerian ones are almost always used. If the distribution function for the Eulerian specification is F[a(x, 1)], the statistical or ensemble average of M[u(x, 1)], a function of the flow field in space and time, is (Mlu(x, \I> = J Flu(x, 1] Miu(x, nav, the integration being over function space. For some flows and for some measurements, the probability average may be the only possible average, but the majority of flows studied in the laboratory are statistically stationary with respect to time, i.e. in a co-ordinate system moving with a suitable uniform velocity, usually zero, the velocity components at a fixed point are stationary random functionsSTATISTICAL DESCRIPTION 5 of time. If this is so, the ergodic hypothesis asserts that the mean value with respect to time, M= lim none ht [ M(t dt, (13.1) is identical with the ensemble average. If the flow possesses sym- metry with respect to a plane or an axis, the flow variables may be stationary random functions of one or more space co-ordinates, and mean values over the appropriate directions are identical with time means. If the flow variables are stationary random functions of any space or time co-ordinate, mean values are independent of that co- ordinate and the flow is statistically homogeneous for the co- ordinate. Mean values of velocity, pressure, temperature and concentration are comparatively easy to measure, and it is convenient to write their instantaneous values as sums of the mean value and the fluctuation from it. Necessarily, the mean of the fluctuation is zero. The physical significance of the distinction between mean values and fluctuations depends on the nature of turbulent transfer of energy and entropy from the mean flow to the fluctuations, which is normally a one-way process. Although the kinetic energy associated with the velocity fluctuations is, considered thermodynamically, free energy, it is most unusual for any substantial part of it to be changed back into kinetic energy of the mean flow and it may be regarded as a degraded form, intermediate between the organised and easily available energy of the mean flow and thermal energy. 1.4 Notation for turbulent flows Two kinds of notation are in common use to describe turbulent flows, the compact suffix notation used for Cartesian tensors and the older notation used by Reynolds which differentiates more obviously between the various components of the vectors and tensors. If the motion is statistically isotropic, the choice of co-ordinate axes is unimportant and one component of a vector has no more and no less significance than either of the other two. In the inhomogeneous flows which are far more common, several directions are marked out by the symmetry and homogeneity of the flow and velocity6 TURBULENT SHEAR FLOW components in these directions are physically distinct. It has become the custom to distinguish co-ordinates and components in these directions by using different letters as symbols rather than the less obvious suffixes, but, in the most gencral treatment of turbulent flow, directions are not specified and the suffix notation is more compact. For these reasons, the notation used below is mixed. Where it is convenient, the suffix notation will be used, i.e. a velocity vector u has components u,, u2, us along the co-ordinate axes Ox,, Ox2 Ox;, but if the axes are those appropriate to the particular flow the components become u, v, w parallel to Ox, Oy, Oz. The axes are chosen as a rule so that Ox is in the general direction of mean flow and Oz is in the direction of maximum gradient of mean velocity or mean temperature. For axisymmetric flows, cylindrical polar co-ordinates are used but the notation is similar. 1.5 Three-dimensional correlation and spectrum functions The first problem is to obtain from experimental measurements a clear idea of the structure and motion of the turbulence. From now on, frequent references will be made to ‘eddies’ of the turbulent motion, a word intended to describe flow patterns with spatially limited distributions of vorticity and comparatively simple forms. Examples are the Hill spherical vortex and the simple vortex ring. It is supposed that real turbulent flows are the superposition of many such eddies of different kinds sizes and orientations. ‘Since the experimental data is always incomplete, the identification of eddy types must be by informed guesswork followed by measure- ments designed to confirm the guess, and then to fit the inferred structure into a coherent dynamical account of the motion. The first stage depends on familiarity with the meaning of the mean- value functions that are used and on an understanding of the relation between the form of the functions and the presence of particular forms of eddy. The mean-value function most used to examine the spatial structure of turbulence and its evolution in time is the co- variance between velocity components measured at two separated points in the flow, the double-velocity correlation function or, more briefly, the correlation function. It is defined as Rix; 1, 1) = u(x, t) (x+y, f+7), (1.5.1)STATISTICAL DESCRIPTION 7 where u(x, t) is the instantaneous value of the ith component of the velocity fluctuation at the position x and time #. Restrictions on the form of Ry are imposed by the condition of incompressibility, div u = 0, and by the interchangeability of u, and u, in the definition. They are é jy, Rucci) = 0 (1.5.2) and ROG r,t) = Ryx+e; —1, —1)- (1.5.3) It is to be expected that the motion in one part of the fluid is statis- tically independent of the motion in a sufficiently distant part, and 8o R,, should become negligibly small for |r] more than some value characteristic of the scale of the flow. For similar reasons, the correlation function becomes small for large values of the time interval. The complete Correlation function is a function of position in the flow and of four displacement variables, which places it beyond experimental measurement unless it exhibits an abnormal degree of symmetry and homogeneity. Most attention has been focused on the simultaneous correlation function with t = 0 and on the time-space correlation with spatial separation in the direction of mean flow. In this section we shall consider the relation of the simultaneous correlation function to the eddy patterns of the turbulence, using for illustration the contributions to the function of random distributions of eddies of similar forms. An eddy velocity distribution that can take many forms is defined by: 6 us agle 1? cos 1,x1 cos 1x2 cos I5x5], 2 lg = efem#* 0s yx, 608 [3x3 608 Iyx3], (5.4) ax uy =0, where Px? = ty2xy2 4022x274 at52K57, representing a finite, three-dimensional array of eddies. If a turbulent flow contains these eddies with their centres distributed randomly@) © 08 odSTATISTICAL DESCRIPTION 9 but statistically uniformly in space, it is easy to show that the con- tribution to the correlation function is a function of r with non-zero components, Ry (t) = ar J, (1.5.5) Ry Alr) = Rar) = where Sr) = Ae Fos yr He) x (cos [pry +e72™*Y(cos Iyry-tem 8"), A is a constant specifying the intensity of the eddy system, and or? = ar azn +457r57 With particular values of the defining constants, the basic velocity pattern can take the form of several eddy structures of physical interest, including the isolated simple eddy, a periodic array of simple eddies and intermediate arrangements such as a finite row of eddies. Three interesting forms are sketched in fig. 1.1: Type A: |, = = 1s = 0. This represents a simple eddy with circulation at right angles to Ox,. The non-zero components are Ry = $Aa,*(1— 42377?) eo, Raz = Aa (1-day? ye t??, (1.5.6) Ryy = Ry = 4Aay2ay? ry, 7 " and become very small for large values of |r| (greater than about 3a"), Type B: a, = a = #; = 0. The motion is periodic in space and infinite in extent. The components are Roz = Aly? 08 Iyr4 C08 fyr2 C05 Isr, Ryy = Aly? 08 yr 608 fyr2 608 Isr, | (5.2) Ryz = Ray = Abd; sin yr, sin br, 605 Irs 1.1, Simple eddy structures. (a) Isolated eddy (type A); section at z a, ~ 4x3. (6) Periodic array of eddies (type B); section at z= 0, J; (©) Finite row of simple eddies (type ©); section at z= 0, as/h numbers are the values of the stream function for two-dimensional flow, =o = 0.10 TURBULENT SHEAR FLOW and are likewise infinite in extent, oscillating with the same period as the velocity in the original disturbance. Type C: ly = I; = 0. The motion is a finite array of simple eddies with centres along a line parallel to Ox,. The components are: Ryy = $4a,7(1 —4an?r2?)(cos Myr, tem) ewe? Ryo on Arter) (1.5.8) 2 =2hr, sin hin=25 cos lyr, fem4"*", Riz = Ray = ban" [dayryra(cos hry be) 4 ]yry sin rJede which oscillate in the Or, direction with nearly the period of the eddy system but decreasing amplitude. In the other directions, they fall off as exp(— 42?r2), While the correlation function for turbulence composed of simple eddies must be a smooth and simple function of separation, a simple correlation function does not imply that the turbulence is composed of simple velocity patterns. For example, consider turbulence com- posed of quasi-periodic flow patterns of type C, each with different values for the characteristic wave number, /,. If the values are normally distributed around a mean value /, with standard deviation B, the correlation component R,, is | (1.5.9) and the periodicity of the patterns is nearly undetectable in the correlation function if B/ls is more than about 0.4. Even if the ratio Bilo is smaller, the extent of the correlation function is considerably less than the length of the basic patterns. For the detection of periodic flow patterns with a wide range of periods, it is necessary to study higher-order multi-point correlations or their equivalents. With simple eddies, the extent in r-space of appreciable values for the correlation function is comparable with the size of the largest eddies of the turbulence, and consideration of the correlation function for large separations may lead to valid conclusions about Ry, = $Aay?(1—$oy7r,7) PP -[° 4? Cos Lor, tm tet F .STATISTICAL DESCRIPTION IL the form of these eddies. Typical turbulent flows contain eddies with a wide range of size, and the distribution of energy over the range of size is an important quantity for the discussion of the motion. Using the correlation function to assign energy to eddies of a particular size (or range of sizes) requires the construction of a function whose magnitude for argument r is clearly related to the energy of eddies with ‘diameter’ r, One method of adapting the correlation function to give a distribution function for eddy size is that of the structure functions used by A, N. Kolmogorov in his theory of local isotropy. A quadratic structure function may be defined as the mean product of velocity differences between points in the flow, Bix; = Tu0o = ux + r))luj(x) — u(x + 1). (1.5.10) It may be expressed in terms of the correlation function as Bi (x51) = Ri fx; 0)+ Ri(x+r; 0) —Ry(xj 1) Ry(x+e; —1). (1.5.11) It is argued that eddies of scale much larger than |r| contribute very little to the velocity difference, u(x) — u(x +1), because their contributions to u(x) and u,(x + r) are almost identical. It is argued also that the contributions of eddies of scale much smaller than |r| is negligible but, although they may contribute less than an eddy ‘matched’ to the separation, the difference is not great and it is better to regard By,(r) as determined by all eddies of size less than or comparable with [r|. For homogeneous turbulence of the kind generated by the model eddies, we say that the contribution to the mean square velocity, u,2 = Ry ,(0), from eddies of size r or less would be, say, R110) — Rirlr, 0, 0) and that the contribution from eddies of size r in unit range of log r is @ ~ 7 [RuO-Risl, 0, 0))- For simple model eddies (type A), the distribution function so ob- tained has a fairly sharp peak around r = 2271, and the function would give a reasonable description of the distribution of eddy size in turbulence containing similar eddies. A practical advantage of12 TURBULENT SHEAR FLOW using structure functions is that velocity differences are com- paratively insensitive to the slow and uncontrollable changes in the flow which always occur during measurements in the atmosphere or in the ocean, Another way of analysing the distribution of eddy sizes is to use the three-dimensional Fourier transform of the correlation function, defined as 1083 &) = 2x)" f Ry(xiem™ dV@), (1.5.12) the integration being over all r-space. The inverse relation, Ryn) = J By0x k)e* AVR) (1.5.13) means that ®,, is the contribution to wju) = Rj,(0; x) from Fourier components of the velocity field with wave numbers in unit volume of k-space. The assumption of incompressible flow leads to the restriction that ky, (x; k) = 0 (1.5.14) ‘i ‘equivalent to equation (1.5.2). If the turbulence is inhomogeneous with scale Z, the values of the spectrum function for values of |k| comparable with or smaller than >? are strongly dependent on the inhomogeneity and should not be used in a description of the turbulent motion. For separations con- siderably less than L, the variation of the correlation function with x is small compared with its variation with r and then the local spectrum function, 0x5 k) = ny? (, Rixse™ avn) (1.5.15) derived from correlation measurements within a cube of side 28, is related to the ordinary function by 2 pepe A 'B sin ky’ oon = (2 B sin ky’B sin ky’B In, aL kB KYB kyB x ,(k-+k') dV(k’). (1.5.16) That is, the local spectrum function is the result of measuring with a resolution given by the bracketed factor, roughly an average over a volume mB? in k-space, and is nearly the same if |k|B is large. For values of the wave number considerably more than L~},STATISTICAL DESCRIPTION 13 the spectrum function has a local significance and is not directly dependent on the inhomogencity. For many purposes and particularly for the discussion of isotropic turbulence, the integrated spectrum function, defined as £0) =|), 100 dst, (15.7 where the integration is over a spherical surface of radius k, is more useful. It represents the contributions to wu, from wave numbers with magnitudes equal to k. The spectrum function for the correlation function of equation (1.5.5) has non-zero components, Oy = kG), On = kPa) 1.5.18) O12 = On = —kykrg(k), (518) where ky? +17 ky? +1? ky? +13? k) = Aexp| Ae th ke + he ks + Is" g(k) ovo me a? ay Ik, 1. Jk. x coat (2) cont?) oo (24). Oy ay ay The components have maxima in the neighbourhood of the wave numbers (+, -t/, +/3), unless the a// are large when maxima occur near the ellipsoid, Lire a? a” ay? The special forms of eddy pattern have the following spectrum functions: Type As h= =0. The components of the spectrum function are ky? ka? ks? = Ak,? exp—( 54. 92543, 11 = Ak,” exp (Gs , Ky? ka? | ks 2 a ka? ks ©, = Ak, one > (1.5.19) ky? ky? ks? O13 = Oy, = —Akyk, oo-(S +455)14 TURBULENT SHEAR FLOW If a, =a, =a, which represents a ‘spherical’ eddy, the integrated spectrum function has non-zero components Ey, = Ey, = Ankt oP? (1.5.20) which have sharp maxima at k = 2'/?a, Since a~' is a measure of the size of the eddies, it is seen that the dominant Fourier com- ponents for eddies of size L have wave numbers near L~?. Type B; a, =a, =a3 =0, This infinite periodic array has a discrete spectrum function which is non-zero only for k = (+h, th, th). Type C: ly = ly = 0, 04 =a = 45. Depending on the relative values of /, and a, the spectrum function has maxima near k = (41,0, 0), or, if « > J, near |k| = #-". To show that turbulence composed of isolated eddies can be represented by a small range of wave numbers, the eddy velocity distributions for a//, = 1/m and J(2/m are shown together with the corresponding integrated spectrum function in fig. 1.2. Whether the basic eddies composing the turbulent flow resemble simple eddies or periodic arrays of eddies, the integrated spectrum function of a random distribution homogeneous in size is con- centrated in wave numbers close to the inverse of the diameter or wavelength of the component eddies, and so the function can be used to express in quantitative form the relative intensities of physical eddies of different sizes. There are two exceptions to this con- clusion: (1) Any finite superposition of eddies with minimum size L leads to a form of £,,(k) such that E,(k) = Cyk* + terms of order k* for kL <1, and no special meaning can be attached to the form of the spectrum at low wave numbers. Its magnitude depends mostly on the largest eddies of size around L. (2) Viscous dissipation sets a lower limit to the size of physical eddies and the spectrum function for wave numbers larger than the reciprocal of the minimum size is determined by that size, say a, and varies nearly as e~*™, These two qualifications to the usual identification of eddies with Fourier components of the velocity field are due to the initialSTATISTICAL DESCRIPTION 15 £0} ro) Kh Fig. 1.2. Spectrum function of a simple isolated eddy. (a) Velocity distributions, (8) three-dimensional spectrum functions: for -~~-, « =/,/7; —, «=hv2/"16 TURBULENT SHEAR FLOW assumption that a physically acceptable eddy must be of finite extent, i.e. resemble the eddies of types A and C rather than type B. It is important to realise that the correlation and spectrum functions form a very incomplete description of a turbulent motion, and that the central role they play in current theoretical and experi- mental work is due to their comparative simplicity and convenience. In principle, the statistical description of a turbulent flow requires a knowledge of the complete joint-probability distribution function for realisations of the velocity field, The correlation function is one of the infinite set of integral moments of the basic function and attempts to use it as a complete description amount, mathematically, to making hypotheses concerning the nature of the complete function, and, physically, to making statements about the eddy structures that exist in fully developed turbulence. For this and other reasons, inference of velocity patterns from observed spectra and cor- relations is an uncertain process, usually involving preconceptions whose validity is always open to doubt. 1.6 One-dimensional correlation and spectrum functions Incomplete though the specification of a turbulent field by its cor- relation function may be, the observations necessary to determine it are so numerous that the primary information nearly always concerns only the particular correlations Rir(K57,0,0), RirKi0,7,0), — Rarlx; 0,0, 7) or the Fourier transforms of Ris(x57,0,0), — Ra2(Xi 7, 0,0), Raa(x; 1, 0, 0) with respect to r (Ox; is in the direction of mean flow). Unless the turbulence is isotropic, a knowledge of any or all of these functions is insufficient to determine R,,(x; r), but they are useful for setting scales to the motion and for inferring the eddy structure. A useful result can be obtained from the condition of incom- pressibility (1.5.2). Integrating with respect to all values of r, and ry gives a wf Roce 1) dr, dry = 0, (1.6.1)STATISTICAL DESCRIPTION 17 since R,(x; r) is expected to be small for all large values of |r|. It follows that t c Ry(X3 Py, Pa, 73) dr, drs = a constant (1.6.2) which must be zero. In practice, we are interested in Ry sx; 0, 72,73) for which [ f° Riss 0, ra, 73) dr2 drs = 0. (1.6.3) The physical meaning is that the instantaneous flux across any closed surface is zero and that the compensating inflow across a plane x, = constant in response to an outflow at a particular point takes place mostly at points displaced by values of rz, ry such that Ry, (x; 0, rz, 75) is negative. In isotropic turbulence, fF 7R,4(0, 7,0) dr = 0 (1.6.4) ° and the return flow takes place at distances that make rR, ,(0, 7, 0) a minimum. In anisotropic turbulence, the return flow may be concentrated in a plane and then negative values of R, (0, r, 0) may be numerically much larger (or smaller) than those of R11(0,0,r). Further, if the eddies are all much the same size, return flow takes place over a limited range of r, but a wide range of eddy sizes implies return flow over a wide range of r and consequently smaller negative values for the transverse correlations. If eddies of the kinds considered in § 1.5 are taken to be typical of simple eddies, it is clear that the three-dimensional and one- dimensional correlation functions are smooth, i.e. the curvature is nowhere large if all the component eddies are of about the same size, The occurrence of locally high curvature implies the presence of a wide range of eddy sizes, which may take two forms. In the first form, a wide and continuous range of eddy sizes leads to large curvature near r = 0, a feature of nearly all correlation functions. for turbulence of high Reynolds number. The other form occurs when there are two distinct ranges of eddy size present, and arises from the addition of two correlation functions of very different scales, Then the curvature is large near some positive value of r and the composite function has a characteristic two-component18 TURBULENT SHEAR FLOW form. Examples occur in free turbulence and even more strikingly in correlation functions for atmospheric turbulence. Os os Rule, 0, 0) ° , 1 Fig. 1.3, Correlation functions for isotropic turbulence with eddies of uniform size, The statements made above are illustrated in figs. 1.3-1.6, which show schematically correlation functions for the following types of turbulence: (1) isotropic turbulence of uniformly sized eddies, (2) isotropic turbulence of a wide range of eddy sizes, (3) isotropic turbulence with two distinct ranges of eddy sizes,STATISTICAL DESCRIPTION 19 (4) ‘two-dimensional’ turbulence in which us = 0, (5) ‘two-dimensional’ turbulence for which u, = u2 everywhere. If the velocity fluctuations are small compared with the mean velocity of flow, the changes in the velocity pattern as it sweeps past Ru(.7,0) Ril, 0, 0) 0 1 , 2 3 Fig. 1.4. Typical transverse and longitudinal correlation functions for isotropic turbulence with a wide spectrum of eddy sizes. a fixed point are negligible and a time displacement of ¢ is equivalent to a displacement in the flow direction of —U,t, where U; is the mean velocity. Then the Fourier transforms of R,,(x;r,0,0), 220 TURBULENT SHEAR FLOW Ru(,7,0) s 0 os 1 Ls Fig. 1.5. Correlation functions for isotropic turbulence with eddies of two distinct sizes. (Equal intensities, size ratio Ra2(x; 7, 0,0), Rsx(X;r,0,0) are nearly proportional to the fre- quency spectra of the velocity components, 14, «, and ws, for frequency kU,. The frequency spectrum is readily measured with electrical spectrum analysers. In terms of the three-dimensional function, the one-dimensional spectrum function is difky) = i J ° Ry fr, 0, 0) e™" dr (1.6.5) = af i ©, dks dks,STATISTICAL DESCRIPTION au Ru, 0,0) 0 as , 1 15 Fig. 1.6. Correlation functions for turbulence with motion parallel to a fixed plane, —, us = 0 (motion parallel to x10x2); ~-—, 1 = Ma. showing that a single value of ¢,,(k,) represents the sum of values of ©, for a wide range of |k|, from k, to infinite wave number. So while a simple eddy structure leads to a three-dimensional spectrum function which is large only over a limited range of |kl, it leads to a one-dimensional spectrum function hardly more confined in extent than the corresponding correlation function. For isotropic turbulence, the integrated spectrum function (1.5.14) is related to $r1(k1) by 2.) = fo Pee® _ , Mout die dk (16.6)22 TURBULENT SHEAR FLOW which shows the size distribution to be determined more by the curvature of the spectrum function than its magnitude at any par- ticular wave number. For example, $,,(0) is almost always not zero and its value is determined exclusively by values of ,, at non-zero values of k. It does not indicate a finite intensity of very large eddies. The one-dimensional spectrum function may be used in a similar way to the correlation function to reveal anisotropy of the com- ponent eddies. If the motion of the turbulence is almost confined to the x,Oxz plane, backflow will take place in the Ox, direction for a uz outflow and JP Rests 0, 0 ar should be nearly zero. The corresponding spectrum function #0) =f" Ralr.0,0)6°H ar will also be nearly zero at kK, = 0. Roughly, spectra of the shape 1 in fig. 1.7 imply motion in the x,Ox, plane, of shape u, motion in Onlks) ol o ky Fig. 1.7. One-dimensional spectrum functions for isotropic and two-dimensional turbulence. Curve 5, ts = 0; curve it, us = 0; curve mn, isotropic. the x,0xs plane, and of shape 1m, motion either equally distributed or at 45° to Ox,. Isotropic turbulence gives spectra of shape 111. If two distinct ranges of eddy size exist, the spectrum functions take characteristic two-component forms similar to those of the correlation function. It is worth noting that a low-intensity group ofSTATISTICAL DESCRIPTION 23 large eddies in a background of smaller eddies is much easier to detect from the appearance of the spectrum function than from the appearance of the correlation function. 1.7 Correlations and spectra with time delay The simultaneous correlation and spectrum functions provide in- formation about the instantaneous flow patterns, but a description of their growth and development requires comparison of flow patterns at different times, conveniently by appropriate use of the complete space-time correlation function. Changes of flow pattern that affect the correlation can arise either by displacement of individual eddy patterns or by change of the patterns themselves. In turbulent flow, localised eddies move with respect to the surrounding fluid with a self-induced velocity that depends on their structure, a familiar example being the motion of a vortex ring. So the ‘centre’ moves with a velocity compounded of the local mean velocity, the instantaneous local velocity of the larger eddies in which it is im- bedded, and the self-induced velocity. If it were possible to follow an eddy centre, the evolution of a single eddy could be studied but Eulerian measurements allow only a determination of the average displacement. In general, it is not possible to say whether the apparent change of eddy pattern after allowing for the average dis- placement is a real change or merely the effect of a deviation of the eddy centre from its mean position. Addition of a time variable adds to the complexity of experi- mental measurements, and most of the available measurements refer to the space-time correlation function for a single velocity component and spatial separation in the direction of mean flow. The general appearance of the function, Ry (x; 7, 0,0, t), is shown in fig. 1.8 as variations with r for several fixed time delays. With increasing time delay, (1) the heights of the maxima with respect to variation of r become less; (2) rm» the position of the maximum, increases nearly proportion- ally to time delay, the ratio r,,/t being defined as the convection velocity; (8) the radius of curvature at the maximum becomes greater; (4) the variation about the maximum may become asymmetric.24 TURBULENT SHEAR FLOW ‘The first two effects may be interpreted as the consequence of a velocity pattern that moves with the convection velocity but also changes and loses identity in a way described by the variation of maximum correlation with time delay. Since u, is a characteristic of the whole motion, the convection velocity and loss of correlation ~ Curve of maximum Ril’, 0, 05 1)] correlation Fig. 1.8. Time delay correlations, as functions of streamwise separation for various time intervals. are averages over eddy components of all sizes but the last two effects are indicative of different behaviour of smaller-scale com- ponents and suggest that more information can be gained from a closer study. The object is to recast the correlation function into a form that discriminates between eddies of different sizes. In §1.5, it was argued that the velocity difference, u(x,t) — u(x + 4,1), is determined mostly by eddies of size |r| or less. As a rule, the smaller eddies of turbulent flow hold less energy than the larger ones and the major contribution to the difference comes always from eddies of size comparable with the separation. Hence the velocity difference may be interpreted as a measure of the in- fluence of eddies of size r, on the flow velocity at the point x + 4r and the space-time structure function, Bys(% 85 8,1) = [uy uxt, 1) x [us (k+s, 1+1)—u,(x+s+r, t+1)] 11(X5 ST) Ry X48; §, 7) —Rix(xs 8t4, )—Ry (K+; 84,0), (1.7.1)STATISTICAL DESCRIPTION 25 ie. the covariance between velocity differences distant apart s in space and t in time, measures the changes in eddies of size r. The relation between the changes in the eddies and the changes of the structure function can be discussed by supposing that the velocity pattern for eddies of size r 1s known so that a single eddy with centre at Xo has the velocity differences, my (x) — u(x + 1) = af(% — Xo, 8), where f is the same for all eddies of size r and a is the velocity amplitude of the eddy. (Variations of orientation and type can be introduced if it is thought necessary.) For a particular eddy, the amplitude and centre position are functions of time and the con- tribution of one eddy to the structure function is {a(e)a(e + 1)} x {fEx — xo(*)IIx + 5 — xo(t + 7)]}- The first factor depends on a real change in eddy amplitude and the second depends on translation of the eddy. To make up the whole structure function, we need the probability distribution function, Pla(t), a(t + 1), xo(t), xo(t + 7D], defining the probability of observing particular values of the am- Pplitudes and centre positions in one realisation of the flow. For a steady flow, By,(%, 85 8,1) = J Plavo), A(t) —a(0), X00), Xo(t)—X0(0)] x a(O)a(e)fOx—xo(O))fCx—Xo(#)-+8) AV, (1.7.2) the integration being over all values of a(0), a(t), xo(0) and xo(¢). Consider now a homogeneous distribution of eddies of size r, with centres distributed randomly in space with uniform number density N and with amplitudes independent of the centre positions. The distribution function factorises to NP, (a(0), a(z) — a(0)]P,[x0(t) — xo(0)] and Bis(s.2)= Jf Prla(O), a(e)—a(O)Ja(O)a(e) dato) date) XN IJ Palxo(s)—xo(O1 fIx— x00] xS1X—Xo(t) +5] d[xo(t) —X0(0)] aoe festa ~ xo(0)]B a8 — x) +9(0), 0] x d[xo(t)—x9(0)]- (1.7.3)26 TURBULENT SHEAR FLOW The first factor is simply the autocorrelation coefficient for indi- vidual cddy amplitudes with time delay t and describes the intrinsic changes in the eddies. The second factor depends on the movements of the eddy centres in the time interval, and the effect can be de- scribed as the combination of a translation in s-space of X(t) — xo(0) and a diffusive spread in s-space by random movements of the eddy centres. The diffusive spread is a problem in Lagrangian diffusion, but experimental studies suggest that a fair approximation to P, is P2[Xo(t)~X0(0)] = 2x) 220-9 expl—H(xo(t) —X0(0) — (Xo(t) — Xo(0)))?/07] (1.7.4) where a is the standard deviation of the centre displacement about its mean value. Then the maximum value of B,,(s, t) occurs where Sq, = Xo(t) — Xo(0) = 5 (1.7.5) defining a convection velocity, U, = S,/t, and the magnitude of « could be found by comparing the shapes of the structure functions for r = Qand for time delay t. With a knowledge of ¢, the reduction in height of the maximum by diffusion can be calculated and used to find the autocorrelation coefficient for the amplitude of individual eddies. The smaller eddies are carried around by the larger ones and, for them, the convection velocity of their eddy centres is nearly the local velocity. If up and Lo are the scales of velocity and length for the main turbulent motion, particle velocities are expected to remain nearly unchanged for times short compared with Lo/uo and then Xo(t) — Xo(0) = Ur + w(O)r, (1.7.6) It follows that o = (u2)'2t (1.7.72) and we sce that the effects of diffusion are small if (4,2)'/2z < r, and that they reduce the structure function to a negligible value if (u'r > r. Unless the autocorrelation coefficient changes ap- preciably in a time interval of r/(12)"?, it is difficult to distinguish the effect of its variation on the delayed structure function from the effect of diffusion. In inhomogeneous turbulence with spatial variation of mean velocity, the interpretation is less simple, For example, eddies of aSTATISTICAL DESCRIPTION 27 particular size may tend to aggregate near a plane parallel to the direction of mean fiow and the convection velocity of the centres will be nearly the mean velocity at the plane. If the convection velocity is determined from structure functions for x in another part of the flow, it will differ from the local mean velocity, being in general intermediate between the local velocity and the convection velocity of the centres. Further, the maximum value of the structure function taken over a plane through x and parallel to the preferred plane can be increased if the point of delayed measurement at x + $ is moved towards the preferred plane where it receives a larger contribution from each eddy. The apparent convection velocity defined by the condition of maximum structure function is then directed towards the preferred plane although the eddies move parallel to it. Usually it is only the larger eddies that show a prefer- ence for a particular location and show convection velocities different from the local mean velocity either in magnitude or direction. For some purposes, it is better to represent the contribution to the flow from eddies of a particular size by a group of Fourier components which may be isolated from an electrical signal by a band-pass filter. Several forms have been used, in particular the” space-time spectrum function, baths 0) = aes [f Ri(r, 0,0, te" de de, (1.7.8) where the wave number k refers to displacement in the direction of mean flow. The identification of eddies of size k~' with Fourier components of wave number k is justified by the observation that eddies of size much larger than k~! make negligible contributions while the contributions from smaller eddies are widely dispersed in waye number and are less in total energy. Then the evolution in time of an eddy is described by the spectrum function for a range of k covering perhaps an octave of wave number. The usual appear- ance of the spectrum function is indicated in fig. 1.9 by lines of constant spectral intensity on the kw plane. Typically, large in- tensities are concentrated near a line defined by the condition, seule) 6, ars)28 TURBULENT SHEAR FLOW and the slope of the line, —da,/dk, is the convection velocity for eddies of size k-. The variance of spectral intensity about the central line is caused both by the variability of convection velocity and by real decay of individual eddies, but the separation of the two effects is not as clear as with the structure function. The value of the spectral representation lies in the simple and direct presen- tation of the magnitudes of convection velocity for eddies of the whole range of eddy size. ey Fig. 1.9. Contours of equal spectrum intensity in (k;, #) plane, Other representations of the information contained in the space- time correlation function are possible and have been used. Some of these are discussed by Wills (1964). 1.8 Homogeneity and symmetry of turbulent flows Study of turbulent flows is much casier if they possess properties of symmetry and homogeneity that reduce the complexity of their statistical description. The symmetry and homogencity may be properties of the flow boundaries and the forces that drive the flow, or it may arise from the tendency of turbulent flows to ‘forget’ details of their initiation and to assume as homogeneous and sym- metrical conformation as is possible. For example, a jet from a round nozzle is hardly distinguishable from one of similar mo- mentum flux issuing from a nozzle of irregular shape except close to the plane of exit. Although the boundary conditions of practicalSTATISTICAL DESCRIPTION 29 (013) Mean velocity x (or) ‘(or x3), «@ Mean velocity Mean velocity Fig. 1.10, Co-ordinate systems for (a) Two-dimensional mean flow. (b) Axi- symmetric mean flow. (c) Flow with circular or helical streamlines (between rotating cylinders.) flows may have no particular symmetry, the majority of them resemble ideal flows which are homogeneous and symmetrical in some respects. The ideal flows are best described in co-ordinate systems so chosen that the boundary conditions are easily stated30 TURBULENT SHEAR FLOW and any symmetry or homogeneity can be expressed as invariance with respect to co-ordinates or interchange and reversal of co- ordinates. The system of Cartesian co-ordinates used to describe flows with plane symmetry and the cylindrical polar co-ordinates for axisymmetric flows is shown in fig. 1.10. Most turbulent flows are strongly inhomogencous with respect to variation of one co-ordinate, 2 in plane flow and r in axisymmetric flow, and the velocity com- ponent in this direction is w. Many are unidirectional (or nearly so over the significant part of the flow) flows and the co-ordinate axis Ox is in the direction of flow and is the axis of axisymmetric flows. u is the corresponding velocity component. The remaining co- ordinate, » for plane flow and ¢ for axisymmetric flow, defines a direction in which the flow is homogeneous and the corresponding velocity component is v. Flows with strongly curved streamlines, c.g. the flow between concentric rotating cylinders without axial pressure gradient, have their velocity in the direction of variation of ¢, and it is logical to use u for the velocity component in this direction and to use v for the axial component of velocity and y for the axial co-ordinate. The more important flows may be classified by their homogeneity and symmetry, and the table shows the principle classes with the appropriate co-ordinate systems and the more important members.STATISTICAL DESCRIPTION 3 Classification of turbulent flows by homogeneity and symmetry Description Members of class System of Homogeneous Mean flow co-ordinates in Homogeneous Isotropic Cartesian x, y,7 x direction axisymmetric turbulence Grid turbulence Flow behind a Cartesian wz x direction uniform grid in a uniform stream Unidirectional Flow between Cartesian x, y x direction flow parallel planes, plane Couette, pressure flows Flow ina Cylindrical x, ¢ x direction circular pipe or polar (6) between concentric cylinders Rotating flow Flow between Cylindrical 4, $ direction rotating cylinders polar (c) without pressure gradient Skewed flow Combined Cartesian x,y in xy plane pressure and Couette flow Two-dimensional Plane jetsand Cartesian» in xy plane or plane flows wakes, boundary but nearly layers, mixing in layers x-direction Skewed flow Boundary layer Cartesian y nearly in on yawed wing ay plane Axisymmetric Circular jets and Cylindrical In axial developing flows wakes, flow in polar (6) plane but conical diffusers nearly in x direction Swirling flow Swirling jets and Cylindrical $ - wakes polar (6) Secondary flow Secondary flows Cartesian x Nearly in ‘non-circular pipes x directionCHAPTER 2 THE EQUATIONS OF MOTION FOR TURBULENT FLOW 2.1. Assumption ofa continuous fluid It is usual to assume that the turbulent motion of gases and liquids can be described by the equations of motion for a continuous fluid without a molecular structure. The assumption has been queried but no inconsistencies have been found so far, and, in gases, whose structure is well understood, it may be shown that departures can occur only for scales of motion and fluid velocities outside normal experience. The essence of the continuum approximation is that the flow velocities and other continuum properties can be defined as averages over regions of space and intervals of time that are large compared with the scales of the molecular motion and small com- pared with the scales of the continuum flow. Separate molecular realisations of a particular flow deviate from the development pre- dicted by the continuum equations in a way that may be represented by ‘random noise’ terms describing molecular fluctuations of the continuum averages. If a continuum flow is essentially uniform over lengths less than /, and times less than fo, the averages may be taken cover cubes of side /, and time intervals of #9. Then the density at a ‘point’ will have a standard deviation from its expected value p of order B= plitly?)” 1(c47to/v)-*7, 1.1) where n is the number-density of the molecules, ¢, is the root-mean- square of one component of the molecular velocity, and v is the kinematic viscosity. Similarly, a component of fluid momentum has a standard deviation from its expected value of about (puy’ = pes(alo?)“*7(ex7tolv)- 7 (2.1.2) Finally, the thermodynamic equation of state can be used to relate density, temperature and pressure only if the number of molecules in the volume /° is sufficient to define a molecular distributionEQUATIONS OF MOTION 33 function and if little change takes place in a time comparable with the relaxation time v/c,?. The two conditions are nlp? > 1, Cy2fo/v > 1. (2.1.3) If the largest eddies of a turbulent flow have a characteristic size Land a characteristic velocity V, it is known that the smallest scales of motion are of size (v3L/V3)"* = Jy, of duration (vL/V2)"/? = to, and with characteristic velocity (vV*/L)'/*. Substitution in (2.1.1, 2.1.2, 2.1.3) shows that deviations from the continuum values due to molecular fluctuations are small if both 9/4 nly? = nu(“) >i and (2.1.4) ety 0?/VL\"? 7 * a >i. In the flow of air along a pipe of 1 em radius with a central velocity of 10*cms™', the velocity and size of the smallest eddies near the wall are nearly 200 cm s~! and 5 x 10-3 cm. In standard conditions nly? w 3 x 10! and c,to/v & 2 x 10%, both large numbers, Only when the velocities involved exceed greatly the molecular velocities will the continuum approximation fail to describe turbulent motion. 2.2 The equations of fluid motion The equations describing the motion of a continuous fluid are derived from the conservation laws and the fluid properties, in the form of the equation of state and the relations between stress and rate of strain and between conducted heat-flux and temperature gradient. The vector flux of mass at any point in the fluid is pu, and conservation of mass is expressed by equating the divergence of the mass flux to the rate of decrease of local density, i.e. by or by (2.2. + The usual summation convention is used, that a repeated free suffix implies summation over the three possible values, e.g. bu, Ou, | ou, us Bx, 7 Bx, * ay By”34 TURBULENT SHEAR FLOW where 6 6 ate Fa measures the rate of change following a fluid particle, ie. a point that moves always with the continuum velocity at its current position. ‘The tensor pu,u, is the flux of fluid momentum in the Ox, direction across a surface element normal to Ox, and conservation of mo- mentum is expressed by equating the difference between the rate of increase of momentum and the divergence of the flux to the rate of gain from external forces and by molecular forces and migration across the boundaries of the control volume. For external forces described by a potential ¢, the condition for conservation of momentum is ous) “a ap pa 2 Flows) = — Fea Pe ob ax; ax, 7? oe, (2.2.2) where p is the thermodynamic pressure, and p,, is the stress tensor arising from departure from thermodynamic equilibrium consequent on straining of fluid elements. Using the condition for conservation of mass (2.2.1), pou Dr ‘Ou, ou) op apy ah of Gen =i tae, Pde 223) also obtainable directly by considering the acceleration of a fluid clement. From (2.2.3), an equation for the kinetic energy of the streaming motion can be obtained, a ie (2.2.4) ° Peau’) = ~uge +02 pu The remaining conservation law is that of energy. The energy per unit mass of the fluid is E + 4u,? +¢, the sum of the internal energy E, the kinetic energy of streaming and the potential energy. The total energy flux is the sum of the convected flux of energy, pu{E + 4u,? + ¢), the flux by working of the stresses on the fluid velocities uj(5yp — Py), and the conducted flux of thermal energy, —k aT/@x,, where T is the local temperature and k is the thermal conductivity. Conservation of energy is expressed byEQUATIONS OF MOTION 35 HAsO onde Apu), Ara) , 6 (, ar +7 . “Gx, Ox, ox, By using equations (2.2.1) and (2.2.4), equations for the internal energy E may be obtained, DE_ a au, af, aT" Dr Pax, Pas, aC z) 026) and for the total heat (enthalpy), H = E + p/p, DH Dp, ay, a/, aT ’ De = Det? nage k m: @27 The interpretation is simply that changes of internal energy are composed of work done by expansion against the pressure, heat generated by work against the fluid stresses p,, and the net gain of heat by conduction. The heat generated by working against fluid stresses is denoted by pe = py dujax,. (2.2.8) For the present purposes, the equation of state of the fluid is assumed to be the ideal gas equation, p= RIp, (2.2.9) where, if the fluid is not a perfect gas, the zeros of pressure and temperature are chosen so that the average value of the pressure is the isothermal bulk modulus and the average ‘absolute temperature’ is the reciprocal of the coefficient of thermal expansion at constant pressure, Nearly all the following refers to fluids with Newtonian viscosity, i.e. the fluid stresses are linearly related to the rate of distortion by du, , du, du, mu = (5 Hy +H). ay (2.2.10) where and 4 are coefficients of viscosity. The second law of thermodynamics requires that the rate of conversion of mechanical energy to heat by viscous action should always be positive, and so, since = 7, ote Guy uy 2 pe= me in tu ‘ +3, uz, my +Gu- age zy. (2.2.11) # must be positive and A must be less than $y. For a perfect gas, (2.2.5)36 TURBULENT SHEAR FLOW kinetic theory predicts that 4 = $y, indicating that uniform dila- tation produces no viscous stress. If the viscosity is Newtonian, the viscous term in the equation of motion (2.2.2, 2.2.3) is (2.2.12) 2.3 Approximate forms of the equations of motion In most turbulent flows, the variations of velocity are small compared with the speed of sound and variations of density are small compared with the average density. Then the variations of density may be ignored in so far as they affect the inertia of the fluid or its heat capacity, and the equations of motion take a simpler form. The mass density in the equations for the fluid momentum and internal energy is replaced by p,, an average density over the horizontal plane of constant gravitational potential ¢ through the point con- sidered. Similar average values for the pressure and temperature are defined by the hydrostatic equation, Pa a dx, ~ PB, 23.4) and by the equation of state, Pa = Rely (2.3.2) if the variations of pressure and density are small compared with the average values at the particular level. These average values are independent of time in ordinary circumstances. Using the average values for pressure and density related by (2.3.1), the momentum equation (2.2.3) becomes P= Pa Pu wees s+ 2.3.3) Ox, p Ox, pox 233) without approximation. The buoyancy term, involves the variations of density but, by considering the orders of magnitude of the terms, it can be shown that either the density variations are too small to cause appreciable buoyancy forces or the fractional variations of pressure are small compared with those of density and temperature. Then,EQUATIONS OF MOTION 37 _ «Pape _T-T pT =pTy ite Bt (2.3.4) and, substituting p, for p in the other terms, the momentum equation becomes au, 4 8 (pope) T= Tad, 2 (pu Sty. --(?oP 2.3.5) at! x," al p, J? Te amt amhp,) >> if the vertical variation of p, is small over the characteristic scale- length of the flow. Over small ranges of temperature, total heat is related to tem- perature by H = c,T + constant, and, to the approximation of constant density, equation (2.2.7) becomes Dr D(p—P.) Drs], Pu Ou, k oT Dr “pr Dr} acpax pe xt? 23-9 Since p, and @ are independent of time, Dp. u Pe ob Do Dr ax, ~ Me ae, 8 De and the equation for the total heat turns into an equation for the potential temperature, 7 + ¢/cy, Pirséle,) =nZ, (T+ ¢/e,) (23.7) De Ox? » omitting terms 1 D(p~p.) Pu Ouy pcp Dt and Pat p OX, Here x = k/(p,c,) is the thermometric conductivity. To the approximation of constant density, mass conservation is expressed by the continuity equation, au, ax, ‘omitting terms (D/Dt)(p — p,) and Dp,/Dt. For a Newtonian fluid, the equation of motion becomes ou, du, ap. @u, T-T,d Feu etm Bay Te 4 oe 239) 8 = (2.3.8)38 TURBULENT SHEAR FLOW with a change of notation so that p now denotes (p — p,)/Pa the ‘kinematic’ pressure variation. In all that follows, kinematic pres- sures and stresses are used and mechanical values can be recovered by multiplying by the fluid density. The three equations (2.3.7-9) describe most kinds of turbulent motion. If the flow has characteristic scales for variation of velocity, position and temperature, vo, L and Oo, the ratio of the terms omitted in equation (2.3.6) to those retained is of order up?/a? (where a is the speed of sound) or @/7,. The terms omitted from the continuity equation are of order uo?/a?, Oo/7, or L/L,, where L, = RT/g is the scale-height of the fluid. The terms omitted from the temperature equation (2.3.7) are smaller in ratios of order %02/a?, 0o/T L/L, or g?/(¢ 0c). The conditions that the approximate equations should describe the motion accurately are: (1) that the square of the flow Mach number, uo?/a?, is small, (2) that the temperature loading, 90/7,,, is small, (3) that the scale of the flow is small compared with the scale- height of the fluid. The temperature field described by the equations refers only to variations induced by external heat sources or by interaction with a non-uniform distribution of the ambient potential temperature, T, +c, =T,. The equations do not describe the temperature fluctuations induced by pressure changes or by viscous dissipation of mechanical energy. The ratio of these fluctuations to the ambient temperature is of order u9?/a? for turbulent flows. 2.4 Mean value equations for momentum, energy and heat From now on, the flow variables are expressed as the sum of a mean value and the fluctuation from the mean value, the velocity is U,+ u, where U, is the mean velocity. By definition the mean value of the fluctuation is zero. Then taking the mean value of the continuity equation (2.3.8), AU reu) _ 20s dur Us _ og (2.44)EQUATIONS OF MOTION 39 and the velocity fluctuations satisfy a continuity equation, (2.4.2) Taking the mean value of the momentum equation (2.3.9) and using the continuity condition for the fluctuations, we obtain the equation for the mean velocity in the standard form, Uy, , Us , Sumy _— _ OP gfT-T) » FU, = +U, ox tox Tae TY axe (2.4.3) where P + p is the pressure difference from the ambient Pressure, T + @ is the potential temperature, and T, the ambient potential temperature. The equation for the mean velocity may be put into the form, Be !4U, oe me Te), +3]- oyP+ Sy 1) ww] (2.4.4) showing that the mean flow is accelerated “ foros arising from the mean buoyancy, the gradient of the mean pressure, the viscous stresses developed by the mean flow alone, and by a virtual force which is the gradient of the Reynolds stress, —izju. The additional force describes the effect of the turbulent fluctuations on the mean flow and is simply interpreted as a consequence of the mean rates of transport of momentum by the turbulent movements of the fluid. If it could be determined, the mean flow would be known and the first problem of turbulent flow, the nature of the mean motion, could be solved. The kinetic energy of the velocity fluctuations is a quantity of clear physical significance, and an equation for it can be obtained by multiplying the equation for the total velocity, SOD U4) Ce ~P +P) g(T-T.+8) PU + ud) Ox, T. axe by the velocity fluctuation and taking the mean value. It is Sah +0 OD 2 pn stv Fang Pa = Hien © (2.4.5)40 TURBULENT SHEAR FLOW where q? = u,u,. ‘Fhe equation amounts to a statement of energy conservation for the velocity fluctuations and the terms have simple interpretations. They are: (1) rate of increase of fluctuation energy, (2) gain through advection of energy by the mean flow, 3) production of turbulent energy by working of the mean flow on the turbulent Reynolds stresses, (4) transport of turbulent energy by turbulent pressure gradients and by turbulent convection, (5) gain of energy through working of the buoyancy forces, and (© transformation of fluctuation energy to heat plus a smaller amount of energy diffusion by the working of viscous stress fluctuations. An equation for the total kinetic energy of the flow is also useful. Itis 8a yd 9 aay? Alia +UPY+U, ae +U/)) a — — + ag ligt uaa, + PU, + pu] #U, a = -[ mim weal Ster—TU,+G4] (246) and equates the rate of change to the divergence of an energy flux, with contributions from the buoyancy forces and the viscous dissipation of mechanical energy. The equation for the potential temperature, T + 0, leads to the equation for the mean temperature, oT oT ou eT HUG kaa (2.4.7) ‘Compared with the equation for a flow without fluctuations, the only difference is the presence of the term, (@/2x,)(u0), representing the effect of turbulent transport of heat and analogous to the Reynolds stress term in the equation for the mean velocity. Again a knowledge of the term would permit calculation of the distribution of mean temperature. For the temperature field, the mean square of the temperatureEQUATIONS OF MOTION 41 fluctuation plays a part similar to that of the fluctuation energy. Physically, it is closely related to the contribution of the fluctuations to the mean entropy. To the approximation in use, the difference of the local entropy from the entropy of fluid at pressure p, and tem- perature T,, is SS, = cy loge(T + O/T), (2.4.8) and the mean entropy is S=S, +c, log. T/T, — 4¢,0°/T?. (2.4.9) The quantity 407 is proportional to the (negative) contribution to the mean entropy arising from the presence of temperature fluc- tuations, and the destruction of fluctuation entropy by molecular conduction of heat is analogous to the dissipation of fluctuation kinetic cnergy by viscous stresses. An equation for 30° is casily obtained by multiplying the equation for the temperature by the temperature fluctuation and taking the mean value. It is ae 9 mag ots a) = KV x? )+U;, a(t Dud 5 +30 u,) = KOV70 (2.4.10) and the interpretation of the terms is similar to that of the terms in the energy equation, i.e. the term 4,0 O7/Ax, represents production of ‘entropy’ by turbulent flux of heat along the gradient of mean temperature, the term $6%u, turbulent convection of fluctuation entropy, and the term k0V70 mostly a destruction of fluctuation entropy by heat conduction down temperature gradients. The equation for $[(T ~ 7,)* + 87] is sometimes useful. It is a 2 aR a 21914 2 oar gO 4107)4+ Us LT 1)? +367] + 5 Oud TTD) 6, ot — +3,,0° w+ ge UT TU r+ Oui] = K(T-T)V(T-T,)+0V°6], (2.4.11) 2.5 Energy dissipation by viscosity The equation for the kinetic energy of the fluctuations (2.4.5) con- tains the term vu,(é*u,/dx,?) representing the effect of viscous forces42 TURBULENT SHEAR FLOW on the energy. It is related closely to the local rate of conversion of mechanical energy to heat, in kinematic form, UU), (a, Buy? eee =3(Zi4 ay +(e zy (2.5.1) where aU, 2U,¥ E= 7 i (2.5.2) is the part of the dissipation of mechanical energy due to the mean velocity gradients, and Gu, Ou; = (2.5.3) e (ee) (2.5.3) is the mean turbulent energy dissipation. The viscous term in the energy equation can be “ee as # a mu a Bui Ge) (2.5.4) 172) Puy & aaa Ox; et) after using the continuity equation to show that Buy Guy Fuguy es) ax, dx, ~ dx,ex," ‘The first term of (2.5.4) involves only second derivatives of mean values of velocity products, and has the nature of a generalised diffusion term. If the flow field is bounded or homogencous, the volume integral of the term is zero and the term contributes nothing to actual dissipation of energy. Further, if uo is a typical velocity fluctuation of the flow and the width of the flow is L, the magnitude of the first term is vo?/L? while the actual energy dissipation is of order t3/L. Provided that uoL/v, the Reynolds number of the turbulent flow is large, the turbulent energy dissipation is Fu _ | [om uy wage (ere b oi, = (2) = v0. (2.5.6)EQUATIONS OF MOTION 4B 2.6 Conductive dissipation of temperature fluctuations The equation for the intensity of the temperature fluctuations con- tains a conductive term which is almost proportional to the rate of entropy production by heat conduction down gradients of the temperature fluctuation. It may be written in the form, Kav = Kv7(4ey—1( & . (2.6.1) i)? 6. where the first part describes a viscous diffusion of fluctuation intensity by conduction, and the second measures the rate of entropy production. The quantity, a= (Re) (26.2) &, is the dissipation rate for temperature fluctuations and, in the dis- cussion of the fluctuations, plays the same role as ¢ does for velocity fluctuations. The ‘diffusion’ part of KOV76 is negligible if the Péclét number of the turbulence is large, and then t = k(@6fex,. (2.6.3) 2.7 The relation between the pressure and velocity fields Taking the divergence of the equation of motion (2.4.3) leads to AU tu) AUytuy) _ _O(P+p) ex, ox Gx, O @74) after using the continuity condition (2.3.8). The pressure is so determined by the velocity and temperature fields, and a formal solution of (2.7.1) is 1 Ca Pee glee Ortunuseup) ix, OX, _ G(T +8) aV(x) T x, |x'—x|’ (2.7.2) where x’ is the position at which the pressure is to be found, It should be noticed that pressure is not a local quantity but depends44 TURBULENT SHEAR FLOW on an integral over the entire field of velocity and temperature. The pressure fluctuation, p, is given by 1 OU, du, — 00] d¥(x) ax) = [2 Ixy wae Bes 7 z| ix ae (2.73) as the sum of contributions from three effects: (1) the term in the bracket, 2(@U;/éx,)(éu,/ax,), represents an interaction between the mean velocity gradients and turbulent velocity gradients, (2) the term, (62/dx, 0x,(ua, — iu), represents the effect of fluctuations of Reynolds stress about the mean value and, in shear flows, it is usually much smaller than the first term, (3) the term, — (g/T)(00/2x), represents the induction of pressures in response to the buoyancy forces.CHAPTER 3 HOMOGENEOUS TURBULENT FLOWS 3.1 Introduction Making a distinction between the mean values and the fluctuations of velocity and temperature carries with it an implication that there is a physical difference between the parts of the kinetic energy and entropy densities that are associated respectively with the mean fields and with the fluctuation fields, essentially because transfer from the mean value forms to the fluctuation forms is normally irreversible and is the first stage in a cascade process ending in trans- formation or destruction by molecular transport processes. Both the broad features and the details of the transfer processes are of vital importance for the full understanding of turbulent flow, and the necessary information is obtained most easily from study of homo- geneous turbulent flows. Most turbulent flows are inhomogeneous but the more important features of the energy transfer process are the same whether the flow is homogeneous or not, and the com- parative simplicity of the statistical description of homogeneous flows makes possible experimental and theoretical studies in a detail that is not feasible for inhomogeneous flow. We shall examine some flows which are homogeneous, or nearly so, in the turbulent fluctuations and in the gradients of mean velocity, that is to say, the mean values of functions of the fluctuations and of the mean velocity gradients are independent of position in the flow. In order of increasing complexity, they are: (1) nearly homogeneous and isotropic turbulence with uniform mean velocity; (2) nearly homogeneous but strongly anisotropic turbulence with uniform mean velocity; (3) nearly homogeneous turbulence with uniform gradient of mean velocity.46 TURBULENT SHEAR FLOW 3.2 Eddy interactions in homogeneous turbulence For homogeneous turbulence with uniform gradients of mean velocity, equation (2.4.5) for the turbulent kinetic energy, 497, becomes 05 — dU, u,\? Sug =~, iy) : 2) showing that energy is generated by working of the mean flow against the Reynolds stresses and is dissipated as heat by working of the turbulent velocity gradients against the viscous stresses. Since the Reynolds stress tensor specifies the turbulent energy, the eddies containing most of the energy also contribute most to the Reynolds stresses and presumably receive most of the energy that is transferred from the mean flow. On the other hand, the rate of energy dis- sipation is proportional to the mean square of the velocity gradients which is determined by eddies much smaller than those containing most of the energy. So a simple consideration of the energy budget raises two problems - the nature of the energy transfer from the mean flow to the turbulent eddies and the nature of the transfer from the large energy-containing eddies to the much smaller dis- sipating eddies. Some understanding of the problems, though unfortunately not their solution, comes from examining the equations for the rates of change of Fourier components of the velocity field. In chapter 1, it was shown that an eddy of limited spatial extent can be described by a group of Fourier components with wave numbers of com- parable magnitudes and that, with some caution, conclusions about Fourier components of wave number k can be applied to eddies of ‘size’ k~!. The velocity fluctuations within a large volume V can be expressed as the sum of Fourier components, u(x) = Y afk) exp(ik. x), (3.2.2) where the allowed values of k satisfy cyclic boundary conditions and are distributed uniformly in wave number space with number density (2n)-3V. The condition of incompressibility, du,/x; = 0, requires that kadk) = 0 (3.2.3) and, since u(x) is real, a(k) = a,*(-K). (3.2.4)HOMOGENEOUS TURBULENT FLOWS 47 In homogeneous turbulence, the equation for the velocity fluctuation is ou, OU, tu, ap. Oy > zs => sca 3.2.5) a Gite tHE a ae age? O25) where @p _ , aU, ou, —oF 2S Ox; Ox~ OX, (see 2.3.8, 2.7.1). It follows that the variation with time of a single Fourier component of the fluctuation field is described by an equation for its amplitude, daj(k) _ dr oe (3.2.6) us ate au —vk2a(k)— Fy HOt 5 AW +i com kak’) = 3.2.7) ean! and by an equation for the rate of change of the wave number of the component, dk, dU, an i (3.2.8) The second equation describes the rotation and distortion of the velocity pattern by the mean velocity gradients and, since the divergence of the ‘velocity’ dk,/dt in wave-number space is a (dk) _ _2U, aa) -F-o, (3.2.9) the motion of the allowed wave numbers in wave-number space is solenoidal and the originally constant number-density is preserved. The first three terms in (3.2.7) are linear in the component am- plitude and describe changes in amplitude due to viscous stresses and to interaction between the turbulent motion and the mean flow. The remaining terms are quadratic in the amplitude and describe changes due to interactions between components of different wave numbers, here expressed as a sum over pairs of components whose wave numbers satisfy K+k =k, t 8im is the substitution tensor, =1 if = mand Oif # m.48 TURBULENT SHEAR FLOW Since each Fourier component is a plane wave pattern with motion at right angles to the wave normal, the scalar amplitude (a,a,*")"/? and the direction and ellipticity of the polarisation determine the amplitude. Using (3.2.7) to form an equation for the square of the scalar amplitude, we find - raat Woat+aia) xy i ky DE alka (k)a,"(k)—a,*(k)a;*(k")a(k)], (3.2.10) noticing that the net contributions of the terms Fe an) +15 kPa kay") d qn) in (3.2.7) vanish since k,a(k) = am = 0. The implication is that these terms, which describe the changes in a, caused by fluid acceleration by pressure gradients, alter the polarisation of the component without affecting its scalar amplitude. It does not follow that the interactions causing changes in scalar amplitude, i.e. in kinetic energy, are physically distinct and separable from those causing changes in polarisation. For example, the second term on the right of (3.2.10) may be written in the form _oU, aa," +a a “ey aa where (aa,* + a,*a))/(a,a,*) is a dimensionless quantity specifying the polarisation, and we see that the rate of variation of aa," depends on the scalar product of the polarisation tensor and the mean velocity gradient tensor. The quantity, 4(aa;* + a,*a,), averaged over all realisations of the flow, is the mean contribution of the component with instan- taneous wave number k to the Reynolds stress ju), and, since the distortion in wave-number space described by (3.2.8) preserves the original density of allowed wave numbers, the three-dimensional spectrum function is Ok) = —Xa,a,*+0,%a,), (3.2.11) im where the angle brackets signify an ensemble average over all realisations, It is not difficult to derive equations for the variationsHOMOGENEOUS TURBULENT FLOWS 49 of ©,, using (3.2.7), but the most interesting one is the equation for © ,,(&), the spectrum function for the mean square velocity fluctu- ation, satisfying J O(ke) dk = 9? Ttis @ aya au, HOM) = ~21P 4K) -2 51 Oth) UV, 20,(8)_ 8 th ae ae 5M) 8212) and the terms on the right represent (1) loss by direct viscous dissipation, (2) energy transfer from the mean flow to components of wave number k, (3) redistribution of energy in wave-number space consequent on distortion of the velocity pattern, and (4) redis- tribution of energy by non-linear interactions between different components. To emphasise the essentially conservative nature of the energy transfer between components, the last term has been written as the divergence in wave-number space of S,(k), the flux vector of the total intensity g?. One interpretation of the spectrum equation (3.2.12) is that turbulent energy flows in wave-number space from the region of comparatively small wave numbers, where most of the energy is produced and resides, towards much larger wave numbers where the rate of viscous dissipation is sufficient to convert the flow to heat. A fundamental problem of turbulent motion is the relation between the flow of energy in wave-number space and the energy distribution in that space. No satisfactory solution for the region containing most of the turbulent energy has been found, but conditions are simpler for the larger wave numbers which contain only a small part of the total energy but which include nearly all the components involved in the process of viscous conversion of energy to heat. 3.3. Experimental approximations to homogeneous turbulence In truly homogeneous flow, all the mean values in the full equation for the turbulent energy (2.4.5) are independent of position in the50 TURBULENT SHEAR FLOW flow and the equation reduces to the simple form (3.2.1). Unless production of turbulent energy by the Reynolds stresses happens to be exactly equal to the rate of dissipation, the energy varies with time and the flow is not stationary. Even if the practical problems of initiating spatially homogeneous turbulent motion could be over~ come, the transience of the flow would make measurements very tedious and all experimental studies of ‘homogeneous turbulence’ have used flows which are stationary with respect to time but necessarily inhomogeneous in one direction. These flows are pro- duced by placing grids of suitable forms across the entrance to ducts of constant or varying section, and the turbulent motion near any cross-section of the flow is assumed to be similar to that in a truly homogeneous flow which has existed for a time equal to the ‘time- of-flight’ from the grid to the particular section, _ fen! ula ean where U, is the local mean velocity. In flows with gradients of mean velocity, a clear definition of t,is possible only if the lateral variations of U, are small. The equivalence of the stationary grid flow and the theoretical homogeneous flow depends on the possibility of defining a volume that moves with the local mean velocity and that has dimensions small compared with the scale of the streamwise inhomogeneity but large compared with the size of the turbulent eddies. Then the motion Within the volume interacts with gradients of mean velocity and with a turbulent motion outside that are much the same as they would be in a homogeneous, non-stationary flow after the decay time f, determined by its present position. The condition for equivalence is that the scale of inhomogeneity, say 3.3.2) should be large compared with the size of the energy-containing eddies, specified by an integral scale Ly = i, Ry (1,0, 0) dr. (3.3)HOMOGENEOUS TURBULENT FLOWS 5 As we shall see, the rate of turbulent energy dissipation, = (my P= Nex)? is approximately = GPL 3.4) and substitution in the energy equation for stationary grid turbulence, aU, 25 U ax, dtu, 5x, tix, 4 uy + Puy) ox = Tay \2 x00) ) » G35) : ; leads to the relation +terms of order (q)"/2/U, or less. (3.3.6) For equivalence, it is necessary that () @? «U,; and ie. both the turbulent velocity fluctuations and the variation of mean velocity over an eddy ‘diameter’ should be small compared with the local mean velocity. 3.4 Isotropic turbulence: general The study of isotropic turbulence began in 1935 when G. I. Taylor defined it by the condition that all mean values of functions of the flow variables should be independent of translation, rotation and reflexion of the axes of reference. It is the simplest form of turbulence that is relevant to ‘complete’ turbulent flows such as jets and boundary layers, but it still presents unsolved problems in spite of intensive theoretical and experimental study. The importance of the theory of isotropic turbulence lay in the demonstration that it is 352 TURBULENT SHEAR FLOW possible to derive from the equations of motion and continuity relations between mean values connected with the intensity and scale of the turbulence, mean values whose measurement had be- come possible through the previous development of the hot-wire anemometer by H. L. Dryden and A. M. Kuethe. The presently accepted view that a satisfactory theory of turbulent flow must be based on an adequate and realistic account of the turbulent motion may be attributed equally to the development of the theory and of the hot-wire techniques. A comprehensive account of the present knowledge of isotropic turbulence would occupy considerable space and we will consider mostly features that are common to all tur- bulent motion. For more detail and, indeed, for an exact statement of much of the following, reference should be made to the literature (in particular, Batchelor 1953). The usual experimental approximation to isotropic turbulence is produced by placing a uniform grid across the entrance to the working section of a wind tunnel. The grids most commonly used are of the ‘biplane’ type, with two layers of uniformly spaced, circular cylinders with axes in the two layers at right angles. The grid is specified by the mesh length M, the separation between the axes of adjacent cylinders, and by d the diameter of the cylinders. The ratio M/d is usually in the range 4-6. Within a few mesh lengths of the grid, the individual wakes of the cylinders merge and the flow becomes statistically homogeneous over planes at right angles to the direction of mean flow along the Ox, axis.t Grid turbulence produced in this way is noticeably anisotropic, the ratio of the turbulent intensity in the stream direction to that normal to the stream direction being approximately 1.25. Comte- Bellot & Corrsin (1966) have shown that the intensities may be made equal by passing the grid turbulence through a 1.27:1 con- traction, but most of the experimental work has been analysed assuming complete isotropy. The consequent errors are unlikely to affect the conclusions drawn here. ‘The special feature of isotropic turbulence is the simplicity of its specification. By definition, the correlation tensor, R,,(x; ), has a + If the flow resistance of the grid is too large (too small values of Mjd), flow through it is not uniform and persistent lateral variations of turbulent intensity are found (Grant & Nisbet 1957, Bradshaw 1965).HOMOGENEOUS TURBULENT FLOWS 53 form independent of the position and orientation of the axes of reference, i.c, Ry = rjAQ) + 5yBO, 4.1) where A(r) and B(r) are functions of the scalar separation r. The condition of continuity for R,,, that a = 3 =0, a, RyQs 0) leads to a differential equation connecting the two functions, and the whole correlation tensor may be expressed in terms of a single scalar function of r. The usual choice of function is the longitudinal correlation function, I(r) = Rule, 0, 0/Ri10, 0, 0) (4.2) and then Ry) = m[- FE po Ast arf | . (3.43) In a similar way, the three-dimensional spectrum function can be expressed in terms of the (scalar) integrated spectrum function defined in § 1.5, Ok) = aa uk? — kik )E11(k)- (3.4.4) The equations for the rate of change of the correlation and spectrum functions include third-order tensors describing the non- linear interactions between components of the motion. The triple velocity correlation tensor is Tyne) = ux) + 8) and may also be expressed in terms of a single scalar function of the separation, conventionally We) = UPR, Xa 8 + XM? 3.4.5) 3.5 Reynolds number similarity in isotropic turbulence Perhaps the most significant fact about turbulent flows is that, while geometrically similar flows are expected to be dynamically and structurally similar if their Reynolds numbers are the same, their