Statistical Method
3. Methods of Studying Turbulent Flows
Characteristic function
The characteristic function is the Fourier transform of the PDF
*+
𝑓" 𝜅 = % 𝑒 '() 𝑓 𝑣 𝑑𝑣 = 𝑒 '()
,+
The characteristic function of the sum of two independent variables
is the product of their individual characteristic functions
68
Nasr Azadani
Statistical Method
3. Methods of Studying Turbulent Flows
Characteristic function
The 𝑛th moment is related to the derivatives of the characteristic
function by
𝑑 0 𝑓" 𝜅
0
1 = 𝑖0 𝑣0
𝑑𝜅
(23
+ +
1 𝑑 𝑓" 𝜅
0
𝑖𝜅 0
𝑓" 𝜅 = 5 0
1 𝜅0 = 5 𝑣0
𝑛! 𝑑𝜅 𝑛!
023 (23 023
69
Nasr Azadani
Statistical Method
3. Methods of Studying Turbulent Flows
Joint probability density functions (JPDF)
The JPDF, 𝑓8 𝑢, 𝑣 , of variables 𝑢 and 𝑣 is the probability of
finding the first random variable between 𝑢 and 𝑢 + 𝑑𝑢, and the
second one between 𝑣 and 𝑣 + 𝑑𝑣
The basic properties of the JPDF are
Ø 𝑓8 𝑢, 𝑣 ≥ 0
+ +
Ø ∫ ∫ 𝑓8 𝑢, 𝑣 𝑑𝑢 𝑑𝑣 = 1
,+ ,+
+ +
𝑓 𝑢 = % 𝑓8 𝑢, 𝑣 𝑑𝑣 , 𝑓 𝑣 = % 𝑓8 𝑢, 𝑣 𝑑𝑢
,+ ,+
70
Nasr Azadani
Statistical Method
3. Methods of Studying Turbulent Flows
Moments of 𝒖 𝒂𝒏𝒅 𝒗
First moment
+ +
𝑢𝑣 = % % 𝑢𝑣𝑓8 𝑢, 𝑣 𝑑𝑢 𝑑𝑣
,+ ,+
Covariance
is a measure of the asymmetry of the JPDF
𝒞 𝑢, 𝑣 = 𝑢𝑣 − 𝑢 𝑣 = 𝑢G 𝑣 G
Correlation function
is used to quantify the degree of correlation between 𝑢 and 𝑣
𝑢G 𝑣 G
𝑟 𝑢, 𝑣 =
𝑢G I 𝑣 GI
71
Nasr Azadani
Statistical Method
3. Methods of Studying Turbulent Flows
Correlated and related variables
q For perfectly correlated variables, the correlation function is ±1
q Two variables whose covariance is zero, or equivalently whose
correlation is zero, are said to be uncorrelated
q Two uncorrelated variables need not to be independent
q Statistical independence of two variables requires that the JPDF
can be expressed as the product of the individual PDFs
𝑓8 𝑢, 𝑣 = 𝑓 𝑢 𝑓(𝑣)
q Independent variables are uncorrelated, but the reverse is not true
72
Nasr Azadani
Statistical Method
3. Methods of Studying Turbulent Flows
Ergodicity and statistical symmetries
Stationarity
q Turbulence is stationary if all mean quantities are invariant under
a translation in time
q A stationary variable 𝑣 is ergodic if the time average of 𝑣
converges to the mean 𝑣 as the time interval extends to infinity
1 O
% 𝑣 𝑡 𝑑𝑡 = 𝑣 𝑎𝑠 𝑇 → ∞
𝑇 3
q In this case a time average is equivalent to an ensemble average
73
Nasr Azadani
Statistical Method
3. Methods of Studying Turbulent Flows
Ergodicity and statistical symmetries
Homogeneity
q Turbulence is homogeneous if all the mean quantities are
invariant under any spatial translation
q Then an ergodic hypothesis allows an ensemble average to be
calculated as a spatial average
1 U
% 𝑣 𝑡 𝑑𝑡 = 𝑣 𝑎𝑠 𝐿 → ∞
𝐿 3
74
Nasr Azadani
Statistical Method
3. Methods of Studying Turbulent Flows
Ergodicity and statistical symmetries
Isotropy
Turbulence is isotropic if all the mean quantities are invariant under
any arbitrary rotation of coordinates
Axisymmetry
Turbulence is axisymmetric if all the mean quantities are invariant
under a rotation about one particular axis only
75
Nasr Azadani
Statistical Method
3. Methods of Studying Turbulent Flows
Central Limit Theorem
Many quantities in turbulence can be thought of as a sum of
random variables. If the incremental changes 𝑥0 all have the same
unknown PDF and zero mean value, then the net change in variable
is given by,
X
𝑧 = 5 𝑥0
02Y
If all the individual incremental changes are independent and
uncorrelated, the variance of 𝑧 is given by
X X
𝑧 I = 5 5 𝑥0 𝑥Z = 𝑁𝜎 I
02Y Z2Y
76
Nasr Azadani
Statistical Method
3. Methods of Studying Turbulent Flows
Central Limit Theorem
𝑢 = 𝑁 ,Y⁄I 𝑧
X
" '(^ '(X_`⁄a ∑e
df` cd = 𝑓" 𝜅𝑁 ,Y⁄I
𝑓^ 𝜅 = 𝑒 = 𝑒
𝜅I𝜎 I
𝑓" 𝜅𝑁 ,Y⁄I =1− + 𝑂 𝜅 i 𝑁 ,i⁄I
2𝑁
X
𝜅I𝜎 I (a ma
𝑓"^ 𝜅 = lim 1 − = ,
𝑒 I
X→+ 2𝑁
1 ^a
,
𝑓 𝑢 = Y⁄I
𝑒 Ima
2𝜋𝜎 I
77
Nasr Azadani
Statistical Method
3. Methods of Studying Turbulent Flows
Central Limit Theorem
The sum, normalized by 1/ 𝑁, of 𝑁 random, independent and
identically distributed variables of zero mean and finite variance,
𝜎 I , is a random variable with a PDF converging to a Gaussian
distribution with variance 𝜎 I
78
Nasr Azadani
Statistical Method
3. Methods of Studying Turbulent Flows
Intermittency
q A signal is said to be intermittent if rare events of large magnitude
are separated by long periods with events of low magnitude
q Spatial intermittency implies that the signal displays localized
regions with events of large magnitude, and wide areas with events
of low magnitude
q PDFs of intermittent flows are not Gaussian
q If turbulence is dominated by coherent structures localized in
space and time, then PDFs are not Gaussian
q Flows characterized by intermittency are not self-similar 79
Nasr Azadani
