-
Some solutions to exerciseswith apologies for any mistakes
PART I
Chapter 1
Exercise 1.1, Origin of the closure problem:The closure problem
arises in any non-linear system for which one attempts to derive
anequation for the average value. Let ξ correspond to the result of
coin tossing, as in the text,and let
u3 + 3u2 + 3u = (u + 1)3 − 1 = 7ξShow that if u3 = u3 and u2 =
u2 were correct, then u = (9/2)1/3 − 1. Show that the correctvalue
is u = 1/2. Explain why these differ, and how this illustrates the
‘closure problem’.
Solution to Ex. 1.1, If u3 = u3, averaging the equation would
give
u3 + 3u2 + 3u = (u + 1)3 − 1 = 7ξ = 7/2
so u = (9/2)1/3 − 1. However, the equation has the exact
solution u = (7ξ + 1)1/3 − 1. Sinceξ = 1 with probability 1/2 , and
ξ = 0 with probability 1/2 ,
u = (81/3 − 1) × 1/2 + (0) × 1/2 = 1/2
Exercise 1.2, Eddies:Identify what you would consider to be
large and small scale eddies in photographs 1.4 and1.7.
Solution to Ex. 1.2, Descriptive
Exercise 1.3, Turbulence in practice:Discuss practical
situations where turbulent flows might be unwanted or even an
advantage.Why do you think golf balls have dimples?
1
-
Chapter 2
Exercise 2.1, Dissipation range scaling:
In the legend of figure 2.1 let Rλ = R1/2T . Assuming that the
energetic range begins where
the data leave the −5/3 line, do these data roughly confirm the
RT scaling of η/L?Solution to Ex. 2.1,
η = (ν3/ε)1/4; η/L = (ν3/L4ε)1/4 = (ν3/k3/2L3)1/4 = R−3/4T
From the figure, at Rλ = 600, ηκ ≈ 7 × 10−4 and RT = 3.6 × 105.
At Rλ = 1, 500,ηκ ≈ 2 × 10−4 and RT = 2.25 × 106. Assume κ ∝ 1/L;
the data give
(η/L)600(η/L)1,500
= 7/2 = 3.5
while the dimensional analysis says this should be proportional
to R−3/4T .(
(RT )1500(RT )600
)3/4= (22.5/3.6)3/4 = 3.95.
Good, order of magnitude, confirmation of the scaling.
Exercise 2.2, DNS:One application of dimensional analysis is to
estimating the computer requirements for DirectNumerical Simulation
of turbulence. The computational mesh must be fine enough to
resolvethe smallest eddies, and contain enough points to resolve
the largest. Explain why thisimplies that the number of grid
points, N , scales as N ∝ (L/η)3 in 3-dimensions. Obtainthe
exponent in N ∼ RnT . Estimate the number of grid points needed
when RT = 104.
Solution to Ex. 2.2, L/η ∝ R3/4T −→ N ∼ R9/4T When Rt = 104, N ∼
109.Exercise 2.3, Relative dispersion:The inertial range velocity
(εr)1/3 can be described as the velocity at which two fluid
elementsthat are separated by distance r move apart (provided their
separation is in the inertial range,η � r � L). Deduce the power
law for the time-dependence of the mean square separationr2(t). Use
dimensional analysis. Also infer the result by integrating an
ordinary differentialequation. The scaling v2 ∝ r2/3 is often
called Richardson’s 2/3-law.
Solution to Ex. 2.3, Solution by dimensional analysis: r2 ∝
εt3.A solution by integrating an o.d.e.: From Kolmogoroff’s
2/3-law
dtr2 ∝ ε2/3r22/3
The solution to this equation is r21/3 ∝ ε2/3t + r21/30 . If
ε2/3t >> r2
1/3
0 then r2 ∝ εt3.
2
-
Exercise 2.4, Averaging via p.d.f.’s:Let the p.d.f. of a random
variable be given by the function
P (x) = Ax2e−x for the range ∞ > x ≥ 0P (x) = 0 for x <
0.
What is the value of the normalization constant A? Evaluate
sin(ax) where a is a constant.
Solution to Ex. 2.4, A = 1/2.
sin(ax) = 1/2Im
∫ ∞0
x2e−x(1−ia)dx = Im[(1 + ia)3/(1 + a2)3] =3a − a3(1 + a2)3
Exercise 2.5, Taylor microscale:Let u be a statistically
homogeneous function of x. A microscale, λ is defined by
limξ→0
d2ξ
[u(x)u(x + ξ)
]= −u
2
λ2,
if the correlation function is twice differentiable. Show that
u2/λ2 = (dxu)2. Assume that(dxu)2 follows dissipation range scaling
and obtain the RT dependence of λ/L.
Solution to Ex. 2.5,
−u2
λ2= u(x)d2ξu(x + ξ)|ξ=0 = ud2xu = dx(udxu) − dxudxu
= dx(udxu) − (dxu)2 = −(dxu)2
because homogeneity implies dx(any average) = 0. In dissipation
range (dxu)2 ∼ ε/ν leadsto
λ/L ∼ R−1/2T
Exercise 2.6, Toor’s analogy, (2.2.13):To what does the term
‘analogy’ refer?
The p.d.f. P (m) = (1 + m)a−1(1 − m)b−1/B(a, b), for −1 < m
< 1 and P (m) = 0 for|m| ≥ 1 is referred to as the ‘beta’
probability density. The normalization coefficient isB(a, b) =
2a+b−1Γ(a)Γ(b)/Γ(a+b) where Γ is the factorial function Γ(a) =
(a−1)!, extendedto non-integer arguments.
The beta distribution is a popular model for the mixture
fraction p.d.f. in reacting flows.Given that Γ(1/2) =
√π, evaluate γA for a = b = 1/2, a = b = 2 and a = b = 4.
Plot
the β-distribution for these same three cases. Based on these
cases, describe how the p.d.f.evolves as mixing and reaction
proceed.
Solution to Ex. 2.6, The analogy is between non-reactive scalar
mixing and reactantconsumption. The elegance is that one can
predict properties of reacting flow withoutexplicitly analyzing the
reaction process.
3
-
Mean reactant concentrations (normalized by initial value):
γA = Γ(2a)/Γ(a)Γ(a + 1)22a in general for a = b
γA = 1/π for a = b = 1/2
γA = 3/16 for a = b = 2
γA = 35/256 for a = b = 4
When a = b the variance is proportional to1/1+a+b, per p.22. As
a increases, variancedecreases, in accord with time-evolution.
m
p(m
)
Beta distributions
-1.0 -0.5 0 0.5 1.00
0.5
1.0
1.5
2.0 a=b=2 a=b=4 a=b=1/2
Exercise 2.7, Langevin equation:Why does the expansion (2.2.20)
begin with 1 and why is the next term negative?
In (2.2.16) and subsequent equations let s2 = (1 − r2)σ2. Derive
the equation
1/2dtu2 = −u2
TL+
σ2
TL.
from (2.2.18). Solve for u2(t) with initial condition u2(0) = 0,
σ and TL being constants.Is u2 a statistically stationary random
variable? The exact solution for the variance that isderived here
will be used to test a Monte-Carlo simulation in the next
exercise.
Solution to Ex. 2.7, r(0) = 1 by (2.15), r is a correlation
function so |r| ≤ 1. Solutionis u2 = σ2(1 − e−2t/TL); this is not
statistically stationary because u2 is a function of time.Exercise
2.8, More on stochastic processes:Most computer libraries have a
random number algorithm that generates values in the range{0, 1}
with equal probability; i.e. P (ũ) = 1, 0 ≤ ũ ≤ 1.
Show that ũ = 1/2 and that u2 = 1/12 where u is the fluctuation
ũ − ũ. Deduce thatξ ≡ √12 (ũ − 1/2) has ξ = 0 and ξ2 = 1, as is
needed in the model (2.2.16). A Gaussianrandom variable can be
approximated by summing N values ξi and normalizing by
√N . For
N = 16
ξg = 1/4
16∑1
ξi.
Program this and verify by averaging a large number (10,000 say)
of values that ξg ≈ 0 andξ2g ≈ 1. Let s in (2.2.16) be as in
exercise 2.7, normalize t by TL and u by σ and solve
(2.2.16)numerically starting with u = 0 and integrating to t/TL =
10 by steps of 0.05. Computeu2(t) by averaging 100 such solutions,
by averaging 1,000 such solutions and by averaging4,000 such
solutions. Plot these estimates of u2 vs. t and compare to the
exact result foundin exercise 2.7. Does it look like the average is
converging like 1/
√N?
4
-
Solution to Ex. 2.8, ũ =∫ 1
0ũdũ = 1/2; u2 =
∫ 10(ũ − ũ)2dũ = ∫ 1
0(ũ − 1/2)2dũ = 1/12
t
u2(t
)
0 2.5 5.0 7.5 10.00
0.25
0.50
0.75
1.00
1.25
Analytical 4000 samples 1000 samples 100 samples
Langevin equation solutions.
Exercise 2.9, Isotropy:Verify that (2.3.35) is the most general
fourth order, isotropic tensor.
Exercise 2.10, Solving equations via Cayley-Hamilton:Use the
Cayley-Hamilton theorem to solve
bij = Sikbkj − 1/3δijSklblk + Sijin which S is a given,
trace-free matrix (Skk = 0). Why is b also trace-free? This
solution isonly valid if S is such that the solution is not
infinite; what is this solvability criterion?
Solution to Ex. 2.10, In matrix form, solve
b = S · b − 1/3Trace(S · b)I + S
Let b = AS + B(S2 − 1/3IS2) where S2 = Trace(S2). Substituting
this into the equationgives
AS + B(S2 − 1/3IS2) = AS2 + B(S3 − 1/3SS2) − 1/3(AS2 + BS3)I +
S
Cayley-Hamilton is S3 = 1/3IS3 + 1/2S2S. Equating coefficients
of S2 gives B = A.
Equating coefficients of S gives A = 1/(1 − 1/6S2). So
b =S + S2 − 1/3IS2
1 − 1/6S2
and the solvability condition is S2 �= 6.
5
-
Exercise 2.11, Solving equations via generalized
Cayley-Hamilton:τ is the solution to
τ = τ · ΩΩ − ΩΩ · τ + Trace(S · τ )δ − Sin which S is symmetric
and trace-free and ΩΩ is antisymmetric and trace-free. Show that
τis symmetric. Use (2.3.46) to solve this equation in two
dimensions.
6
-
Chapter 3
Exercise 3.1, Derivation of Reynolds stress transport
equation:This exercise may seem laborious, but it is a good
introduction to the use of Reynoldsaveraging, and to the Reynolds
averaged Navier-Stokes (RANS) equations.
Derive (3.1.4) from (3.1.3) and (3.1.2) and then obtain (3.1.5).
Symbolically the stepsare
RSij = uj
[NS(U + u)i − NS(U + u)i
]+ ui
[NS(U + u)j − NS(U + u)j
]where ‘NS’ represents the Navier-Stokes equations. How many
equations and how manyunknowns are there? Why not also form the
moment of the continuity equation?
Solution to Ex. 3.1, uj∂iui = 0 introduces a new unknown vector
uj∂iui that is un-related to other unknowns in the RANS equation:
so the continuity equation is simplyirrelevant to the single point,
averaged equations. 6 equations, 28 unknowns (given U) – 22more
unknowns than equations.
Exercise 3.2, Production of k:Using (3.1.3) show that the rate
at which mean energy (per unit mass) 1/2UiUi is lost tothe
turbulence is uiuk∂kUi = −P . (Conservation terms are not an
‘energy loss’.) Thisdemonstrates that the term ‘production’ is
actually referring to the transfer of energy fromthe mean flow to
the turbulence, and not to a net source of energy.
Solution to Ex. 3.2,
1/2DtU2 = ∂k(UkP ) − Ui∂kuiuk
= ∂k (UkP − Uiuiuk) + uiuk∂kUior, in vector notation
1/2DtU2 = ∇ · (U P − U · uu) − Pk
The first term on the right is in conservation form, and simply
transports energy. Pk is theenergy lost to (or gained from)
turbulence.
Exercise 3.3, The mixing length rationale:Consider the mean
concentration C(xi) of a passive quantity that is convected by a
turbulentvelocity vector uj(t). Derive an eddy diffusion formula
analogous to (3.2.9). The eddydiffusivity should come out as a
second order tensor, i.e., its a matrix of components. Isthis
matrix symmetric in general? How about when the turbulence is
statistically stationary(see the discussion of stationarity above
(2.2.16))? Give your reasoning in answer to thesequestions.
Solution to Ex. 3.3,
uic ≈ uij∂jC =∫ t
0
ui(t)uj(t′) dt′∂jC
7
-
The eddy diffusion tensor is defined by uic = κTij∂jC, so
κTij =
∫ t0
ui(t)uj(t′) dt′
In general κTij �= κTji because ui is evaluated at the later
time t and uj at the earlier time t′in κTij , while uj is at the
later and ui the earlier time in κTji . However, if the turbulence
is
statistically stationary ui(t)uj(t′) is a function only of |t −
t′| and
κTij =
∫ t0
uiuj(|t − t′|) dt′ = κTji
so it doesn’t matter which component is earlier and which is
later.
Exercise 3.4, Anisotropy equation:The Reynolds stress anisotropy
tensor is defined as bij = uiuj/k − 2/3δij. Using (3.2.15)and
(3.2.16) derive the evolution equation for bij in homogeneous
turbulence. The equationshould involve bij , ∂jUi, φij, εij and k,
with no explicit or implicit dependence on uiuj .
‘Isotropy’ means complete lack of any directional preference.
Hence the identity matrix[δij ] is isotropic because all the
diagonal components are equal; or more correctly, becauseif the
coordinate system were rotated, the identity matrix would remain
unchanged. Thetensor bij measures the departure of uiuj from
isotropy.
Solution to Ex. 3.4,
ḃij = −φij + εijk
− 2/3(∂jUi + ∂iUj)−bik∂jUk − bjk∂iUk − (bij + 2/3δij)(P − ε)
where P = −bkl∂kUl.
8
-
Chapter 4
Exercise 4.1, Turbulent kinetic energy equation in the
channel:Write the form of the turbulent kinetic energy equation
(3.1.6) for this special case of parallelflow: U(y), homogeneity in
x, z and stationarity in t. Show that∫ 2H
0
Pdy =∫ 2H
0
εdy.
What boundary condition on k is did you use, and why? Recall
that P is the rate at whichturbulent energy is produced from the
mean shear. This exercise suggests that in shearlayers, production
and dissipation are similar in magnitude.
Solution to Ex. 4.1, The boundary condition ∂yk = 0 = k follows
from the definitionk = 1/2u2 + v2 + w2 and the no-slip condition u
= v = w = 0: k has a quadratic zero.
Exercise 4.2, Magnitude of Cf :Let RH = 10
4. Use formula (4.1.11) to estimate the value of Cf .
Solution to Ex. 4.2, A printout of√
(2/Cf) versus 2.43 log(104√
Cf/2)+5 gives Cf =5.7 × 10−3.Exercise 4.3, Pressure drop across
a channel:Because of the favorable pressure gradient, the log-law
can be considered valid practically tothe center of a fully
developed, plane channel flow. The centerline velocity, UCL , and
volumeflux, Q, are related by
Q ≈ 0.87UCLAwhere A is the cross-sectional area of the channel.
Its height is 2H . The channel is longand narrow, so the sidewalls
can be ignored. A certain volume flux, Q, of airflow is
desired.Write a formula that can be used to estimate the necessary
pressure drop across a channelof length L — do not try to solve it
for ∆P ; that could be done numerically.
Solution to Ex. 4.3,From momentum conservation
ρu2∗ = 2H∆P/L
Drag formula
u∗ =UCLκ
log(u∗H/2) + Bκ=
Qκ
0.87A[log(u∗H/2ν) + Bκ]
Squaring, and using κ = 0.41 and B = 5.0
u2∗ =0.22Q2
A2[ 1/2 log(u2∗H2/4ν2) + 2.05]2
∆P
ρ=
0.44Q2L
HA2[log(∆PH3/2µνL) + 4.1]2
9
-
Exercise 4.4, Zones of the boundary layer:Make a sketch of U(y)
as in figure 4.5. Indicate the law-of-the-wall and
law-of-the-wakeregions in both log-linear and linear-linear
coordinates. In zero pressure gradient, what isthe approximate
magnitude of U/U∞ at the top of the log-layer? What is it for the
adversepressure gradient data?
Solution to Ex. 4.4, 0.75 on ZPG. 0.5 APG
Exercise 4.5, The momentum integral:Derive the momentum integral
equation (4.2.20) for the case where dxU∞ �= 0. Substituteδ∗ = Hθ.
Will a favorable pressure gradient (dxU∞ > 0 and dxP < 0)
increase or decreasethe growth rate of θ? Answer the same question
for an adverse pressure gradient.
Exercise 4.6, Law of the wake:Infer a drag law by evaluating
(4.2.16) at y = δ99. Use Coles’ wake function for w(y). SetU99 = U∞
and compare this to (4.2.15). Comment on the effect of pressure
gradient on skinfriction.
Exercise 4.7, Drag law:The power law form U = Ud(y/d)
a, y < d is sometimes used to fit the mean flow profile. dis
the boundary layer edge, so Ud = U∞ and U = U∞, y ≥ d. The exponent
a is a smallnumber, about 1/7, although it varies with Reynolds
number, as you will show. Calculateθ, δ∗ and H = δ/θ for this
profile. For small a, (y/d)a = ea log(y/d) ≈ 1 + a log(y/d). If
thisis matched to the log-law (4.1.7), how is a related to u∗/U∞?
What is the correspondingskin friction law, written like (4.1.11)
in terms of friction coefficient, Cf , and momentumthickness
Reynolds number, Rθ = Udθ/ν?
Solution to Ex. 4.7,
Θ = ad/(1 + 3a + 2a2) → d = Θ(1 + 3a + 2a2)/a
Comparing U = U∞ + aU∞ log(y/d) to the log-law (4.1.7)
U =u∗κ
log(y+) + Bu∗
gives a = u∗/κU∞ =√
Cf/2/κ and U∞ + u∗/κ log(ν/u∗d) = Bu∗. The latter can
berearranged to
U∞u∗
=
√2
Cf=
1
κlog(u∗d/ν) + B =
1
κlog
[κRθ(1 + 3a + 2a
2)]+ B
Alternatively, a depends on Reynolds number via
1/a = log[κRθ(1 + 3a + 2a
2)]+ Bκ
10
-
Exercise 4.8, Measurement of u∗:Measurements in a particular
turbulent boundary layer give a thickness δ99 = 3.2cm. Thevelocity
U is measured at three heights with the results, U = 40.1m/s at y =
2.05cm,U = 34.7m/s at y = 0.59cm and U = 30.7m/s at y = 0.22cm.
i Estimate the friction velocity u∗.ii Let the fluid be air with
ν = 0.15cm2/s. Is this an adverse or favorable pressure
gradient boundary layer?
Solution to Ex. 4.8,The lower two points are in the log-layer;
hence u∗ log(y2/y1)/κ = U1 − U2 =⇒ u∗ =
4. × 0.41/ log(.59/.22) = 1.66 m/s. The lowest point is at y+ =
243 which satisfies thecriterion 40 < y+ < 0.2δ99.
At y = 2.05cm, y+ = 2269, u∗(log y+/κ + 5.1) = 39.75m/s. Hence
the law-of-the wakecorrection is (40.1 − 39.75)/1.66 = 0.21. The
wake correction is 2Π/κ sin(π/2(2.05/3.2))2 =4.32Π. Hence Π(β) =
0.21/4.32 = 0.049. The empirical formula (4.2.17) in the text
givesβ = −0.48. Since β < 0 the pressure gradient is
favorable.Exercise 4.9, Governing equations for self-similar
flow:Substitute the assumed forms
U = ∆U(x)f(y/δ(x)); V = ∆U(x)g(y/δ(x)); uv = ∆U2h(y/δ)
for the case of a 2-D jet into (4.3.25) using (4.3.31) but
ignore (4.3.34). Let ζ = y/δ. Rewritethe equations that you derived
as ordinary differential equations involving f(ζ), g(ζ) andh(ζ). In
order for the independent variable to be ζ alone, the coefficients
in this equationcannot be functions of x or y — because then the
solution would be of the form f(ζ, x) orf(ζ, y), which are not
self-similar. What condition must be imposed on ∆U and dxδ for
thisto be the case? Do you think that all the results (4.3.34) can
be derived by this approach?
Solution to Ex. 4.9, Substitution into momentum equation
gives
δ
∆U
d∆U
dxf 2 − dδ
dxηff ′ + gf ′ = −h′
For similarity, the coefficients must have proportionate x
dependence: in this case they mustbe constant:
dδ
dx= const. :
δ
∆U
d∆U
dx= const.
The former gives δ ∝ x; the latter would be satisfied for ∆U ∝
xn, for any n. n = −1/2 isobtained from the momentum integral.
All scaling can be derived by this more formal approach. In the
case of the wake it showsthat similarity can only be approximate,
under a small deficit assumption.
Exercise 4.10, The momentumless wake:For a zero momentum
deficit, plane wake, the constant in (4.3.30) is zero. I.e., in the
smalldeficit approximation, U∞
∫∞−∞(U − U∞)dy = 0.
11
-
Use the eddy viscosity approximation −uv = νT ∂yU ; assume νT to
be constant; usethe small defect approximation for the convection
velocity; and form the second moment of(4.3.28) with respect to y2,
after making these approximations, to find the similarity scalingin
this case. (Form the second moment means multiply by y2, then
integrate between ±∞).
Solution to Ex. 4.10, Consider∫ ∞−∞
y2[∂xU(U − U∞) + ∂yV (U − U∞)dy =∫ ∞−∞
y2νT ∂2y(U − U∞)dy
In the present case of zero momentum flux the right side
integrates to 0,∫ ∞−∞
y2νT ∂2y(U − U∞)dy = 2νT
∫ ∞−∞
(U − U∞)dy = 0.
Requiring U = U∞−udf(η) and ud
-
One integration gives
−a2f
∫fdη = cf ′
Let∫
fdη = h; note that h′ → 0 as η → ∞.
− 1/2ah′h = ch′′
Integrate:− 1/4a(h2 − h2∞) = ch′.
Integrate one last time, or intuit that the solution is of the
form h = 1/k tanh(kη) (orf = sech2(kη))
−a4(tanh2 − 1) = ck2sech2.
The functions cancel, leaving k =√
a/4c.The solution is
U/UCL = sech2η√
a/4c
The 50% thickness is defined such that at η = 1, 0.5 =
sech2(√
a/4c). Hence√a/4c = 0.8813 and a = 0.05 −→ c = .016
Exercise 4.12, Enthalpy thickness:Show that the enthalpy
thickness evolves according to
d∆
dx= St
in analogy to the evolution equation (4.2.23) for momentum
thickness. Compare the evo-lution of ∆(x) to that of Θ(x) for a
zero pressure gradient boundary layer in air over therange 103 <
Rθ < 10
4. The comparison can be in the form of computed curves using a
datacorrelation for Cf .
Exercise 4.13, Reflected plume model:Because the equation of a
passive scalar is linear, superposition is applicable. The
meanconcentration of a sum of sources is the sum of their
individual mean concentrations. TheGaussian cloud
C =Q√
(2π)3X2 Y 2exp
[− x
2
2X2− (y − ys)
2
2Y 2
]is the mean concentration of a line source located at x = 0, y
= ys. If a no-flux wall exists aty = 0, a fictitous source can be
added at y = −ys to satisfy the boundary condition ∂yC = 0on y = 0.
Write the concentration distribution for this ‘reflected plume’
model. Show thatthe concentration integrated across the physical
doman, which is y > 0, −∞ < x < ∞, is
13
-
constant in time. There is no mean flow, just turbulence, so dtX
= u and dtY = v. Theheight of the plume centerline is defined
as
YL(t) =
∫ ∞−∞
∫ ∞0
yC(y)dy dx/∫ ∞
−∞
∫ ∞0
C(y)dy dx.
Show that this increases with time, even though there is no mean
velocity in the y-direction.For the exponential correlation
analyzed in the text find the ‘Lagrangian mean velocity’,dtYL when
ys = 0. Why is the Lagrangian mean velocity not zero, even though
there is noEulerian mean velocity?
Solution to Ex. 4.13,
C =Q√
(2π)3X2 Y 2exp
−x22X2
[exp
−(y − ys)22Y 2
+ exp−(y + ys)2
2Y 2
]; y > 0
Integral is ∫ ∞0
exp−(y − ys)2
2Y 2+ exp
−(y + ys)22Y 2
dy
=
∫ ∞0
exp−(y − ys)2
2Y 2dy +
∫ 0−∞
exp−(y − ys)2
2Y 2dy =
√2πY 2
after changing y to −y in the second integral. x-integral
is√
2πX2 so area integral isconstant.
In for non-zero ys, YL is expressed in incomplete
Gamma-functions. However, for ys = 0
YL =
∫ ∞0
exp−y22Y 2
ydy/∫ ∞
0
exp−y22Y 2
dy =
√2Y 2/π.
and (4.4.52) can be substituted to find YL(t).
14
-
Chapter 5
Exercise 5.1, Useful mathematical relation:Prove that εilpεjnp =
(δijδln − δinδlj), as used in the analysis of (9.2.31). The brute
forceapproach is to write out all 81 components. However, all but
12 are zero. Use this identityto show that ν|ω|2 = ε in
homogeneous, incompressible turbulence. The vorticity vector isωi =
εijk∂juk.
Solution to Ex. 5.1,
ν|ω|2 = ν�pil�pjn∂iul∂jun = ν(δijδln − δinδjl)∂iul∂jun=
ν∂iul∂iul − ∂iul∂lui = ε − ∂i(ul∂lui) = ε
Since ∂i(ul∂lui) = 0 due to homogeneity. So the dissipation and
vorticity spectra are inti-mately connected. |ω|2 is often called
‘enstrophy’.Exercise 5.2, Vorticity and dissipation:Suppose that
the scale of non-homogeneity is δ ∼ L. Show the magnitude of the
first termon the right of (5.2.3) relative to the second is O(RT
).
Exercise 5.3, The balance between stretching and diffusion:Show
by dimensional analysis of the stretching (ωj∂jui) and diffusion
(ν∇2ωi) terms of thevorticity equation that a balance is reached
when the vortex radius is O(η). I.e., if λ is theradius then λ ∼ η.
Note that the rate of strain at small scales is of order √ε/ν.
Solution to Ex. 5.3, ωj∂jui ∼ ω√
ε/ν ≈ ν∇2ωi ∼ νω/λ2. So λ2 ∼√
ν3/ε = η2
Exercise 5.4, Introduction to RDT:Find a solution of the
form
ω3 = A(t)ei(k1(t)x1+k2x2)
to the inviscid vorticity equation if the velocity is of the
form u3 = αx3, u2 = v2 andu1 = v1 − αx1 and ∇ · v = 0. The initial
condition is ω3(0) = ω0ei(κ1x1+κ2x2). Because v isthe velocity
generated by the vorticity, it too is proportional to
ei(k1(t)x1+k2x2).
Solution to Ex. 5.4, k̇1 = αk1; Ȧ = αA. v2 = k1ω3/|k|2; v1 =
−k2ω3/|k|2Exercise 5.5, Burger’s vortex:Solve the two dimensional
vorticity equation corresponding to (5.2.5) for ωx(y) when
thestraining flow is U = αx and V = −αy. What is W (y)? Why is this
an exact solution tothe Navier-Stokes equations?
Solution to Ex. 5.5,ωx = ωCLe
−αy2/2ν
W = W−∞ + ωCL
∫ y−∞
e−αy′2/2νdy′
15
-
Chapter 6
PART II
Exercise 6.1, Integral equation closure:Write a program to solve
the model (6.1.1) through (6.1.3). Plot solutions for θ(x/a),H(x/a)
and Cf(x/a) with U∞/U0 = (1 − x/a), where a is a characteristic
length and U0 acharacteristic velocity. Start with Cf (0) = 3 ×
10−3, H(0) = 1.35 and Ra = 105. (Whatis θ(0)/a?) Stop the
computation when the boundary layer separates; at what x/a
doesseparation occur? Also plot these variables with U∞/U0 = (1 +
x/a) in the range 0 ≤ x/a ≤0.5. Discuss the effect of favorable and
adverse pressure gradients on Cf and H .
Solution to Ex. 6.1, Separation at x = 0.227. APG causes Cf to
drop; FPG increasesit. Separation criterion is H = 3.
x
APG
0 0.05 0.10 0.15 0.20 0.250
1
2
310ThH1000Cf
x
FPG
0 0.1 0.2 0.3 0.4 0.50
2
4
610ThH1000Cf
Exercise 6.2, Golf ball revisited:
16
-
In chapter 1, exercise 1.3 asked why golfballs have dimples. An
extension to fig-ure 6.2 provides an answer: the dimplespromote
transition. Provide a qualitativeexplanation of the figure at
right.
Reynolds number
smooth sphere
golf ball
Dra
g co
effic
ient
Exercise 6.3, The mixing length model:Show that if m = κy then
the mixing length formula (6.1.6) gives the log-law in the
constantstress layer (−uv = u2∗). Johnson & King (1985)
suggested using νT = u∗m, which gives thesame result in a constant
stress layer. Compare the mean velocity obtained from these
twoformulations in a linear stress layer, −uv = u2∗ + αy.
Solution to Ex. 6.3, In the constant stress layer (mdyU)2 = u2∗;
hence dyU = u∗/κy,
which gives the log-law when integrated. For JK: u∗mdyU = u2∗,
which gives the same result.In a linear stress layer, JK gives
u∗κydyU = u2∗ + αy, or
U =u∗κ
log y+ +α
u∗κy + Bu∗
where the integration constant was chosen to recover the
standard log-law as α → 0. Themixing length model gives
U = u∗
∫ √1 + αy/u2∗
κydy
=u∗κ
(2√
1 + αy/u2∗ + log
[√1 + αy/u2∗ − 1√1 + αy/u2∗ + 1
])+ u∗const.
To obtain the constant, let α → 0; the above becomes
U → u∗κ
(2 + log
[αy
4u2∗
])+ u∗const.
Setting this to u∗/κ log(yu∗/ν) + Bu∗ gives
const. = B − (2 + log(αν/4u3∗))/κ
Exercise 6.4, The Stratford boundary layer:Further to exercise
6.3: Stratford proposed a clever idea for reducing skin friction.
Hesuggested that a carefully designed adverse pressure gradient
could reduce u2∗ almost to zero(in the linear stress layer) along a
substantial length. The limiting case u2∗ = 0 is called the
17
-
Stratford boundary layer. What mean flow profile does the mixing
length model predict inthis case?
Solution to Ex. 6.4,
(mdU/dy)2 = αy −→ dU/dy = √αy/m =
√α/κ2y
using the fact that in the constant stress region m = κy. Since
this owes to dimensionalanalysis, it is assumed valid in the linear
stress layer as well. Thus we obtain the half powerlaw, U ∝
y1/2Exercise 6.5, Mixing length model with Van Driest damping:The
log-law applies in the region 35
-
the k-ε model can predict such effects. There is an analogy
between rotation and streamlinecurvature, so your conclusions apply
to effects of curvature on the turbulence as well.
Solution to Ex. 6.6, The exact expression for production is P =
−uiuj∂iUj. Themean velocity gradient is ∂iUj = Sij + Ωij is terms
of the rate of strain and rate of rotationtensors. Note that Ωij is
anti-symmetric: Ωij = −Ωji, uiujΩij = 0. (Proof:
uiujΩij = ujuiΩji = uiujΩji = −uiujΩijHence uiujΩij = 0.) So P =
−uiujSij and with the eddy viscosity model P = 2νT SijSji
−2/3k∂iUi. In incompressible flow ∂iUi = 0. Since Ω drops out of
the equation rotation effectsare not represented.
Exercise 6.7, Realizability of k-ε:Consider the k-ε model
(6.2.18, 6.2.19) for the case of homogeneous turbulence. Prove
thatif k and ε are greater than zero initially they can not
subsequently become negative.
Solution to Ex. 6.7, Let ξ = k/ε. Then
dtξ =dtk
ε− kdtε
ε2= 2CµS
2(1 − Cε1)ξ2 + (Cε2 − 1)
Since Cε2 > 1 were ξ to approach 0 this equation gives dtξ
> 0. Hence ξ cannot becomenegative, or even reach 0. Then
dtk =
(2CµS
2ξ − 1ξ
)k
gives
k = k0 exp
[∫ t0
(2CµS2ξ − 1/ξ)dt′
]> 0
Exercise 6.8, Bounds on production:Show that
|P| ≤ 2k|λmax|where |λmax| is the eigenvalue of the rate of
strain tensor S with maximum absolute value.This suggests that eddy
viscosity models should be constrained by
2νT |S|2 ≤ 2k|λmax|.
Let the eigenvalues of S be λ1, λ2 and λ3. For incompressible
flow λ1 + λ2 + λ3 = 0. Ifλ3 = 0, show that
|λmax|2 = |S|2/2and generally that
|S|2 ≥ 3|λmax|2/2
19
-
where λmax is the maximum eigenvalue of S. Hence conclude that
the eddy viscosity oughtto satisfy
νT ≤ 2k3|λmax| .
For the k-ε model Cµ ≤ 2ε/(3k|λmax|) could be imposed.Solution
to Ex. 6.8, Choose the coordinate system in which S is
diagonal:
|P| = |u21λ1 + u22λ2 + u23λ3| < (u21 + u22 + u23)|λmax| =
2k|λmax|
In 2-D, S = diag(λ,−λ, 0) so S2 = 2λ2; or |λ|2 = |S|2/2.In 3-D,
S = diag(λ1, λ2,−λ1 − λ2) so S2 = 2(λ21 + λ22 − λ1λ2). This last
expression is
minimum when 2λ1 = λ2 if λ2 is the maximum eigenvalue. Hence S2
≥ 3/2λ2max.
Exercise 6.9, Front solution:Show that the mean shear at the
propagating front solution corresponding to (6.2.46) is∂yU ∼ (ct −
y)(σε−1)/(2−σε), y < ct.
Solution to Ex. 6.9, From the solution in the text νT = Aµ(ct−y)
and c = Aµ/(2−σε).Near the front
dtU = Aµ∂y[(ct − y)∂yU ]Thus U ∼ const + (ct − y)n with n = 1/(2
− σε) and dyU ∼ (ct − y)n−1.Exercise 6.10, Calibrate an eddy
viscosity transport model:Eddy-viscosity transport models, like
S-A, are an alternative to the k − ε model for full-blown CFD
analysis. They solve a single equation for the dependent variable
νT . Considerthe model equation
DtνT = C1|S|νT − (∂iνT )(∂iνT ) + ∂i[(νT
σν+ ν
)∂iνT
]or, in vector notation
DtνT = C1|S|νT − |∇νT |2 + ∇ ·[(νT
σν+ ν
)∇νT
].
Note that 2|S|2 = |∂yU |2 in parallel shear flow.C1 and σν are
empirical constants. Use log-layer analysis to obtain a formula
relating C1
and σν to the vonKarman constant, κ. What second experimental
datum might be used todetermine values for C1 and σν?
Solution to Ex. 6.10, In the log-region νT = u∗κy, ∂yU = u∗/κy
and the Reynoldsnumber is high, so ν can be ignored in comparison
to νT . Substituting νT = u∗κy
0 =C1√
2− κ2 + κ
2
σν
20
-
This gives σν = 1/(1 − C1/κ2√
2). C1 could be obtained from data on homogeneous shearflow, for
which dtνT = C1|S|νT .Exercise 6.11, Channel flow by k − ε:Use the
velocity and length scales u∗ and H to set up the non-dimensional
k-ε equationsthat must be solved in conjunction with (4.1.4) for
fully developed channel flow. Non-dimensionalize the wall-function
boundary conditions (6.2.35). [The U boundary conditionsimplifies
to U(y/H = 40/Rτ) = log(40)/κ + B in this non-dimensionalization.]
Solve theresulting problem numerically for steady channel flow with
Rτ = 10, 000. Plot your solutionfor U+ and k+.
Solution to Ex. 6.11,
log-linear linear-linear
y+102 103 104
0
10
20
30
40U+10k+100ε+
y+0 2000 4000 6000 8000 10000
0
10
20
30
40U+10k+100ε+
Figure 1: Channel flow: k − ε with wall function.
See figure 1. The numerical centerline velocity is U+ ≈ 27. or
Cf = 2.7 × 10−3.Exercise 6.12, Channel flow by 2-layer:Repeat
previous exercise for the 2-layer k-ε model.
Exercise 6.13, Turbulent dispersion:Stochastic models provide a
concrete physical representation of the ensemble averaged
smooth-ing effect of random convection. The following is simpler
than Taylor’s model: the positionof convected fluid element is
given by
Y (t + dt) = Y (t) +√
α(t)dt ξ(t)
in which ξ(t) the random process defined below equation
2.2.16.1) Show that α(t) = 2KT where Kt is the eddy diffusivity.2)
Use the k − ε formula KT = Cµk2/ε to show that in decaying grid
turbulence
Y 2 ∝ (t + t0)(2Cε2−3)/(Cε2−1)
21
-
Exercise 6.14, Non-existence of 1/2-power law:In a linear stress
layer the total shear stress varies as τ = αy, with α being a
constant thathas dimensions /t2. By dimensional reasoning U ∝
(αy)1/2 and
k ∝ αy; ε ∝ α3/2y1/2; ω ∝ (αy)−1/2.
Show that the k − ε and k − ω models do not admit a power law
solution of this form.
22
-
Chapter 7
Exercise 7.1, Three-dimensional boundary layers:Let αRS =
u2u3/u1u2 and αU = S23/S12 denote the tangents of the direction of
Reynoldsstress and mean rate of strain. Show that (7.1.1) implies
the evolution equation
dtαRS =−2kCµC1S12
u1u2(αU − αRS)
If the initial state is an equilibrium shear flow, U(y), then at
t = 0
−2kCµC1S12u1u2
=−CµC1k∂yU
uv
=−CµC1k
νT=
C1T
.
Suppose that initially αU = 0 = αRS but the flow subsequently
veers so that it developsa component W (y) and αU > 0. Show that
if αU increases monotonically with time thenαRS ≤ αU . Thus the
angle of the Reynolds stress lags that of the mean rate of
strain.Exercise 7.2, The anisotropy tensor:Why must the diagonal
components, b11, b22, b33 of the anisotropy tensor (7.1.5) lie
between− 2/3 and 4/3? In fact, generally the eigenvalues of bij
have to lie between − 2/3 and 4/3 .
Solution to Ex. 7.2, u2 > 0 and k > 1/2u2. From the first
b11 = u2/k − 2/3 > −2/3.From the second b11 < 4/3
Exercise 7.3, Return to isotropy:Use the Rotta model for Fij to
solve (7.1.7) for the relaxation of bij in homogeneous turbu-lence
with no mean flow, ∂jUi = 0, and initially bij = b
0ij . Obtain the turbulence time-scale,
T = k/ε, from the k-ε solution for decaying isotropic turbulence
§6.2.1. Why is it OK touse the isotropic solution, even though bij
�= 0? Rewrite your solution for bij as a solutionfor uiuj . Also
explain why C1 > 1 is necessary.
Solution to Ex. 7.3,
dtbij = (1 − C1)bijT
For decaying grid turbulence k = t−n and ε = −dtk = nk/t; hence
T = t/n. The aboveequation becomes dtbij = n(1 − C1)bij/t with
solution bij = b0ijtn(1−C1). The decay law for kapplies to decaying
non-isotropic turbulence as well. C1 > 1, or else the anisotropy
wouldgrow unbounded with time (see the previous problem).
Exercise 7.4, Invariants of the anisotropy tensor:The second
(II) and third (III) invariants of bij are defined in (2.3.41) as
II = − 1/2bijbji =− 1/2b2ii and III = 1/3bijbjkbki = 1/3b3ii. (The
first invariant bii is identically 0.) For the samecase of
relaxation toward isotropy via Rotta’s model as in exercise 7.3,
write the evolution
23
-
equations for II and III. The two dimensional plane with
coordinates II and III is calledthe ‘invariant map’. Find the
equation for III as a function of II for the Rotta model byforming
the ratio
dIII
dII=
dtIII
dtII.
Then show that this model predicts III ∝ II3/2.Solution to Ex.
7.4, dtIII = −3(1−C1)III/T ; dtII = −2(1−C1)II/T so dIII/dII =
3III/2II.
Exercise 7.5, Solution to SSG for homogeneous shear:Derive
(7.2.39) with (7.1.35) included. Substitute the SSG values for the
coefficients of therapid model. Use either the empirical value PR =
1.6 or the k-ε value (6.2.24). Set theRotta constant to C1 = 1.7.
Compare numerical values to the experimental results. (Thevalues
cited in the text include a non-linear term in the slow model.)
Exercise 7.6, εijk:Show that
εijkεilm = δjlδkm − δjlδmk(if you have not done so in exercise
5.1) From this show that if Ωij = εijkΩk then ωi =εijkΩjk =
2Ωi.
Exercise 7.7, GLM:Show that (7.1.31) can be written in terms of
the production tensor as
℘rapidij = [ 4/5 − 4/3(C2 + C3)]kSij−C2(Pij − 2/3δijP) − C3(Dij
− 2/3δijP) (0.1)
where Dij ≡ −uiuk∂jUk − ujuk∂iUk. From this show that the IP
model (C2 = 3/5 andC3 = 0) corresponds to canceling the terms
involving Dij and Sij to leave
℘rapidij = − 3/5(Pij − 2/3δijP)This is a simple, and for that
reason among others, popular model.
Solution to Ex. 7.7, This is essentially given in the text.
Exercise 7.8, Solution for axisymmetric, homogeneous rate of
strain:Find the equilibrium solution to (7.2.36) for the
incompressible, homogeneous straining flow,U = Sx, V = − 1/2Sy and
W = − 1/2Sz.
Solution to Ex. 7.8, From
∂1U1 = S, ∂2U2 = ∂3U3 = − 1/2Sit can be seen that the only
non-zero b’s are b11 and b22 = b33. Since bkk = 0, by
definition,
b33 = b22 = − 1/2b11.
24
-
Substituting the two above equations into (7.2.36) with dtbij =
0 (equilibrium condition)gives
0 = (1 − C1)b11 − 2b11Sk/ε − (b11 + 2/3)PR − 8/15Sk/ε + (C2 +
C3)(2b11Sk/ε + 2/3PR)PR (≡ P/ε) is known from (6.2.24b) in terms of
the k-ε constants. Rearranging gives
b11 =8/15Sk/ε + 2/3(1 − C2 − C3)PR
(1 − C1) + 2(C2 + C3 − 1)Sk/ε − PRHowever, we also know that PR
= −bijSijk/ε which is PR = − 3/2b11Sk/ε in the present
case. Hence, a consistent equilibrium requires that Sk/ε be the
solution to the quadraticequation
PR =4/5(Sk/ε)2 + (1 − C2 − C3)PRSk/ε
(C1 − 1) + 2(1 − C2 − C3)Sk/ε + PRwhere PR = Cε2−1Cε1−1 is a
known constant.Exercise 7.9, Equations in a rotating frame:If
equations (7.2.36) are referred to a reference frame rotating about
the x3-axis, the i, j-component of the mean velocity gradient
becomes
∂jUi − εij3Ω3in the absolute frame. Here Ω3 is the rate of frame
rotation and ∂jUi is the velocity gradi-ent in the rotating frame.
Evaluate the components of the production tensor for
rotating,homogeneous shear flow, Ui(xj) = δi1δj2Sx2 and compare to
(7.2.41).
The time-derivative of bij in the stationary frame is replaced
by
∂tbij + Ω3(bikεkj3 + bjkεki3)
relative to the rotating frame. Substitute this and the
transformed velocity gradient tensorinto (7.2.36) to derive the
evolution equation (8.3.25) for bij in the rotating frame. Thereare
two choices for the last part of this exercise: either
(i) Solve this equation numerically for rotating, homogeneous
shear flow. The initialcondition is isotropy, bij = 0.
Non-dimensionalize time by S so that the equationscontain the ratio
Ω/S. Investigate how the solution depends on this
parameter—experiment with its value. The k-ε equations
dtk = −kSb12 − ε ; dtε = − (Cε1kSb12 + Cε2ε) εk
will be needed as well. Include a plot of b12 vs. time. Or(ii)
Derive the closed-form equilibrium solution (8.3.33) for b11, b22
and b12 and Sk/ε as
functions of Ω/S in homogeneous shear flow. You can use (6.2.24)
for the equilibriumvalue of P/ε.
In either case, discuss the stabilizing or destabilizing effects
of rotation.
25
-
Exercise 7.10, Wall functions for SMC:State the wall function
boundary conditions for k, ε, U and uiuj that can be applied
some-where inside the log-layer. Write a program to solve plane
channel flow with one of the SMCformulations and wall functions.
Compare the solution to channel flow data at Rτ = 590presented in
figure 4.3.
Exercise 7.11, Limiting budget:Consider the Reynolds stress
budget (3.1.5). By examining the scaling of each term with yas y →
0 at a no-slip wall, derive the limiting behaviors
(7.3.57)ff.Exercise 7.12, Near wall analysis:Verify that the first
of equations 7.3.70 has the behavior
εuiujk
− ν ∂2uiuj∂y2
= ℘nn = kfij = O(y3)
as y → 0. Show that if fij(0) = O(1) and uiuj is not singular,
the only possible behaviorsare uiuj = O(y
4) or uiuj = O(y2).
Solution to Ex. 7.12, Let uiuj ∼ Ayn and note k ∼ εy2/2ν. Then
A[2 − n(n −1)]yn−2 ∼ y2. Hence n = 4 by equating the exponents.
Otherwise the left side must vanishgiving n = 2. Negative n = −1 is
singular.
26
-
Chapter 8
Exercise 8.1, Objective tensors:Show that the absolute vorticity
tensor Ωij = 1/2(∂Vi/∂xj − ∂Vj/∂xi) + εjikΩFk transformsproperly
under the arbitrary change of reference frame (8.1.5).
Exercise 8.2, Realizability by stochastic equations:If a model
can be derived from a well defined random process, then by
definition it isrealizable. Consider a random velocity vector that
satisfies
ui(t + dt) =
(1 − a dt
T
)ui(t) +
√bkdt
Tξi(t).
Here ξ is a vector version of the random variable used in
equation (2.2.16). It has theadditional property that
ξiξj = δij .
a and b are coefficients that will be determined and k is the
turbulent kinetic energy, asusual.
i) Derive the evolution equation for the Reynolds stress tensor
uiuj .ii) Assume the high Reynolds number formula T = k/ε and
determine a and b such that
the Reynolds stress evolution equation is identical to the Rotta
model for decayinganisotropic turbulence.
iii) What realizability constraint is implied by the square root
in equation (8.5.71)?
Exercise 8.3, Bifurcation:Verify (8.3.35). Substitute constants
and plot the bifurcation curve for the LRR model.
Exercise 8.4, Non-linear constitutive equations:Non-linear
constitutive equations In constitutive modeling, the coefficients
in equation (8.3.29)are regarded as unknown functions of |S| and |W
| that can be formulated with completefreedom, provided they are
dimensionally consistent. However, useful constraints can be
ob-tained by invoking realizability. Consider a general
two-dimensional mean flow field wherethe mean rate of strain and
mean vorticity tensors can be written according to (8.3.24)
andlet
b = 2CµS + B(Ω · S − S ·Ω) + C(S2 − 1/3 |S|2δ)Show that for the
flow (8.3.24), the constraints
B2 ≤ [( 2/3 + 1/3cS2)2 − (2CµS)2] /(2ΩS)2and
C ≤ min [1/S2, 3 [ 4/3 − 2Cµ|S|] /S2]ensure that a
(quadratically) non-linear model is realizable. S and Ω are
understood to benon-dimensionalized by k/ε
27
-
Solution to Ex. 8.4, For this flow
uiujk
= 2/3
1 0 00 1 00 0 1
+2CµS 1 0 00 −1 0
0 0 0
−2ΩSb 0 1 01 0 0
0 0 0
+cS2 1/3 0 00 1/3 0
0 0 −2/3
The Schwartz inequality
uv2 ≤ u2 v2gives
(2ΩS)2b2 ≤ (2/3 + cS2/3)2 − (2CµS)2Hence
b2 ≤ [(2/3 + 1/3cS2)2 − (2CµS)2] /(2ΩS)2.The condition that u2
and v2 < 2k gives
2/3 + 1/3cS2 ± 2CµS < 2
orc ≤ 3( 4/3 − 2Cµ|S|)/S2
The condition w2 ≥ 0 gives2/3 − 2/3cS2 ≥ 0
orc ≤ 1/S2
A rigorous approach is the show that these conditions ensure
that the eigenvalues of uiuj/kare non-negative and less than 2.
28
-
Chapter 9
PART III
Exercise 9.1, Familiarization with δ:Show that lim→0
∫∞−∞ e
−|x|eikxdx = 2πδ(k) by explicitly evaluating the integral, then
apply-ing the definition of the δ-function.
Solution to Ex. 9.1, Just evaluate the integral∫ ∞−∞
e−|x|eikxdx =2�
�2 + k2
and reason: This → 0 as � → 0 if k �= 0; this → ∞ as � → 0 if k
= 0; and ∫∞−∞ 2�/(k2+�2)dk =∫∞−∞ 2/(x
2 + 1)dx = 2π. So all the properties of 2πδ(k) are
satisfied.
Exercise 9.2, Direction of independence:Let r = (rx, ry, rz) be
the components of the separation between two hot-wire
anemometers.
Suppose that Rii(r) = e−(r2x+r2y)/L2 , independently of rz. This
means that the eddying
motion is independent of the z-axis. What is Φ(kx, ky, kz)? What
is the kz-space analog toindependence of z in physical space?
Solution to Ex. 9.2,q2L2
4πδ(kz)e
−(k2x+k2y)L2/4
Exercise 9.3, Integral length scale:The integral length-scale
provides a measure of the size of energetic eddies. This
length-scalecan be defined as a vector with x1-component
L1 =
∫ ∞0
Rii(x1; x2 = 0, x3 = 0)dx1,
similarly for L2, L3. For instance Ly = 1/2∫∞−∞ Rii(0, y, 0)dy.
Show that
q2Ly = π
∫ ∞∫−∞
Φ(kx, 0, kz)dkx dkz
The 1-D energy spectrum in the y-direction is defined by
Θ(ky) =
∫ ∞∫−∞
Φ(kx, ky, kz)dkx dkz
so Θ(0) = q2Ly/π. Often it is easier to measure, or compute,
Θ(0) than to obtain L byintegrating the correlation function.
29
-
Solution to Ex. 9.3,∫ ∞∫−∞
Φ(kx, 0, kz)dkx dkz =
∫ ∞∫−∞
[q2
(2π)3
∫ ∞∫−∞
∫ei(kxx+kzz)Rii(r)dxdydz
]dkx dkz
=q2
2π
∫ ∞∫−∞
∫δ(x)δ(z)Rii(r)dxdydz =
q2
2π
∫ ∞−∞
Rii(0, y, 0)dy = q2Ly/π
Exercise 9.4, One dimensional spectrum:Derive a formula that
relates Θij(k1) to the Fourier transform of a two-point
correlationfunction.
Solution to Ex. 9.4, By definition
Φ(kx, ky, kz) =q2
(2π)3
∫ ∞∫−∞
∫ei(kxx+kzz)Rii(r)dxdydz
So it is just a matter of playing with the order of integration∫
∞∫−∞
Φ(kx, 0, kz)dkx dkz =
∫ ∞∫−∞
{q2
(2π)3
∫ ∞∫−∞
∫ei(kxx+kzz)Rii(r)dxdydz
}dkx dkz
=q2
(2π)3
∫ ∞∫−∞
∫ {∫ ∞∫−∞
ei(kxx+kzz)dkx dkz
}Rii(r)dxdydz
=q2
2π
∫ ∞∫−∞
∫δ(x)δ(z)Rii(r)dxdydz
=q2
2π
∫ ∞−∞
Rii(0, y, 0)dy = q2Ly/π
Exercise 9.5, Practice with isotropy:Derive (9.2.35) in physical
space by assuming that uiuj(r) is a function of rk and δkl: i)
Writethe most general tensoral form and apply incompressibility;
ii) Why is the incompressibilityconstraint now ∂riuiuj(r) = 0 =
∂rjuiuj(r)? In your derivation simply obtain the form(9.2.35) with
f an unspecified function; you don’t have to relate it to the
energy spectrum.
Exercise 9.6, Experimental test of isotropy:The 1-D spectra
Θ11(k1) and Θ22(k1) can be measured with a fixed probe. The method
is tomeasure frequency spectra and then use Taylor’s hypothesis to
replace ω by k1Uc, where Ucis the mean convection velocity measured
by the anemometer. Show that if the turbulenceis isotropic then
Θ22(k1) = 1/2 [Θ11(k1) − k1∂k1Θ11(k1)]
30
-
Where the 1-D spectrum is defined by
Θij(k1) =
∫ ∞∫−∞
Φij(k)dk2 dk3
This relation between spectral components has been used to test
experimentally for isotropy.
Solution to Ex. 9.6, Let F (k) = E(k)/4πk4 then, integrating in
cylindrical coordi-nates (r2 = k23 + k
22)
Φ22 =
∫ ∞∫−∞
k21+k22F (k)dk2 dk3 =
∫ 2π0
k21
∫ ∞0
F (√
r2 + k21)rdrdθ+
∫ 2π0
∫ ∞0
F (√
r2 + k21)r3drθ
= k21
∫ ∞0
F (√
r2 + k21)2πrdr + 1/2
∫ ∞0
F (√
r2 + k21)2πr3dr
Compare this to 1/2(Φ11(k1) − k1dk1Φ11(k1))
Φ11 =
∫ ∞∫−∞
k23 + k22F (k)dk2 dk3 =
∫ ∞0
F (√
r2 + k21)2πr3dr
One half of this is the second term in Φ22. Differentiating
this
∂k1Φ11 = k1
∫ ∞0
drF (√
r2 + k21)2πr2dr = −2k1
∫ ∞0
F (√
r2 + k21)2πrdr
Since drF (√
r2 + k21) = rF′(√
r2 + k21) and dk1F (√
r2 + k21) = k1F′(√
r2 + k21). − 1/2k1times this is the first term in Φ22.
Exercise 9.7, The VonKarman spectrum:What is the proportionality
between the length scale L that appears in the VonKarmanspectrum
and the 1-D integral length-scale L111? This notation means ‘the
integral scale inthe x1-direction of the 1-1 correlation’, L
111 =
∫∞0
u1(x)u1(x + rx)drx.
Solution to Ex. 9.7, Cs − CKε2/3Ls+5/3. ε ∝ q2/t ∝ q3/L → ε ∝
L2/t3.Exercise 9.8, One dimensional spectra:Evaluate the 1-D
spectrum, Φ122(k1). Use this to show that the integral scale in the
x1direction of the 22-correlation is one half the integral scale in
the x1 direction of the 11-correlation. In general the transverse
integral scale in isotropic turbulence is one-half thelongitudinal
integral scale.
Exercise 9.9, Synthetic spectra:Another way to synthesize an
isotropic, incompressible random field is
û =
[ξ − (k · ξ)k|k|2
]√E(k)
4πk2
31
-
Verify that (9.2.41) has an isotropic spectrum.
32
-
Chapter 10
Exercise 10.1, Triad interactions:The Navier-Stokes equations
contain the term uj∂jui. Write the Fourier transform of thisterm as
a convolution integral. Also write it as an integral over
wave-number triangles.
Exercise 10.2, Energy decay:Compte-Bellot & Corrsin
considered a crude model for the spectrum. Let
E(k) = Csks k < 1/L(t)
E(k) = CKε2/3k−5/3 1/η > k > 1/L(t)
E(k) = 0 k > 1/η
(0.2)
CK is supposed to be a universal constant. Consider the case η ≈
0. How is Cs related toCK , ε and L(t)? How is q
2 related to ε and L(t)? Assume that ε and q2 follow
power-lawdecays. Derive the formula relating the exponent in L ∝ tm
to s.
Solution to Ex. 10.2, L ∝ t2/(3+s)Exercise 10.3, Two-point
correlations:Consider the two point correlation q2Rii =
ui(x)ui(x
′) in homogeneous turbulence. If r =x − x′ then
∂xjRii(r) = −∂x′jRii(r)and ui(x
′)∂xui(x) = ∂x[ui(x)ui(x′)] because x and x′ can be regarded as
independent vari-ables. From such considerations, derive a formula
relating ε = ν∂iuj∂iuj to the behavior ofthe two-point correlation
function, Rii(r), as r → 0.
Solution to Ex. 10.3, ε = −νq2∇2rRii|0Exercise 10.4, Physical
space:The spectral evolution equations have a physical space
corollary. Correlations and spectraform Fourier transform pairs.
Because of this, spectral closure models are called
‘two-point’closures. An equation for the two-point auto-correlation
in homogeneous, isotropic turbu-lence, analogous to (10.2.9),
is
∂tuiui(r) + 2ν∇2ruiui(r) = −2∂rjujuiu′i(r)
whereujuiu
′i(r) = uj(x)ui(x)ui(x
′)
Derive equation (10.2.16).
Exercise 10.5, A dissipation formula for isotropic
turbulence:Streamwise derivatives of streamwise velocity (∂1u1) are
the easiest to measure. Show thatthis derivative can be used to
infer the rate of dissipation by deriving the formula (∂1u1)2 =
33
-
ε/15ν for isotropic turbulence. Your derivation must start from
the energy spectrum tensor,equation (9.2.23).
This relation can also be derived by invoking the definition of
isotropy in physical space,rather than via Fourier space. The
general form for an isotropic fourth order tensor is (2.3.35)
∂iuj∂kul = Aδikδjl + Bδilδjk + Cδijδkl
The ‘velocity derivative skewness’ is defined by
Sd = (∂1u1)3/((∂1u1)2)3/2
High Reynolds number experimental values of Sd are about 0.4. In
isotropic turbulence theexact equation for dtε contains a term
proportional to ν(∂1u1)3 (the self-stretching term).
Show that this term non-dimensionalized by ε2/k is of order
R1/2T , where RT = k
2/εν.
Solution to Ex. 10.5,For integral approach, in spherical
coordinates
∂1u1∂1u1 =
∫ ∞∫−∞
∫k21
E(k)
4πk2
(1 − k
21
k2
)d3k
=
∫ 2π0
∫ π0
∫ ∞0
E
4π
(1 − cos2 φ) cos2 φ sin φdφdθk2dk
= 2/15
∫ ∞o
Ek2dk = ε/15ν
Symmetry demands that B = C; incompressibility demands that
A+B+3C = 0. HenceA = −4C and
∂iuj∂kul = C[δilδjk + δijδkl − 4δikδjl]The coefficient C is
related to the streamwise derivative of the streamwise velocity by
the1111 component: −2C = (∂1u1)2. Upon contraction, this leads
to
ε = ν∂iuj∂iuj = 15ν(∂1u1)2.
Exercise 10.6, Scale similarity:Strict scale similarity would
require that E(k) be of the form q2(t)L(t)Ẽ(η) with η = kL(t)and
where Ẽ is a non-dimensional, similarity solution. Assume that q2
∝ t−n and find n byrequiring strict self-similarity in equation
(10.2.9).
34
-
Piston
H(t)
U = 0
U = dH/dt
Figure 2: Schematic for compression of turbulence by a
piston.
Chapter 11
Exercise 11.1, Turbulence in a piston; 1-D compression:RDT
analysis can be be applied to a uni-directional compression (Hunt,
1977). This modelsthe effect of compression on turbulence in a
piston, away from the walls because the analysisis of homogeneous
turbulence. Consider the mean flow
U =xdH/dt
H(t)
corresponding to figure 11.9. Show that s1 = H/H0, i.e., that
the total strain is just theexpansion ratio. This permits the
solution to be written in terms of the physical dimensionH instead
of time.
Show that the distorted wavevector is
κ1 =H0H
k1; κ2 = k2; κ3 = k3
and from Cachy’s formula, that
ω̂1 = ω̂01 ; ω̂2 =H0H
ω̂02 ; ω̂3 =H0H
ω̂03
By invoking (11.1.16) obtain a complete solution for the
component û1 and obtain the inte-grand in
u21 =
∫ ∞∫−∞
∫û1û∗1d
3k
explicitly. Evaluate the integral for initially isotropic
turbulence.
Solution to Ex. 11.1,
u21 =q20β
2
4
[1
β2 − 1 +(β2 − 2) tan−1√β2 − 1
(β2 − 1)3/2]
35
-
where β = H0/H and β ≥ 1 is assumed. As β → 1 this gives
u21 → 1/3q20β2[1 − 2/5(β2 − 1)].
As β → ∞ one findsu21 → q20βπ/8; u22 = u23 → q20βπ/16.]
Exercise 11.2, Homogeneous shear RDT:Use a two-term series
solution to (11.2.31), starting with isotropic turbulence, to
obtain theshort-time solutions (11.2.33) for v2 and uv.
36
-
Chapter 12
PART IV
Exercise 12.1, Energy conserving schemes:Consider the
equation
∂tφ + u∂xφ + ∂x(uφ) = 0.
Show that φ2 is conserved. The velocity is stored at the half
grid points: x = (j + 1/2)∆x,j = 1, 2, 3 . . . J and φ is stored at
the grid points x = j. Consider the discretizations
uδxφ|j = 12
(u(j + 1/2)
φ(j + 1) − φ(j)∆x
+ u(j − 1/2)φ(j) − φ(j − 1)∆x
)
δx(uφ)|j = 1∆x
(u(j + 1/2)
φ(j + 1) + φ(j)
2− u(j − 1/2)φ(j) + φ(j − 1)
2
).
Show that φ2 is conserved for this discretization.
Solution to Ex. 12.1, Multiplying by φ(j) gives
1/2∂tφ2 = (u(j + 1/2)φ(j + 1)φ(j) − u(j − 1/2)φ(j)φ(j −
1))/∆x
= [F (j + 1) − F (j)]/∆x
Exercise 12.2, DNS via Burger’s equation:Solve Burger’s equation
(12.2.16) numerically with ν = 0.01 and 0.001, imposing the
initialcondition
1/2 +{
e−20x2
+ e−20(x+1)2
+ e−20(x−1)2}
(0.5 + 0.5ξ)
where ξ is a random number between 0 and 1. The flow domain is −
1/2 ≤ x ≤ 1/2 . Applyperiodic boundary conditions: u(1/2 + j∆x) =
u(−1/2 + j∆x). Use about 160 grid points.Integrate to a
non-dimensional time of 0.12, plotting the solution at t = 0, 0.04,
0.08, 0.12.Looking ahead to §13.1.1: compare the DNS velocity field
to its filtered field at 4 equallyspaced times, t = 0, 0.4, 0.8 and
0.12. Try the running average filter (13.1.2) with N = 2and 3, and
the Pad’e filter (13.1.4) with α = 0.45.
Note: DNS commonly employs the Runge-Kutta method for time
integration. The discrete equations
∂tuj = −u2j+1 − u2j−1
4∆x+
1∆x
(νj+1/2
uj+1 − uj∆x
− νj−1/2 uj − uj−1∆x)
are a system of ordinary differential equations of the form
∂tuj = RHS(ui).
The second order Runge-Kutta method is
up(:) = RHS(u(:),t)u1(:) = u(:)+up(:)∆t/2up(:) =
RHS(u1(:),t)u(:) = u(:)+up(:)∆t ; t = t+∆t.
37
-
At the end, u has been advanced one time step. This is a simple
integration method, which facilitates testsof various convection
schemes.
Exercise 12.3, Spectral methods:Use fast Fourier transforms
(FFT) to evaluate ∂xu
2 and ∂2xu. In the pseudo-spectral methodu2 is evaluated in
physical space to avoid convolution sums; derivatives are evaluated
inFourier space, as in (12.3.25). Specify u by the initial
condition (12.3.26) of the previousproblem. Optional: replace the
finite difference discretization used in the previous problemby
pseudo-spectral evaluation and solve Burger’s equation.
38
-
! Solve Burger’s equation by Runge-Kutta integration
! g95 -r8 burger_DNS.f90 -o burger_DNS
!*************************************************************
MODULE common1
REAL :: vnu=0.000,dx
LOGICAL :: lg
END MODULE common1
!*************************************************************
PROGRAM BURGER_EX
USE common1
PARAMETER (N=128)
REAL ::
U(N),UF(N),S(N),US(N,30),x(N),t=0.,dt,DTprin,Tprint,CC
INTEGER :: i,itm,icur=1,NT
EXTERNAL RHS
pi = 4.d0*atan(1.)
! WRITE(*,"(a)",ADVANCE="NO")"READ c.f.l. " ; READ(*,*)cc
cc = 0.05
dx = 1.d0/dfloat(N-1)
dt = cc*dx
Tend = 0.2 ; DTprin = Tend/5. ; Tprint = DTprin
NT = Tend/dt + 1.01
!**************** Initial Condition ************************
DO i=1,N
x(i) = -0.5+(i-1)*dx
jj = 1000*rand(i/2)
U(i) = (exp(-20.*x(i)**2)+exp(-20.*(x(i)-1.)**2) &
+exp(-20.*(x(i)+1.)**2))*(0.5+0.5*rand(jj))+.5
ENDDO
U(N) = U(1)
US(:,1) = U(:)
!**************** Integrate in time ************************
time: DO itm = 1,NT
sum=0.
DO i=1,N-1
sum = sum+u(i)**2
ENDDO
IF((itm-1)/5*5.eq.itm-1)PRINT *,t,sum*dx
39
-
call rk4(N,RHS,t,U,t+dt)
IF(t.gt.Tprint)THEN
Tprint = Tprint+DTprin
icur = icur + 1
US(:,icur) = U(:)
ENDIF
ENDDO time
!********************* Output *****************************
OPEN (1,file="plotfil")
DO i=1,N
WRITE(1,’(12f9.4)’) x(i),(US(i,ic),ic=1,icur)
ENDDO
WRITE(1,’(/a)’) ’ ’
DO ic=1,icur
CALL FILTER(N,US(1,ic),UF)
US(:,ic)=UF(:)
ENDDO
DO i=1,N
WRITE(1,’(12f9.4)’) x(i),(US(i,ic),ic=1,icur)
ENDDO
WRITE(*,"(a,8i3)")"gplot plotfil ",(ic,ic=1,icur)
STOP
END PROGRAM BURGER_EX
!******************************************************
SUBROUTINE RHS(N,t,U,S)
!******************************************************
USE common1
REAL :: U(N),S(N),work(2*N),data(2,N)
INTEGER :: j,jm,jp,N
DO j=1,N
jp = mod(j-1,N-1)+2
jm = N-1-mod(N-j,N-1)
! S(j) = -(U(jp)**2-U(jm)**2)/(4.*dx) & ! central
S(j) = -(U(jp)**2-U(jm)**2)/(6.*dx) -U(j)*(U(jp)-U(jm))/(6.*dx)
& ! energy conserv
! S(j) = -U(j)*(U(j)-U(jm))/(2.*dx) & ! upwind
+ (U(jp)-2.*U(j)+U(jm))*vnu/dx**2
ENDDO
RETURN
END SUBROUTINE RHS
40
-
!******************************************************
SUBROUTINE FILTER(N,U,UF)
!******************************************************
USE common1
REAL :: U(N),Uf(N)
INTEGER :: j,jm,jp,N,NF=3
DO j=1,N
Uf(j) = U(j)
DO i=1,NF-1
jp = mod(j+i-2,N-1)+2
jm = N-1-mod(N-j+i-1,N-1)
Uf(j) = Uf(j)+U(jp)+U(jm)
ENDDO
jp = mod(j+NF-2,N-1)+2
jm = N-1-mod(N-NF+i-1,N-1)
Uf(j) = (Uf(j)+.5*(U(jp)+U(jm)))/(2.*NF)
ENDDO
END SUBROUTINE FILTER
!*********************************************************
! 4th order Runge-Kutta routine *
!*********************************************************
SUBROUTINE rk4(N,RHS,t,y,tend)
REAL :: y(N),yp(N),y1(N),y1p(N),y2(N),y2p(N),y3(N),y3p(N)
&
,h,t,tend,tmid
EXTERNAL RHS
h = tend-t
tmid = t+h/2.d0
CALL RHS(n,t,y,yp)
y1 = y+0.5d0*h*yp
CALL RHS(n,tmid,y1,y1p)
y2 = y+0.5d0*h*y1p
CALL RHS(n,tmid,y2,y2p)
y3 = y+h*y2p
CALL RHS(n,tend,y3,y3p)
y = y+h/6.0*(y3p+2.0d0*y2p+2.0d0*y1p+yp)
t = tend
RETURN
END
41
-
Chapter 13
Exercise 13.1, Filtering in physical space:The top-hat filter
is
û =1
∆f
∫ ∆f /2−∆f/2
u(x + ξ) dξ
Let u = sin(kx). Evaluate û and ̂̂u. Letu =
{ |1 − x|, x < 10, |x| ≥ 1.
Evaluate û and compare it to the unfiltered function.
Solution to Ex. 13.1, For the sine function
sin(kx) =2sin(k∆f/2)
k∆fsin(kx); sin(kx) =
[2sin(k∆f/2)
k∆f
]2sin(kx)
For the triangular function, if 1 − ∆f/2 < x < 1 + ∆f/2
averaging will not alter the linearslope. If |x| > 1 + ∆f/2
filtering is over a region where u = 0. Hence
û =
{0 |x| > 1 + ∆f/2
1 − x −1 + ∆f/2 < x < 1 − ∆f/2
When x is within ∆f/2 of 1 or −1 the function is smoothed
û =
1
2∆f
[∆f2
+ 1 − x]2
1 − ∆f/2 < x < 1 + ∆f/21
2∆f
([∆f2
+ x]2
− 1)
−1 − ∆f/2 < x < −1 + ∆f/2
Exercise 13.2, Filter in Fourier space:The Gaussian filter is
expressed by the running average
û =1√π∆f
∫ ∞−∞
e−ξ2/∆2f u(x + ξ) dξ
What is Fk of this filter in Fourier space?
Solution to Ex. 13.2,
Fk =1√π∆f
∫ ∞−∞
e−ξ2/∆2f e−ikξ dξ =
e−k2∆2f/4√π∆f
∫ ∞−∞
e−(ξ/∆f +ik∆f/2)2
dξ
42
-
=e−k
2∆2f/4√π
∫ ∞−∞
ex2
dx = e−k2∆2f/4
Exercise 13.3, Filtered spectrum:Convince yourself that the
filtered and unfiltered one-dimensional energy spectra (see
page242) are related by
Θ̂11 = FkF∗k Θ11.
On a single graph, plot the filtered and unfiltered spectrum
(9.2.28) with p = 17/6 for variousratios of ∆f/L. Use either the
top-hat or Gaussian filter.
Solution to Ex. 13.3, ûi(k) = Fku(k) → ûi(k)û∗j(k) = FkF ∗k
ui(k)u∗j(k) see (9.2.13)and (9.2.18) to relate this to spectra.
Exercise 13.4, Filtering and vorticity:Let
u = A cos kx sin ky cos kz
v = −A sin kx cos ky cos kzw = 0.
This is called the Taylor-Green vortex. Evaluate the vorticity
vector field. Average |u|2 and|ω2| over one period of the sine
waves in each direction. Let
A2 =k3
(1 + (kη)2)2.
Filter the velocity and vorticity fields with a filter width
equal to 5η. Average |û|2 and |ω̂2|over a period and compare to
their unfiltered values.
Exercise 13.5, WALE subgrid model:Show that
SijSji = 1/6(S2S2 + Ω2Ω2) − 2/3Ω2S2 + 2S2ijΩ2ji
where Sij is defined in (13.1.19). Ω2ij = ΩikΩkj , Ω2 = Ω2ii and
similarly for S. The Cayley-
Hamilton theorem (2.3.38) with the invariants defined in
equation (2.3.40) is needed. Whyis Ω2 ≤ 0? Show that SijSji is zero
in parallel shear flow, u(y).
Solution to Ex. 13.5,
|S 2| = |S4| + |Ω4| + 2S2 · Ω2 + 1/3(|S2| + |Ω2|)2 − 2/3(|S2| +
|Ω2|)2
= |S4| − 1/3S2S2 + |Ω4| − 1/3Ω2Ω2 + 2S2 · Ω2 − 2/3 |S2||Ω2|where
|(·)| denotes the trace of a matrix, not an absolute value. From
Cayley-Hamilton
S4 = S · S3 = 1/2 |S3|S + 1/2 |S2|S2
43
-
Thus|S4| = 1/2S2S2 and |Ω4| = 1/2Ω2Ω2
which gives the desired result. In parallel shear flow Ω2 = −S2
so Ω = −S and |S 2| = 0Exercise 13.6, LES via Burger’s
equation:Repeat 12.3, adding the Smagorinsky model (13.1.12) and
using a 4 times coarser grid.Select a value of cburger to optimize
agreement with the filtered DNS field. Compare the LESsolution to
the filtered DNS data at several times.
44