Ertesvag - Turbulent Flow and Combustion

Ivar S.
Ertesvåg
Turbulent Flow
and Combustion
From turbulence theory to engineering tools
Trondheim 2000/2008
Preface
This book is written for the course Turbulent combustion, heat and mass transfer at Department of
applied mechanics, thermodynamics and fluid dynamics, NTNU (formerly Department of engineer-
ing thermodynamics, Norwegian Institute of Technology). It started in 1993, when I took over the
teaching when Professor Ståle Byggstøyl (f. 1955) abruptly passed away in February. A first draft
was ready the following year. Since then the manuscript has regularly been revised and extended.
Some sections are written on the basis of lecture notes left by Ståle Byggstøyl. Some of the material
can be found in my dissertation (1991). Particularly in the combustion part, I have received good
help from Nils Inge Lilleheie and Inge R. Gran. The content of the chapter on EDC is, naturally,
developed in close contact with Bjørn F. Magnussen. Students have given remarks that are often
followed. Several good and eager helpers have read the entire or parts of the draft and given many
advices and corrections.
This book is an outcome of the research activity in the group around Bjørn F. Magnussen at NTNU
and Sintef Energy Research (Thermodynamics and Fluid Dynamics, formerly Engineering Thermo-
dynamics). For two to three decades this group has been working on modeling and simulation of
turbulent flows with computers. The book also contributes to the documentation of the basis and the
methods we and others are using for solving engineering problems.
Several researchers have contributed figures to exemplify how theory and method can be used. Their
names are mentioned in the figure caption.
December 1999
Ivar S. Ertesvåg
The present version is a draft translation of the Norwegian text. The attempt is to give a translation, not a
new edition. However, nothing is added, re-evaluated or removed (with exception of a few short comments
on Norwegian terminology, the addresses on the next page and a few corrections). The reason for this policy
is that an English text is required for the teaching of classes with foreign students, while Norwegian readers
benefit from the complete book in Norwegian. I realize that during the last decade there has been considerable
development both on the field of science and in my own understanding and knowlege of it.
Later, in some years, I plan to revise the book - that is, also the content.
April 2008
Ivar S. Ertesvåg
v
For the reader
In addition to the book, I have made a collection of exercises with solutions. These are to comple-
ment and deepen some of the matter here. This compendium will be available from the Department.
An information page will be put on the web, with address
http://folk.ntnu.no/ivarse/bok99. Here, you will find various relevant information.
Experience tells that there is always one more error. When you find errors or flaws, have hints or
comments, I would like to hear from you. A list of corrections, and other questions and comments
of general interest, I will put on the web page above.
Besides this, I refer to Sec. 1.2 About reading the book.
Ivar S. Ertesvåg
Phone: +47 73 59 38 39
E-mail: Ivar.S.Ertesvag@ntnu.no
Address:
Department of Energy and Process Engineering
Norwegian University of Science and Technology (NTNU)
NO-7491 Trondheim, Norway
vi
Contents
Preface v
List of symbols xi
1 Introduction and basics 1

1.1 Objective, background, overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 About reading this book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 This section you do not need to read . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Combustion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6 Numerical tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.7 Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.8 Chemical reactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2 Turbulence modeling 31
2.1 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 Reynolds-decomposition and averaging . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3 Turbulence equations, Reynolds equations . . . . . . . . . . . . . . . . . . . . . . . 36
2.4 Turbulent transport and exchange . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.5 Turbulence viscosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.6 Prandtl’s mixing length model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.7 More on turbulent transport and length scales . . . . . . . . . . . . . . . . . . . . . 42
2.8 Characteristic scales: – Into the fog? . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3 The k -εε turbulence model 45

3.1 A note on turbulence models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Turbulence energy: Definition and development . . . . . . . . . . . . . . . . . . . . 48
3.3 Modeling of the turbulence energy equation . . . . . . . . . . . . . . . . . . . . . . 50
3.4 k-ε model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.5 Buoyancy and turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.6 What can we expect from a k-ε model? . . . . . . . . . . . . . . . . . . . . . . . . 58
3.7 Formal requirements to turbulence models . . . . . . . . . . . . . . . . . . . . . . . 61
4 Some simple flows 63

4.1 Homogeneous and isotropic turbulence . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2 Isotropic decaying turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.3 Boundary layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
viii Contents
4.4 A logarithmic region in boundary layers – wall functions . . . . . . . . . . . . . . . 70

4.5 Constants i k-ε model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5 Energy and mass transfer (scalar transport) 79

5.1 Mixing by turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2 Model for turbulence fluxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.3 Diffusion model with u ′i u ′j equation . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.4 Mixture fraction, conserved scalar . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.5 Scalar variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.6 Thermal boundary layer – wall functions . . . . . . . . . . . . . . . . . . . . . . . . 88
5.7 Wall function for mass transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6 Equations for the Reynolds stresses 93

6.1 Models for Reynolds stresses – a summary . . . . . . . . . . . . . . . . . . . . . . . 93
6.2 Development of the equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.3 The terms of the equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.4 Modeling of the equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.5 Algebraic simplification of the equation . . . . . . . . . . . . . . . . . . . . . . . . 103
6.6 Why (not) use the Reynolds stress equations? . . . . . . . . . . . . . . . . . . . . . 105
7 Statistical functions in turbulence 107

7.1 Statistical functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.2 Isotropic turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3 Intermittency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8 Energy transfer in turbulent flow 121

8.1 Break-up of eddies in turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8.2 Energy spectrum for turbulent flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8.3 Kolmogorov’s hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.4 Energy spectrum and combustion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.5 Molecular mixing; scalar-dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.6 More on minor scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
9 Various topics in turbulence 135

9.1 Length scales and timescales in turbulent flow . . . . . . . . . . . . . . . . . . . . 135
9.2 Length scales from turbulence models . . . . . . . . . . . . . . . . . . . . . . . . . 138
9.3 Reynolds number for turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
9.4 Alternative methods and models for numerical simulation of turbulence . . . . . . . 141
9.5 Simple estimates for turbulence quantities . . . . . . . . . . . . . . . . . . . . . . . 143
10 Turbulent combustion 149

10.1 Premixed and non-premixed flames . . . . . . . . . . . . . . . . . . . . . . . . . . 149
10.2 Characterization of turbulent flames . . . . . . . . . . . . . . . . . . . . . . . . . . 151
10.3 Chemical kinetics, extinction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
10.4 Combustion models – classification . . . . . . . . . . . . . . . . . . . . . . . . . . 156
10.5 Series expansion of the Arrhenius expression . . . . . . . . . . . . . . . . . . . . . 159
10.6 Eddy models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
Contents ix
10.7 Prescribed probability density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

10.8 Flamelet models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
10.9 Transport equation for probability density . . . . . . . . . . . . . . . . . . . . . . . 165
10.10Direct numerical simulation of turbulent combustion . . . . . . . . . . . . . . . . . 166
11 Magnussen’s combustion model, «Eddy Dissipation Concept» 171

11.1 Basis and overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
11.2 Energy transfer – cascade model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
11.3 Cascade and energy spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
11.4 The fine structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
11.5 Reactor model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
11.6 Fast reactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
11.7 Extinction and blow-off . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
11.8 Model expressions listed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
12 Etterrakst 189
12.1 Numerical simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
12.2 Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
12.3 Soot and unburned hydrocarbons . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
12.4 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
12.5 Fuels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
12.6 Two and multiphase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
12.7 Out into practice – What now? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
Appendix: Equations that we love 195
A The fundamental equations – and something more 197

A.1 Mass balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
A.1.1 Continuity equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
A.1.2 On mass, amount of substance and volume . . . . . . . . . . . . . . . . . . 198
A.1.3 Equation for species mass fraction, concentration – Mass diffusion . . . . . . 199
A.2 Momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
A.2.1 Equation for momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
A.2.2 Euler equation, Bernoulli equation . . . . . . . . . . . . . . . . . . . . . . . 202
A.2.3 Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
A.2.4 Boussinesq approximation for buoyancy . . . . . . . . . . . . . . . . . . . . 204
A.3 Energy balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
A.3.1 General equation of energy . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
A.3.2 Enthalpy and internal energy in combustion – specific heat capacity . . . . . 208
A.3.3 Enthalpy gradient, energy gradient . . . . . . . . . . . . . . . . . . . . . . . 211
A.3.4 Model for heat flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
A.3.5 Enthalpy equation with enthalpy gradient . . . . . . . . . . . . . . . . . . . 213
A.3.6 Heat flux with gradient of internal energy . . . . . . . . . . . . . . . . . . . 214
A.3.7 Equation for temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
A.3.8 Equation for thermal enthalpy and heating value . . . . . . . . . . . . . . . 217
A.4 A short note on tensor notation and coordinates . . . . . . . . . . . . . . . . . . . . 218
x Contents
B Equations for turbulent flow 221

B.1 Decomposition into turbulent and mean fields: Reynolds and Favre . . . . . . . . . . 221
B.2 Continuity equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
B.3 Turbulence equations for momentum . . . . . . . . . . . . . . . . . . . . . . . . . 225
B.4 Equation for species mass fraction . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
B.5 Energy equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
B.6 Turbulence energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
B.7 Reynolds stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
B.8 Turbulence equations in cylinder coordinates . . . . . . . . . . . . . . . . . . . . . 229
B.8.1 Momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
B.8.2 Equations for Reynolds stresses . . . . . . . . . . . . . . . . . . . . . . . . 230
B.8.3 Simplifications, axial symmetry . . . . . . . . . . . . . . . . . . . . . . . . 232
B.9 Series development and average of the reaction term (Arrhenius) . . . . . . . . . . . 233
C Dimensionless parameters 237

C.1 Origin and names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
C.2 Reynolds number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
C.3 Richardson number, Froude number . . . . . . . . . . . . . . . . . . . . . . . . . . 238
C.4 Prandtl number, Schmidt number, Lewis number . . . . . . . . . . . . . . . . . . . 238
C.5 Damköhler number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
C.6 Courant number or CFL number . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
C.7 Other dimensionless parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
C.8 More prominent people of the subject . . . . . . . . . . . . . . . . . . . . . . . . . 240
D Nokre ord om ord 247

D.1 Ordforklaringar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
D.2 Ord om strøymingar og forbrenning . . . . . . . . . . . . . . . . . . . . . . . . . . 256
D.2.1 Særtilfelle av strøyming; Inndeling etter geometri og anna . . . . . . . . . . 256
D.2.2 Storleikar i strøyminga . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
D.2.3 Forbrenningsutstyr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
D.2.4 Flammer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
D.2.5 Ord om likningar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
D.3 Ymse ord i fag og lag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
D.3.1 Turbulent energi eller turbulensenergi? . . . . . . . . . . . . . . . . . . . . 261
D.3.2 Implementere – komplett: berre fyll . . . . . . . . . . . . . . . . . . . . . . 262
D.3.3 Reaksjonsrate og reaksjonsfart . . . . . . . . . . . . . . . . . . . . . . . . . 262
D.3.4 «Fluidized bed»: Kvervelsjikt, sandseng, eller ... svevebed? . . . . . . . . . 263
D.3.5 «Large-eddy simulation»: Storevje-simulering . . . . . . . . . . . . . . . . 263
D.4 Ord om rørsle i luft og vatn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
D.5 Ord om koordinatsystem og nettverk . . . . . . . . . . . . . . . . . . . . . . . . . . 266
D.6 Ein stad å møtast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
D.7 Ord om folk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
D.8 Engelsk – norsk ordliste; forkortingar . . . . . . . . . . . . . . . . . . . . . . . . . 269
List of litterature 273

List of symbols
This list contains symbols that are used in the book. Some symbols that are used only in one section,
are defined in the text and not included here.
Latin symbols
C1 , C2 constants in the redistribution model, Eq. (6.19)

CD1 , CD2 constants in the cascade model, Sec. 11.2
Ci j transient term and convective transport with the mean flow in the equation for
Reynolds stresses, Eq. (6.2) [m2/s3 ]
Ck transient term and convective transport with the mean flow in the equation for
turbulence energy k, Eq. (3.4) [m2 /s3 ]
Cp specific heat capacity, Eq. (1.6) and (A.53a) [J/kg/K]
C p,k specific heat capacity for species k, defined in Eq. (A.53b) [J/kg/K]
Cv specific heat capacity, Eq. (A.58a) [J/kg/K]
Cv,k specific heat capacity for species k, defined in Eq. (A.58b) [J/kg/K]
Cε1 , Cε2 constants in the model equation for ε, Eq. (3.23)
Cµ constant in the k-ε model, Eq. (3.20)
ck amount-of-substance concentration (molar concentration, molarity) of species k,
Eq. (A.12) [mol/m3]
D mass diffusivity, Eqs. (1.5) and (A.18) [m2 /s]
D three-dimensional dissipation spectrum, Eq. (8.7) [m3 /s2 ]
Di j diffusion term in the equation for Reynolds stresses, Eq. (6.2) [m2 /s3 ]
Dk diffusion term in the equation for k, Eq. (3.4) [m2 /s3 ]
Dε diffusion term in the equation for ε, Eq. (3.16) [m2 /s4 ]
E three-dimensional energy spectrum, Eq. (8.5) [m3 /s3 ]
specific internal energy, 2 2
e P Eq. (A.44) [J/kg=m /s ]
for a mixture is e = k Yk ek
et = e + 12 u i u i ; total specific internal energy, Eq. (A.42) [J/kg=m2 /s2 ]
thermal internal energy, 2 2
1e PEqs. (A.50), (A.51), (A.62) [J/kg=m /s ]
for a mixture is 1e = k Yk 1ek
F distribution function, Eq. (7.3)
f longitudinal correlation function, Eq. (7.22)
f probability density (-function), pdf, Eqs. (7.5), (7.7)
fi acceleration due to distant force (mass force) in x i -direction,
Eqs. (1.3), (A.22), (A.42) [m2/s]
xii List of symbols
f k,i acceleration of species k due to distant force in x i -direction [m/s2 ]

Gi j source term in the equation for Reynolds stresses, due to buoyancy,
Eq. (B.43) [m2 /s3 ]
Gk source term in the equation for k, due to buoyancy,
Eqs. (3.28), (B.38) [W/kg = m2 /s3 ]
g lateral correlation function, Eq. (7.24)
g gravity acceleration (approx. 9.8 m/s2 ), Eq. (A.31)
HR heating value for fuel, Eq. (A.64) [J/kg]
h = e + p/ρ ; static specific P enthalpy, Eqs. (1.6), (A.46) [J/kg=m2/s2 ]
for a mixture is h = k Yk h k
hk = h ◦f,k + 1h k ; enthalpy for species k, Eqs. (A.47), (A.52) [J/kg]
h ◦f,k enthalpy of formation for species k at the reference state (T ◦, p◦ ); [J/kg]
ht = h + 12 u i u i = et + p/ρ; total specific enthalpy, Eq. (A.45) [J/kg=m2/s2 ]
1h thermal enthalpy, Eqs. P (A.47), (A.49), (A.105) [J/kg=m2/s2 ]
for a mixture is 1h = k Yk 1h k
jk, j diffusive mass flux of species k in x j -direction, Eqs. (A.16), (A.18) [(kg)k /s/m2 ]
k ≡ 21 u ′i u ′i ; (specific) (kinetic) turbulence energy, Sec. 3.2 [J/kg = m2 /s2 ]
L, ℓ characteristic length scales for turbulence [m]
Le = α/D = Sc/Pr; Lewis number, Sec. C.4
M molar mass for mixture, Eq. (A.8) [kg/mol]
Mk molar mass for species k, Eq. (A.8) [kg/mol]
m, m k mass, mass of species k, Eq. (A.6) [kg]
ṁ mass transfer, Eqs. (11.21)–(11.22) [s−1 ]
n, n k amount of substance, amount of substance of species k, Eq. (A.7) [mol]
O order of magnitude, page 43
P probability, Eqs. (7.1)–(7.2)
Pi j production term in the equation for Reynolds stresses, Eq. (6.2) [m2 /s3 ]
Pk production term in the equation for k, Eq. (3.4) [W/kg = m2 /s3 ]
Pε production term in the equation for ε, Eq. (3.19) [m2 /s4 ]
Pr = ν/α; Prandtl number, Sec. C.4
p static pressure, Eqs. (1.3), (A.23), (A.42), Sec. A.2.3 [Pa = N/m2 ]
Qε destruction term in the equation for ε, Eq. (3.19) [m2 /s4 ]
Q̇ heat, “internal” source term in the energy equation, e.g. radiation, Eq. (A.42) [W/m3 ]
q dissipation in the cascade model, Sec. 11.2 [m2 /s3 ]
qi heat flux per area in x i -direction, Eqs. (A.42), (A.83) [W/m2 ]
R gas constant (for mixture), Eq. (A.63) [J/kg/K]
Rk gas constant for species k in a gas mixture, Eq. (A.60) [J/kg/K]
Ru universal gas constant (approximately 8.31 J/mol/K), Eq. (A.63)
Ri j correlation, Eq. (7.20) [m2 /s2 ]
Rk reaction rate (production rate) of species k, Eqs. (1.5), (A.16) [(kg)k /m3 /s]
ReK = u ′ η/ν; turbulence Reynolds number, Eq. (9.19)
List of symbols xiii
ReT = k 2 /νε; turbulence Reynolds number, Eq. (9.16)

Ret = νt /ν; turbulence Reynolds number, Eq. (9.15)
Reλ = u ′ λ/ν; turbulence Reynolds number, Eqs. (9.12), (9.17)
Re Reynolds number, Sec. C.2
r distance, Sec. 7.2, Eq. (8.14) [m]
r radial coordinate in cylinder coordinates, Sec. B.8 [m]
Sc = ν/D; Schmidt number, Sec. C.4
Ti j = Ci j − Di j ; transport, Eq. (6.21) [m2 /s3 ]
Tk = Ck − Dk ; transport, Eq. (6.22) [m2 /s3 ]
T temperature [K]
t time [s]
UR reaction energy for fuel, Eq. (A.70) [J/kg]
uj velocity component (Euler-) in x j -direction, Eq. (1.3) [m/s]
uL laminar flame velocity, velocity scale for flame, Sec. 10.2 [s]
uτ ≡ (τw /ρ)1/2 ; friction velocity at a wall, Eq. (4.8) [m/s]
V volume [m3 ]
Vk, j diffusion velocity of species k in x j -direction, Eq. (A.18) [m/s]
v = (νε)1/4 ; Kolmogorov micro-scale of velocity, Eq. (8.19) [m/s]
vr velocity component in radial direction in cylinder coordinates, Sec. B.8 [m/s]
vz velocity component in axial direction in cylinder coordinates, Sec. B.8 [m/s]
vθ velocity component in tangential direction in cylinder coordinates, Sec. B.8 [m/s]
w energy transfer in cascade model, Sec. 11.2 [m2 /s3 ]
Xk mole fraction of species k, Eq. (A.7) [(mol)k /mol]
xj Cartesian coordinate in j -direction ( j = 1, 2 or 3) [m]
Yk mass fraction of species k, Eqs. (1.5), (A.6) [(kg)k /kg]
z axial coordinate in cylinder coordinates, Sec. B.8 [m]
Greek symbols
α = λ/(C p ρ); thermal diffusivity, Eqs. (1.6), (A.87) [m2/s]

α (thermal) volume expansion coefficient, Eq. (A.40) [K−1 ]
γ intermittency factor in EDC, Eqs. (11.19)–(11.20)
1m mesh width for grid in numerical simulation, Sec. 9.4 [m]
δ boundary layer thickness, Sec. 4.3 [m]
δr chemical length scale for reaction, Sec. 10.10 [s]
δi j Kronecker delta; =1 when i = j ; =0 when i 6= j (page 220)
δL laminar flame length, length scale for flame, Sec. 10.2 [s]
ε dissipation term in the equation for turbulence energy, Eq. (3.4)
[W/kg = m2 /s3 ]
εi j dissipations term in the equation for Reynolds stresses, Eq. (6.2) [m2 /s3 ]
η = (ν 3 /ε)1/4; Kolmogorov micro length scale, Eq. (8.20) [m]
ηB = (νD2 /ε)1/4 ; Batchelor length scale, Eq. (8.27) [m]
xiv List of symbols
θ tangential coordinate in cylinder coordinates, Sec. B.8 [1]

θ time scale for turbulence [s]
θk = k/ε; time scale for turbulence , Eqs. (8.24), (9.3) (also page 48) [s]
κ von Kármán constant, Eq. (4.9)
κ wave number, Eq. (8.5) [m−1 ]
λ the second viscosity coefficient, Eq. (A.25) [Pas = Ns/m2 = kg/m/s]
λ (thermal) conductivity, Eqs. (1.6), (A.82) [W/m/K]
λ Taylor micro length scale, Sec. 7.2 [m]
µ the first viscosity coefficient, dynamic molecular viscosity, Eq. (1.3) [Pas =
Ns/m2 = kg/m/s]
µB = λ + 32 µ; bulk viscosity, Eqs. (1.4), (A.26) [Pas = Ns/m2 = kg/m/s]
ν = µ/ρ; kinematic molecular viscosity [m2/s]
ξ mixture fraction, Eq. (5.11)
8 = τi j ∂u i /∂ x j ; dissipation function, Eq. (A.44) [W/m3 ]
8i j redistribution term in the equation for Reynolds stresses, Eq. (6.2) [m2 /s3 ]
ϕ general variable, cf. Eqs. (1.1), (5.3), (5.8)
ρ density for the fluid (mixture) [kg/m3]
ρ
P k mass concentration (density) of species k, Eq. (A.11) [(kg)k /m3 ]
k summation symbol, the term should be summarized over all species k of the mixture
σ standard deviation, page 112
σ (= Pr) molecular Prandtl number, Sec. 5.6
σi j stress tensor (total), Eq. (A.22) [Pa = N/m2 ]; σi j = − pδi j + τi j
σh turbulence Prandtl number for enthalpy flux, Eq. (5.1)
σk turbulence Prandtl/Schmidt number for turbulence energy k, Eq. (3.6)
σT turbulence Prandtl number for temperature flux, cf. Eq. (5.3)
σY turbulence Schmidt number, Eq. (5.2)
σε turbulence Prandtl/Schmidt number for dissipation ε, Eq. (3.17)
σξ turbulence Prandtl/Schmidt number for mixture fraction ξ , Eq. (5.21)
σϕ turbulence Prandtl/Schmidt number for ϕ, Eq. (5.3), Sec. C.4
τ shear stress [Pa = N/m2 ]
τ = (ν/ε)1/2 ; Kolmogorov micro time scale, Eq. (8.21) [s]
τB = ηB2 /D; Batchelor time scale, Eq. (8.28) [s]
τc chemical time scale for flame, Sec. 10.2 [s]
τi j stress tensor (viscous), Eqs. (1.3), (A.24), (A.42) [Pa = N/m2 ]
τr chemical time scale for reaction, Sec. 10.10 [s]
τU = η/U ; convection time for microscale, Eq. (9.7) [s]
τU,B = ηB /U ; convection time for Batchelor scale, Eq. (9.8) [s]
τU,r = δr /U ; convection time for reaction length scale, Sec. 10.10 [s]
τw shear stress at a wall, page 67 [Pa = N/m2 ]
χ fraction of the fine structure that reacts, Eq. (11.27)
9 potential of distant force, Eq. (A.30) [m2 /s2 ]
ω characteristic frequency for turbulence, Sec. 11.2 [s−1 ]
List of symbols xv
ωk = Rk /ρ; reaction rate (production rate) of species k, Eq. (A.17) [(kg)k /kg/s]
Superscripts
· time derivative
′ fluctuating value, Eq. (2.1), Sec. B.1
′, ′′ derivative, second derivative
′, ′′ characteristic scale for turbulence
∗ fine structure in EDC, Chapter 11; small-scale turbulence
◦ reference state (0.1 MPa, 25 ◦ C)
◦ surroundings of fine structure reactor in EDC, Fig. 11.3
molar value (to e and h; [J/mol])
mean value, Sec. 2.2 and B.1
∼ mass-weighted average value, Sec. B.1
→ vector
+ dimensionless quantity, Eqs. (4.8), (5.31)
Subscripts
∗ fine structure in EDC, Chapter 11

eff effective
fu fuel
max largest value, Eq. (10.29)
min lowest value, Eqs. (10.29), (11.43)
ox oxidizer
pr product
t turbulence , turbulent (to p, e and h: total)
v viscosity, viscous
w wall
ϕ pertains to the quantity ϕ
Subscripts i , j , k, l are used for indices for coordinate directions, in certain cases also m, n, p, q.
See at page 219 about summation when indices are repeated.
For chemical species, usually k is used as index. When two indices are needed, i is also used. The
summation rule does not apply when the index denotes a chemical species.
Here, it may be an occasion to point out the distinction between subscript/superscript and index
(sub-/super-): A superscript/subscript – of Latin: «scribere»: write) is a sign that is put by or onto
another sign (high or low). Here, it is for instance a “t” that denotes turbulence viscosity and not
molecular viscosity, or “fu” to denote that it refers to fuel. It can also be an accent as in in é and ò, a
ring as in å, or a cedilla as in ç.
An index (from Latin: “index”: fore finger, cf. indicate, of “indicare”: point out, show, mention) is a
xvi List of symbols
subscript or superscript denoting a number, for instance a coordinate direction. From its position we
can distinguish between a low index (subindex) and a high index (superindex), although this is not
relevant in this book. When we, for instance, use general non-orthogonal, curvilinear coordinates,
the distinction between sub- and superindex is important.
The ISO standard (ISO 31-0) recommends to print subscripts with italic (slanted) type when they
represents numbers (i.e. indices) and with upright types otherwise. Physical quantities should be
printed with slanted types.
Chapter 1
Introduction and basics

1.1 Objective, background, overview
Books are supposed to be introduced with some words about the motivation to write and – not the
least – to read the book. All books are important – for the writer. Only rarely is it indispensable and
irreplaceable for the reader. This holds most likely for this book as well.
Combustion is important for us. Abundant access to cheap energy has been one of the prerequisites
for the societal development that we have experienced in our part of the world, with social and
technological development, welfare and prosperity for many. Utilizable energy comes mainly from
combustion. The other side is the darker side: Combustion is the origin of or contributor to serious
environmental problems such as air pollution, forest death and climate change. This harms health,
economy and ecology, and impose restrictions on future generations. Fire, uncontrolled combustion,
should not happen but happens anyway. This is a safety problem and an economical problem.
Combustion involves flowing gases, sometimes liquids as well. Such flows are often, although not
always, turbulent. In many cases without combustion, the flow is important by itself. Natural flows
like wind, rivers and ocean currents are turbulent, as are also the the flow in the main arteries,
ventilation air in buildings, filling of casting molds and drink boxes, the wake after a boat, and – said
short – most other flows.
Indeed, combustion equipment and other flow devices are mainly designed by people that have not
read books like this one – and the equipment does work. Someone may say that such successful prod-
uct development shows that theoretical knowledge is mainly for especially interested academician.
Others may see the same development and say that such ineffective and polluting energy conversion
calls for strong efforts to grow and disseminate fundamental knowledge.
Target group for the book: Students, researchers and engineers that are learning about and
working with turbulent flows with and without chemical reactions.
2 1 Introduction and basics
Background for the book, I: Nearly all technical and natural flows are turbulent. They affect
technical devices, environment, safety and economy. A very large part of air pollution comes
from combustion. Combustion processes dominate in the energy supply, and thereby prosperity
and welfare – and with effects on environment and hopes for the future.
Background for the book, II: Many have taken computer codes into application for calcu-
lating in detail on various flows, with or without combustion. Several such programs have
appeared, which with splendid and user-friendly user-interfaces and powerful computers be-
come good tools for the engineer.
However, a chain is not stronger than its weakest link. Without good insight into the physics
and good knowledge of the mathematical models one is using, the outcome can become really
bad when the results are brought into the real life.
Figure 1.1: Example of flow with visible turbulence. In the smoke plume, we can see eddies
that are about as large as the pipe stack outlet. The winding in the plume itself is due to eddies
in the air flow.
Purpose
This book is written to fill an open space between the existing books. Turbulence books are mainly
on nearly isothermal flows without combustion. Combustion books are often primarily on chemical
1.1 Objective, background, overview 3
Background for the book, III: The traditional method of analysis for thermodynamic systems
is to put up balances for mass and energy, maybe also for momentum and entropy, for major
parts of the system. These balances have then been solved more or less manually or with
simple numerical methods. Now this can be done with a much finer subdivision. Temperature,
concentration and velocity vary within each system unit – and these variations are important.
In principle, we use the same balance equations – however, the treatment is different.
reactions and have little or nothing on flows. On their own field, they are comprehensive and good on
analysis, but says not much about how the equations are achieved and interpreted. There are many
books on particular technical devices, such as furnaces, engines, gas turbines, wings, fans, venti-
lation, fire, ocean currents, atmospheric phenomena or pipeline flows. Such books carry detailed
knowledge and accumulated experience on their field but have not much room for the relations to
other fields or for the general, underlying theory that is common for such fields.
In this book, I have attempted on drawing together knowledge both on flows and on combustion and
put it together as a whole. It should be of help for the reader in understanding more of the physical
phenomena, models and equations. Then he/she will be more able to evaluate different processes
and the numerical results that, nearly said, anyone might produce from a purchased program. It has
also been an aim to draw knowledge out of academic circles – with the associated language – and
bring it forth in a form that is more accessible for those that makes use of it.
The book is intended to give a scientific basis of turbulent flow with and without combustion. On
this basis, one can build further in project and master thesis work, doctoral or other research work,
and in industry and civil service.
The course attempts to give insight and understanding of fundamental physical phenomena that
affects environment, safety and economy. Understanding of these processes also gives a greater
understanding of other phenomena. As most flows are turbulent, this is relevant for everyone that has
to do with gases or liquids in motion. Know-how on this area gives a basis for design of equipment
and processes. It gives a better understanding for the processes and problems others are working
on. This is important both for the one that is conducting the task, for the one that is a cooperating
partner, and for the one that is doing inspection.
Some readers will, for different reasons, not be working further with turbulence and combustion.
For these, the course will give insight into physical phenomena, problem specifications, terminology
etc. that enable them to communicate and cooperate with experts on the field.
In short, the course is about underlying physical processes that most people are in touch with.
The subject: Virtually all flows are turbulent. Turbulence is many small and large eddy mo-
tions in the flow. These motions work to mix different chemical species, even out temperature
differences and change the forces acting in the flow. Combustion, heat transfer and spreading
of substances are to a large extent determined by the turbulence in the flow. In this book, we
will look closer at turbulent flow and combustion as a physio-chemical phenomenon. We will
also look at how such flows can be handled and analyzed in an engineering context.
Target group
I see a four-fold target group for this book: 1) Students, researchers and engineers who take a
course as part of a degree program or as further education and scientific development. 2) Engi-
neers/researchers who work on this field and use the models and theories that are discussed here.
The cross references and index make the book easy to use as support and reference. 3) Engi-
neers/researchers who work on other fields, who are cooperating with those mentioned under 2).
4) Employers for 2), 3) and possibly 1). The grouping is not absolute and most readers will over
time enter two or more of the groups.
The main target group is those working on this or closely related fields. This book deals with the
central issue of nearly all process engineering: The interplay between motions and forces, chemical
reactions, heat and thermodynamical properties. With this outfit, you should be well prepared to
make use of more specialized knowledge and solve the concrete problems you will be working on
in the future.
Within a project or task, there is often no room for an extensive explanation of theory and models
that are used. Both the project proposal/offer and the report should contain something of how the
work is intended to be/was conducted. This often becomes a short version in a “tribal language” that
is not readily comprehensible for outsiders.
Terms as model, simulation, k-ε, Reynolds stress, Eddy Dissipation Concept, flamelet and so on,
easily become senseless labels for those that are working outside the area. Through this book, one
can with some efforts get some insight into what the project workers actually were busy with. Also
for the project employer, the outcome will be greater when he knows more about the matters.
There is, of course, an ulterior motive for the author, to push engineers and researchers to read the
book carefully. You might otherwise risk that the employer knows more about it than you that is
supposed to be the expert.
The models are often fully implemented (by someone else) in a larger computer program. The user
of the program should of course have a good knowledge of theory and methods. The intention is that
this book could give a good basis for this.
Purpose of the book:

Provide knowledge, insight and understanding of
– turbulent flow and combustion as physical phenomena,
– methods for analyzing such phenomena,
– the basis for and development of the models available in the computer programs in
widespread use.
The book attempts on build a bridge between technology and theory by using basic knowledge
in methods for solving practical problems.
1.1 Objective, background, overview 5
Scientific foundations and relations
This book has to be seen in relation to other books and subjects, primarily numerical heat and mass
transfer and combustion theory. The presentation here requires a certain basic knowledge of fluid
mechanics and thermodynamics, including chemical reactions. There are quite a few books that
cover this basis. In the teaching at NTH/NTNU in recent years, we have used Moran and Shapiro
(1993,1998) or Sonntag and VanWylen (1982,1991) in Engineering thermodynamics, Mills (1995),
Bejan (1993) or Incropera and DeWitt (1990) in Heat and mass transfer, and White (1986,1994) in
Fluid mechanics. But as said, there are many books on these subjects.
It is hard to specify a minimum requirement, and equally hard to make a list of “useful” previous
knowledge. Particular insight in one area may compensate lack at another; and a good mind, strong
drive and zeal for work can mend many holes.
The most important basis from fluid mechanics and heat and mass transfer includes
– the equations for mass (continuity), momentum and energy on differential form and on integral
form (control volume); development, interpretation and understanding,
– viscous flow and boundary layers,
– conductive/diffusive transport; gradient models, Fourier’s law, Fick’s law.
From thermodynamics, chemical reactions/combustion is an important issue.
Possible deficiencies can be balanced with other relevant knowledge – either theoretical or practical
– or with more eager reading during the course.
One also have to see this book in relation to books on closely related subjects. A book on combus-
tion theory, e.g. Warnatz, Maas and Dibble (1996, 1999), goes further into the details of combustion
chemistry. A book on numerical (computational) fluid dynamics, e.g. Versteeg and Malalasekera
(1995), describes methods of solving the equations appearing in this book. A third subject is exper-
imental methods. A book on this is Doebelin (1990).
This book is built around the research activity in the group of Bjørn F. Magnussen at NTNU/Sintef
Energy Research, and the selection of material is colored by this. In two-three decades this group has
been working on modeling and simulating turbulent flows with the aid of computers. The group has a
particularly large activity on combustion problems and Magnussens’s combustion model (EDC). We
also work on flows within and around constructions and other devices, and spreading of substances
and heat in the atmosphere. For these purposes, we have mainly been using a “standard” k-ε model
to represent the turbulence.
In many instances, the k-ε model gives a sufficient account of the turbulence; in other cases we solve
the modeled equations for the Reynolds stresses. (There, the “tribal language” appeared again –
you have to read more to learn what the vocabulary words mean.) The models for other physical
phenomena such as reaction rate, radiation, soot formation and heat transfer can be more uncertain.
Relations and foundation:

Turbulent flow and combustion are woven together with other subjects such as numerical and
experimental methods, mathematics and chemical kinetics. We build upon fundamental know-
ledge from thermodynamics and fluid mechanics, from mathematics, physics and chemistry,
and bring this forth to solve practical problems.
In some cases one wishes to use a better turbulence model. It might be that the turbulence is weak
(low Reynolds number), that directional source terms are important, or that the turbulence viscosity
is not satisfactory for directionally-dependent transport properties. When you use a model, it is also
important to know the background and presumptions of the model. And the more complex a model
is, the more knowledge is needed.
Overview of the book
The two first chapters contain introduction to the subject, some fundamental material and various
topics that are useful for further reading. We will look at what turbulence is, what modeling is,
and at some issues on transport and mixing in turbulence. Some parts of the chapters should give
practical information that is useful in the course, for instance notation of equations, averaging rules
and chemical reactions. Other parts of the chapters attempt on bringing understanding of (or, if you
like, philosophy on) treatment of turbulence and combustion, why we do as we do, and what we can
expect. Some parts will be easier to understand when you have read material later in the book.
In Chapter 3, we will develop an equation for turbulence energy and eventually the so-called k-ε
model. This is a widely used turbulence model in research and industry. In Chapter 4 we look at
simple turbulent flows. These are used for model testing and for studying diverse phenomena in the
turbulence. For a boundary layer flow, we can develop simple models, wall functions, that are used
as boundary conditions against solid surfaces.
Chapter 5 deals with modeling of energy and mass transfer. We will discuss transport equations for
scalar variables like energy and concentration. In the equations, the turbulence diffusion and possible
source terms have to be modeled. We look at conserved scalars and define the mixture fraction.
Chapter 6 treats a group of more advanced, and potentially more general, models for turbulence:
Models where equations for the turbulence stresses are solved. These models are somewhat used in
research, and are also being introduced to industry.
In Chapter 7 we go through some statistical functions and how they are used in turbulence.
Chapter 8 describes the spectrum of length and time scales in turbulence. We will look at the break-
down of turbulence to smaller and smaller eddies. This is important for combustion problems as the
transport is mainly carried out with large turbulence scales, while the smallest scales are important
for the molecular mixing and thus for the reactions.
Chapter 9 is a collection of diverse topics. Some of it is summarizing items from different preceeding
chapters. In addition, we will look at different principles for modeling of turbulence and at how to
do simple estimates for turbulent flow and combustion.
Chapter 10 treats models for turbulent combustion. Eventually, in Chapter 11 I will present Mag-
nussen’s combustion model (the Eddy Dissipation Concept, EDC) in more detail. In this chapter, for
the first time, a complete development and explanation of the model expressions of EDC is given.
The appendix is a listing of equations, relations and definitions we make use of in this subject.
Appendix A contains the basic equations and definitions for a number of physical quantities. I have
tried to do this as accurate as possible, so that all terms and all presumptions are included. You
should then be able to deduce the exact equations that applies to the thinkable and (for the time
being) unthinkable special cases that you might decide to work on. Appendix B treats the formal
1.2 About reading this book 7
rules for decomposition into mean field and turbulent field. This is done with conventional averaging
(Reynolds decomposition) and mass-weighted averaging (Favre decomposition). The rules are used
to develop equations for the averaged quantities from the basic equations.
In Appendix C, I present some of the most usual dimensionless parameters. Appendix D contains a
dictionary with explanation of some words that we use in this area and related subjects. At the end
of the book there are a literature list and an index.
In turbulence modeling, the focus is normally on the transport mechanisms and large time and length
scales in the turbulence. This is what determines the mean field of pressure and momentum (mean
velocity) that we seek. The combustion occurs where reactants are mixed molecularly, and this
is mainly in the smallest scales of turbulence. Therefore, we will look at the entire spectrum more
closely than in dedicated turbulence courses. In combustion theory, the emphasis is on understanding
the chemical processes, which are directing the conversion of mass and energy. Chemical reactions
are often discussed in simple flows, or without actually considering a flow at all. In practice, com-
bustion is directed by both flow and combustion, and here we will treat both in relation to each
other.
1.2 About reading this book

You that read this book have read other books before. I can hardly change the good and bad habits
you have adopted. Just you know the best how to approach such a book. If I gave you advises on
this, you would most likely not follow them. Therefore, I do not expect that everyone reads the
chapters in sequence. Actually, the truth is perhaps that nobody does so.
To my best ability I have tried to find a suitable progress and organization of the material. This, of
course, means that much of the matter is based on previous sections. However, much will become
easier to understand when you have read some things further back in the book as well. I have been
careful to include cross-references in the text, both back and forth. The intention is not that you
should look up all. They are there when you need them.
Many equations – little mathematics
Beyond the usual algebra and derivation, there are a few integrals and some logarithms in this book.
Not worse than that. We use one type of differential equations: transport equations, see Eq. 1.1. What
this is, and what it tells, I think will enter your mind in due time. In Chapter 8, I have deliberately
put a limit by keeping Fourier transformation outside, cf. Eq. (8.4). This can be a useful tool to gain
further understanding of turbulence, although the reader have to wander his own way; the literature
is rich. The same holds for techniques on numerical solution of equations. Here, I leave the reader
to himself and to other books.
Those who have got a distaste for “curled derivatives”, hardly keep up to read this paragraph. The
book may seem to contain quite a few equations and mathematical expressions. That is right. Math-
ematical language is also a language, and it is in fact a part of your professional language. I have
chosen to use some mathematical expressions in the text, either together with or in place of literal
words. Call it a bad habit if you like, but you just have to get used to it. Mathematical expressions
can often carry the meaning more unambiguously, and they make the presentation shorter and more
Transport equation: All the differential equations in the book (with one exception) have this
form:
∂ ∂ ∂
(ρϕ) + (ρϕu j ) = − jϕ, j + Sϕ . (1.1)
∂t ∂x j ∂x j
This is the balance equation or the transport equation for some quantity ϕ. It can be mass,
momentum, energy, or a statistical quantity. From the left, the four terms represent storage of
ϕ; convective transport of ϕ, that is what is carried with the flow; diffusive transport of ϕ and
source term. The diffusive term is often on the gradient form − jϕ, j = Ŵϕ ∂ϕ/∂ x j , where Ŵϕ
is the diffusion coefficient for ϕ.
easy to follow.
In some instances, I have chosen to use many equations to describe a development. In particular so
in Chapter 11, which contains developments that you hardly find anywhere else. I could also have
done as many others: “from A you can develop B”, or with the remark “the reader should do the
development as an exercise”. Indeed, you would have learned more from that – if you had done as I
said.
This gives an opportunity to tell about G.I. Taylor, who held a lecture for the Royal Meteorological
Society (Taylor, 1927): He started telling that the president had invited him to explain turbulence in
a non-mathematical language. He had resisted, but the president had persuaded him; it should be
“turbulence with no tears”. This had occurred in a college feast, and the blame for the following had
to be shared between himself, the president and the wine cellar. Without mathematics – he meant –
he could hardly tell the audience anything they did not already knew.
Literature
I have included quite a few literature references. The intention is not that you should look up all
of them, but that you should find use of the book later. The primary purpose is that you might go
further and deeper into particular topics later on. In project and master thesis work, or doctoral work
and other research, you will have to go closer into matter that we here treat relatively superficially.
Then, it is convenient to have some important references at hand.
The literature that is referred contains many references itself – which you can use as road signs into
the literature. Moreover, you can use readily available databases to find articles that refers to those
you already know. In this way, you easily find the latest development in your field.
I use the author/year system, because I like it. For books, and sometimes long articles, I include
page number as well. In my experience, a reference only to a book without page number has nearly
no value.
I have made efforts to include reference to the original literature, to the publications where various
thoughts and models were first presented. I have also tried to include relevant review articles. But I
have not tried to include the latest development. One reason for this is that “the latest” after a short
while no longer is the newest news. Moreover, it often takes some time and efforts before we see
whether these “news” actually are fruitful thoughts and whether it really leads to something useful.
Only in retrospect we can see the directions taken and what the research led to.
1.3 This section you do not need to read 9
Here, it should perhaps be a list of “literature for further reading”. Most likely, we could divide the
list into two: Books and articles 1) that you would never read, and 2) that you would find anyhow.
1.3 This section you do not need to read

Here, I will give you some warnings, which are only bother and trouble to read. Since you still are
reading, I suppose it means that you are curious.
(1) You should not believe that I, who write this book, am an expert on what I write about. (I will
not comment on other writers, you will have to do this on your own.) Turbulence, combustion, mass
transfer and heat transfer are subjects where the amounts of literature increase faster than anyone is
able to read. (This is supposedly the case in many subjects.) Sure enough, the knowledge does not
grow proportionally to the pile of paper, but anyhow . . .
Gradually, as you learn more of a subject, you also see an increasing amount that you knows little
or nothing of. Then, the challenge is to keep a good balance between using what you know and to
learn more of what you do not know.
(2) You should not believe that what is said in this book is the “truth”. What stands here is a
viewpoint, an angle, an interpretation, . . . . . . (find a word yourself). It could be presented in a totally
different way, with a different selection of material, with other interpretations, and so on. In some or
another way, one can find contradictions, and one can question the presentation.
Someone will see the reality in a different way. You will find excellent scientists who think this
presentation leads you the wrong way. – And you will find (I hope) equally excellent scientists who
think that this at least is in the right direction.
(3) You should not believe this is important. You can be a good human anyhow, a good parent, em-
ployee, personnel manager, politician, quarreller, – or whatever ambitions you have. It is a question,
however, whether you can be a good process engineer without a certain understanding of what tur-
bulence is about. This book is intended to give you such an understanding, and some more, although
there are also other means of getting it.
(4) Here, a warning from Lesieur (1990:vii) is appropriate:
Turbulence is a dangerous subject which is often at the origin of serious fights in the scientific
meetings devoted to it since it represents extremely different points of view, all of which have
in common their complexity, as well as an inability to solve the problem. It is even difficult to
agree on what exactly is the problem to be solved.
On the run, I include some words of Chorin (1995):
A quick review is given of some methods available for turbulence modeling. The review is
pessimistic about the present, optimistic about the future, and biased in favor of the author’s
work. [...]
An engineer who has to model turbulence (in particular turbulent combustion) can find thou-
sands of papers that offer modeling strategies. These strategies are often incompatible, often
very complicated and even incomprehensible, and none can be sanely viewed as both reliable
and practical.
The uninitiated observer may find this situation surpriseing. In the last decades turbulence has
attracted attention not only from engineers, but also from mathematicians, physicists, and the
New York Times. Every advance in chaos, fractals, dynamical systems, renormalization, as
well as in the size, speed and parallelization of computers, is heralded as having solved the
“turbulence problem”. [...] Nevertheless, despite the advances, the goal of making reliable
predictions in practical problems has not yet been reached.
(5) Some words primarily to those who are, or is going to be, doctoral students:
As your work progresses, you learn and understand more of your subject. As mentioned, you will
also see more and more that you do not know much about. But you also see clearer and clearer how
you should have set up your work from the beginning. After some years, when your scholarship is
approaching the end, you will become more and more displeased with what you have done, because
you know (you think) something about how it could have been done. Take it easy, this is quite
normal. It indicates that you are maturing scientifically.
The next stage is that you start rejecting what you have achieved. The work is too poor and outdated,
the supervisor is incompetent, and so on. This is also normal. Now you have to keep your mind cold,
complete the writing and submit what you have. We have some experience that those who come too
far into this stage, tend to have some difficulties in submitting their theses.
✬ ✩
(6) Those who are writing something, in English or other languages, may
have good help of one or more of the many books that give guidance in
such exercises. Messy language and foggy text is often caused by messy If no thought
and foggy thoughts. Writing is a process that straighten up both text and your mind does visit,
mind. make your speech
Those who are writing, but without getting started, may perhaps find use not too explicit.
– or at least comfort – of the child book Dustefjerten (in Norwegian) by
Piet Hein
forth, about the author who buries his drafts in the garden and comforts ✫ ✪
Rune Belsvik (1991). The first part is about the story that does not come
himself by eating marzipan.

(7) I have the belief that humans both learn and teach the best in their own language. But knowing
other languages is both useful and necessary. Unknown words makes a text less accessible. Those
who understand the words that are used, also understand more of the contents of the text. On the
other hand, a Norwegian writer can not be blamed when a Norwegian reader have holes in his
Norwegian vocabulary.
And, if you wonder: In this book, except the dictionary and the list of literature, I have (that is
in the Norwegian version) used something more than 5000 written forms (graphemes) – then all
grammatical forms, compounds and personal names counted, but not mathematical expressions. In
Appendix D (dictionary) about one thousand more are added.
I use to say that: If you are writing about something that you do not know anything about, you
should write it in a language you do not know. Quite a few of us like to pick up foreign words and
phrases, in particular from English (and then with Latin origin). Some do this by mere laziness or
lack of knowledge. But if we borrow the word, it does not mean that we understand it; rather on the
contrary. Therefore, I have made a dictionary for this subject, see Appendix D.
Aasmund Olavson Vinje (1818–1870) writes (somewhat ironically) that I am not blind for the ad-
vantage of such a learnedly style; it awes or impresses quite a few readers, who think that such men
1.4 Combustion 11
must be immensely scholared, who write shuch a learned language (Vinje, 1867).
(8) A final warning, as you still are reading on this section: Setting on work on this subject, tur-
bulence and combustion, you expose yourself for a great danger. You will risk becoming interested
in a fascinating subject that may take unbelievable amounts of your time and strenght. No-one can
know where you then end up.
1.4 Combustion
Energy and environment
The greater part of energy use in the World, about 95 %, is converted through combustion. This is
an important source of man-made pollution. Some pollutants come from the fuel itself, for instance
sulphuric and chloride compounds. Other pollutants are caused by poor and incomplete combustion,
for instance soot, unburned hydrocarbons and carbon monoxide. Nitrogen oxides partly originate
from nitrogen in fuel and partly from reactions involving nitrogen of the combustion air.
In recent years, man-made climate changes have come into focus. Nearly all fuels contain carbon,
and large amounts of the greenhouse-gas carbon dioxide is formed by combustion.
Pollution can be reduced by measures before, within and after the combustion. Some substances can
be removed from the fuel, the combustion process itself can be improved, and the flue gas can be
cleaned. Carbon dioxide can hardly be avoided. We can, however, utilize a greater part of the energy
of the fuel, improve the efficiency. Some think of depositing produced CO2 in oil/gas-wells, or to
use it as a raw material for industrial processes. One can also think of separating carbon from oil
and gas, and only burn the hydrogen fraction.
Figure 1.2: Why do we put stones around the fireplace?
Combustion is a physio-chemical phenomenon. Improving the combustion requires knowledge of

the chemical and physical processes that take place. The stone age man most likely learned to
place stones around his fireplace. This gave primarily more stable combustion and slightly better
utilization of the fuel, and may be somewhat less soot in the cave. With the knowledge of today,
we can say that he made an isolating heat reservoir that emitted heat back into the flame. Thence,
he increased the flame temperature and made better use of the fuel. The stones could also give
zones of inhibited air draft. In such wakes, combustion products can re-circulate and stabilize the
flame. In our time, we have greater knowledge, although there is still much that can be done with
the combustion, and there is a lot that we do not know.
Table 1.1: Some examples of energy intensity and fuel conversion (after Bradley, 1992)
Energy conversion Fuel conversiona
rate (MW/m3 ) rate (kg/s/m3 )
Fusion, continuous (expected 2005; NIF project)b 1024
Fusion (achieved; Nova-project) b 1012
Electric spark 104
Laminar hydrocarbon flame in air (peak value) 6000 130
Gas turbine, airplane (take off), primary zone 4000 90
– overall combustor 1500 35
Gas-fired boiler 200 4
Gasoline engine 100 2.5
Pressurized-water nuclear reactor 100
Gas-fired fluidized bed 40 0,8
Cyclone combustor, coal dust 10 0.4
Firec 0.5 0.01
a calculated with typical heating values
b information from Lawrence Livermore National Laboratory, California, January 1999
c after Cox (1995:5)
Technology
By the word technology, we mean use of scientific results or methods on a certain area to achieve
certain goals, and methods and tools applied for this purpose. The aim is often to convert a fuel into
power and/or heat. Examples are gas turbines, engines, furnaces (with burner, grate, bed or similar),
ovens and stoves and burners for direct heating (for boiling, for greenhouses). The aim can also
be light, e.g. gas lights. The aim can be to convert a fuel into more stable end-products for safety
reasons. Burning of gas in a flare is an example of such controlled, although actually undesired,
combustion. Fire is uncontrolled and undesired combustion. Here the aim is to detect the fire and
limit the damage, or preferably, avoid the fire.
Combustion technology is the sum of the equipment, procedures and knowledge that are used to
achieve these goals. Equipment may be “off-the-shelf” or designed for the specific case. Thermo-
dynamics, fluid mechanics and chemistry are knowledge of conversion of fuel into heat and power,
and to formation and spreading of pollutants.
Seen from outside there are substantial differences between flares, fires, furnaces, burners and re-
ciprocating engines. Seen from inside, however, there are no fundamental differences between com-
bustion in the cases mentioned. The physical and chemical processes taking place are the same.
Therefore, the methodological approach can also be the same, even though the geometric differ-
ences are large.
1.4 Combustion 13
Combustion as a physio-chemical phenomenon
By combustion we usually mean an exothermal reaction between a fuel and an oxidizer. The oxidizer
is usually, but not always, oxygen in plain air. The fuel can be coal, oil, gas, wood or waste. In a
fire, the fuel can be any combustible material.
Combustion requires reactants to be mixed molecularly, the temperature to be sufficiently high, and
this situation to prevail long enough.
When a fuel reacts with oxygen, it occurs through hundreds (thousands) of simple partial reactions
where hundreds of components take part. The reaction rates in the different partial reactions depend
on the local temperatures and concentrations of the participating components.
Some reactions are very fast and take place as soon as the reactants get close to each other. We say
that they have a short time scale. Other reactions need much longer time at appropriate temperature
and concentrations (long timescale). If the residence time at reactive conditions is too short, the re-
actions are inhibited or halted (frozen). Unburned hydrocarbons and CO come from such incomplete
reactions.
The requirement of molecular mixing makes the mixing processes as important for combustion as
the reactions themselves. Most flames are turbulent, which means that the turbulence of the flow
determines the mixing and, hence, the combustion.
Since the temperature is important for the reaction, it is also important how the heat is released
and transferred within the flow. Strong radiation from the flame cools the flame. Cooling surfaces
(heating surfaces) near the flame will lower the temperature. If the flame is contained in a heat-
isolated chamber, the surrounding chamber will radiate towards the flame and keep the temperature
high. Much radiation is emitted from soot, while soot is also shielding the flame by absorbing
radiation.
The local temperature determines how the reactions take place. Nitric oxide (NO) often comes from
tiny regions with high temperature. Higher efficiency often requires higher temperature, but then
it is a temperature that is averaged over a larger volume. By controlling the heat transfer, the peak
temperature can be reduced even though the mean temperature is increased.
The physical processes can be expressed mathematically through balance equations for mass, en-
ergy and momentum. They correspond to the continuity equation, the first law of thermodynamics
and the 2nd law of Newton. These equations include terms for chemical reactions, mass transfer,
heat exchange and forces. We like to talk about modeling (“imaging”, “picturing”) of the various
processes. How these balances are formulated and how they are used, depend on what is the aim of
the work. Subsequently, we will come back to the relevant procedures.
Fuels can be in solid, liquid or gaseous form. Liquids evaporate and large fractions of solid fuels
is gasified by heating. Eventually, a residual char remains and reacts directly as a solid substance.
With this exception, combustion occurs in gaseous flames.
(a) Schematic presentation (b) Experiments by Gaunce (1939, unpublished)
Figure 1.3: Investigation of laminar and turbulent non-premixed jet flame. The length of a
laminar flame increases when the flow velocity increases. When the flame becomes turbulent,
the mixing rate is increased and the flame becomes shorter. The length of a turbulent flame
is approximately the same. Higher velocity enhances the turbulence and the mixing, so that
the fuel burning rate increases as fast as the fuel outlet flow. Both figures are from Hottel
and Hawthorne (1949). From a certain outlet velocity, the flame begin to lift off the outlet,
as indicated on the figure to the right. Eventually, the flame will blow off; cf. also Byggstøyl
and Magnussen (1983).
Models and methods for combustion
Above, we visited the stone-age man who placed his fireplace at a stone. He might have noticed
that this had an effect and tried it again later. He found it helpful to put more stones by the fire, and
taught this to others. Even though the practice is different, the principle is still the same. We try, see,
think, try again, learn and teach. With other words (Greek/Latin terms) we talk about experiment,
observation, analysis, hypothesis, verification, method and competence.
To our time, combustion technology has been developed by continual trial and error, new trial in
practice and numerous small, stepwise improvements of the equipment. Experience is gathered over
a long time. We know that mixing and turbulence are important for the combustion. Then, one
have made the equipment to produce turbulence. If the turbulence and mixing become too weak,
one can introduce an obstacle that enhances the turbulence of the flow, a turbulator. Cooling gives
poorer combustion. Therefore, one let low-calorific fuels burn in isolated combustion chambers and
rather extract the heat from the flue gas after the combustion. High-calorific fuels may be burned in
chambers with heating surfaces. Experience tells how much isolation that is required or how much
heat that can be extracted.
As time proceeds, greater efforts are required to continue with small, stepwise improvements. Trial
again and over again becomes very expensive. Here, theoretical knowledge is helpful. We have
mentioned temperature and flow (turbulence). A representative outlet temperature can tell the most
important things about the energy conversion of the combustion. To tell about pollution, we need
1.5 Turbulence 15
knowledge of the peak temperature in the combustion region. Detailed knowledge of temperature
distribution, concentrations and other quantities in a region can be obtained from measurements.
This can, however, be complex and expensive.
Combustion processes follow, as mentioned, certain physical and chemical laws, which can be for-
mulated as a set of mathematical expressions. If we have sufficient knowledge of these laws, we
should, with enough computer power, be able to calculate the values of all the physical quantities.
Physical laws and theoretical models have to be formulated on the basis of specific knowledge of
the combustion processes. Such knowledge can be acquired by experiments and by general physical
relations. Still it is far ahead to a full knowledge of the physics and chemistry of combustion. Also
techniques and computers are still too weak. Nevertheless, we can today model and analyze many
practical problems. In that way, we can draw useful conclusions with respect to design and operation
of a variety of combustion devices and systems.
Laminar flows with no reactions can be simulated with no other models than the molecular transport
models (e.g. Newton’s stress model, Fick’s law and Fourier’s law). Simulations of simple turbulent
flows without special turbulence models (direct simulation) are also conducted. For simple fuels
(up to 3 C-atoms), one can do a detailed simulation of the chemical reactions in combustion. With
a C/H/O/N-system, it is a matter of thousands of partial reactions (elementary reactions). When it
comes to more complex flows and/or fuels, we have to turn to simpler models. Beside flow and
reactions we will face challenges related to radiation, soot, phase boundaries and phase transfer.
Weinberg (1975) describes the combustion history from the stone age to our time. He sees two
separate lines in the development. One is the “technology”, where one tries out something to see if it
works. The other line is “science”, where one studies combustion as a physio-chemical phenomenon
and tries to understand what takes place. The use of the terms can be discussed, in particular as the
methods and approaches in “technology” can be as scientific as in “science”. Moreover, researchers
in “science” have become more aware that they are working with models, not with attempts of finding
the final “truth”. The difference may mainly lie in the aim. One line aims at creating knowledge and
equipment that can be used here and now. The other line attempts to understand the fundamentals of
the phenomena. How this basic knowledge can become “useful”, one knows after long time.
Until today, the two lines have had modest contact. Fireplaces for heating and cooking and furnaces
for metal smelting were developed in a time when fire had only mythological explanations. The
steam engine was developed in a time when natural philosophers discussed phlogistons and basic
ideas from antiquity. Gas turbines, Otto and Diesel engines came forth without noticeable contribu-
tions from the academic sciences. However, to solve or make up for the considerable problems that
relate to combustion, knowledge from both lines has to be utilized. This book belongs to a tradition
that tries to set up a bridge between science and technology.
1.5 Turbulence
The terms turbulent and turbulence belong to everyday language and are used about most issues from
stock market and politics to air travel and flames. When something behaves volatile and agitated and
with large variation, it is referred to as turbulent. A general dictionary explains turbulence, turbulent
as “Characterized by violent disturbance or commotion; violently disturbed or agitated; disorderly,
troubled.” Like many other words, turbulence comes from Latin and is spread to other languages. In
Latin, the word turbulentus has a wider meaning than in modern English (and Norwegian). All the
meanings have in common that something is or is made agitated.
Leonardo da Vinci (1452–1519) was – as far as we know – the first one to systematically investigate
eddying motion in water. He studied and explained formation of eddies and the distribution of
velocity in eddies at and under the surface (Rouse and Ince, 1957:47). It seems that Leonardo
was interested in various eddying motions in water and air, both as an artist, researcher and water
engineer. It also seems that he had ideas of a relation between eddies and heat release. There exist
sketches of a floating marker that Leonardo designed to measure eddying motion at different depths.
Furthermore, he used seeds, dyes etc. to investigate flows in air, rivers, channels and in water tanks
with a glass wall.
Three centuries later, in 1797, another Italian, G.B. Venturi, issued a thesis on fluid motion. Among
other things, he described how eddies were formed, disperged and degraded in certain flows, and he
discussed diffusion caused by eddying motions in duct flows and in the atmosphere.
Gotthilf H.L. Hagen (1839) studied flows in pipes and found a relation between pressure loss and
mass flow rate (flow velocity). This relation was, however, valid only to a certain limit. Then, the jet
flow from the pipe became unsteady and pulsating. In the following 15 years, he proved two types
of flow, what we today call laminar and turbulent flow. He found that the limit between the two
varied with pipe flow, velocity and temperature but was not, however, able to determine a general
criterion for the shift between the two types of flow. It was Osborne Reynolds (1883) that found a
general parameter for the transition from direct motion to sinuous motion, as he calls it. The word
turbulence, for the flow regime, is supposed to be first used by William Thomson (lord Kelvin) in
1887 (Rouse and Ince, 1957:212).
In order to account for the turbulence, Boussinesq (1877:46f) suggests to use the equations of mo-
tion, although with a viscosity that is larger than the molecular. Reynolds (1895) introduces a statis-
tical decomposition of velocity and pressure and deduces equations of motions for turbulence. These
have turbulence stresses in addition to the viscous stresses (see Sec. 2.2).
Figure 1.4: Sketches by Leonardo da Vinci.

1.5 Turbulence 17
Figure 1.5: Eddying motion in water. Drawings by Leonardo da Vinci.

Figure 1.6: Investigation on motion in water, a thin streak of dyed water is led into a pipe
with clear water. Facsimile from Reynolds (1883).
1.5 Turbulence 19
In the following years to 1940, in particular three names are visible in the turbulence literature:
G.I. Taylor, L. Prandtl and T. von Kármán. They created much of the foundations that later work
is built on. Prandtl developed the boundary-layer analysis, and all three proposed mixing-length
models for turbulence viscosity.
Also L.F. Richardson should be mentioned, especially known for studies of turbulent flows in the
atmosphere. Somewhat ahead of time, he also developed a basis for numerical simulation of such
flows by finite differences.
At the end of his book, he writes a chapter on “Some remaining problems”. One of the problems is
the difficulties of conducting all the calculations. Today, this is a fascinating section to read. This
was fifty years before electronic computers seriously entered the area, and 25 years before their
existence. “Let us play with fantasy”, he says, and outlines how a central weather forecast factory
could be organized, with hosts of people conducting numerical calculations by hand (Richardson,
1922:219f).
Taylor (1927) describes turbulence as:
Turbulence is an irregular motion which in general makes its apparence in fluids, gaseous or
liquid, when they flow past solid surfaces or even when neighbouring streams of the same
fluid flow past or over one another.
Hinze defined in 1959 (after Hinze, 1975:2):
Turbulent fluid motion is an irregular condition of flow in which the various quantities show
a random variation with time and space coordinates, so that statistical distinct average values
can be discerned.
Bradshaw (1971:17) defines:
Turbulence is a three-dimensional time-dependent motion where vortex stretching spreads
velocity fluctuations over all wave lengths from a minimum given by the boundary conditions
for the flow. This is the usual state of flow except at low Reynolds numbers.
Not everyone is satisfied with the definitions above. Lesieur (1990:2) wants to include some two-
dimensional flows. He emphasizes the random and uncertain (chaotic) aspects of turbulence. And
he wants to exclude flows where turbulent transport do not dominate, hence, that the turbulence
Reynolds and Peclet numbers should be much larger than unity. His attempt on defining turbulence
is:
– Firstly, a turbulent flow must be unpredictable, in the sense that a small uncertainty as to its
knowledge at a given initial time will amplify so as to render impossible a precise deterministic
prediction of its evolution.
– Secondly, a turbulent flow should be able to mix transported quantities much more rapidly
than if only molecular diffusion processes were involved.
– Thirdly, a turbulent flow should have a wide range of spatial wave lengths.
This definition is more narrow than the former. For instance, the whirling flow close to a wall or
dying eddying motion in a wind tunnel cannot be called turbulent according to this definition –
since molecular transport and mixing becomes important and the spectrum of wave lengths becomes
shorter. The term low-Reynolds-number turbulence model loose meaning – as the flow according to
Lesieur’s definition is turbulent only at high Reynolds numbers.
In everyday work, we can do without a precise definition of turbulence. But what is referred above
tells a lot of what turbulence is. It is a matter of motion; that the turbulence is a phenomenon related
to the flow and not to the fluid. The motion is strongly diffusive; we obtain good stirring/mixing.
The motion varies in time and space, but we regard that we can distinguish statistical mean values.
The length scales, the “diameters” if you like, of the eddies in turbulent flow are distributed over a
wide range. The ratio between the largest and the smallest scale is several decades. We also assume
that the length scale of the turbulence is much larger than the length scale of the molecular motions.
In an engineering context, virtually all flows are turbulent. This can be flow inside pipes or other
devices, or flow around them. Knowledge of the turbulence is decisive in nearly all engineering
flow and combustion problems. Some examples are combustion in gas turbines and reciprocating
engines and formation of pollutants from the combustion; gas explosions; flow and heat transfer
from pipes and heat exchangers; meteorology and spreading of substances in the atmosphere, rivers
and ocean. We see that turbulence relates to energy conversion, safety for humans and pollution of
the environment.
To obtain knowledge of what takes place in technical devices, we can simulate the flows numerically.
Usually, we are interested in forces between fluid and solid parts, and in mass and heat transfer. With
chemical reactions, we are also interested in the mixing of different substances and in the species
that are formed.
If the flow is not turbulent, we can determine these things by solving the basic equations, that is the
equation of state, equations of motion, energy equation and so on. These equations are valid for
turbulent flow as well. However, in order to catch the smallest scales of a turbulent flow, we have to
put a very fine-meshed grid over the calculation domain. This is possible only for very simple flows;
and then only at low Reynolds numbers, that is, when the ratio of the largest to the smallest scales is
not very large.
Present-day computers are not sufficient for the more complex flows, and we have to introduce
simplifications. One way is to apply a grid with wider mesh size, where we solve equations for
averaged values of pressure, velocity (momentum), energy and species concentrations. Now, we
need a relation between these averaged values and quantities that characterize the turbulence. This
we call a turbulence model. Such a model can be more or less complex and more or less general.
A simple model is easier and cheaper to use than a more complex, but is valid for a more restricted
region. In many cases, the simple model is the most accurate for just the type of flow that it is
developed for.
We can imagine that in some years, we will have computers with memory and power to simulate
turbulent flows fully down to the smallest eddy in many engineering devices. Models will, however,
still be desired. One reason for this is that a direct simulation will give very large amounts of data,
which we in many cases do not need. A simpler simulation with a model will cost less and still be
sufficient. Another reason is that humans steadily will extend their wishes so that they always will
be ahead of the development in computer technology.
1.6 Numerical tools 21
Figure 1.7: Example of user interface for a numerical computational tool, excerpt from
screen: Lizard for Kameleon, problem setup, geometrical model (by Bjørn E. Vembe). Here,
one can in a simple way implement geometry and specify boundary conditions, physical
properties and models, numerical methods, etc. This is a simple case made for demonstration:
A gas release at a pole burns and the flame is directed towards a wall. It is subdivided into
25 · 17 · 18 = 7650 volumes.
1.6 Numerical tools

There are quite a few numerical calculation programs that can help us with flow
problems. Here we will look at one example.
If we want to do anything more than “sketch and tell”, we can not avoid solution of equations.
Many equations – thus it is neither practically nor theoretically possible to do this by hand. We can,
however, get assistance from numerical tools, computer programs made to solve the equations that
we will look closer at further back in this book.
There are simple programs that you can obtain (almost) for free, that have scarce documentation,
that require you to work a lot with the presentation of your results, and that you use on your own risk
Figure 1.8: Example of user interface for a numerical computational tool, excerpt from
screen: Lizard for Kameleon, graphical presentation of preliminary results during running
(by Bjørn E. Vembe). Same case as in Fig. 1.7. The upper temperature field is projected such
that we see the highest temperatures in each point in the view direction. This is about as we
see a flame from outside. The other figure shows a cross section through the flame.
1.6 Numerical tools 23
Figure 1.9: Example of presentation of three-dimensional result from Kameleon (by Bård
Grimsmo). The figure shows the visible part of the flame. (In a grey-tone image as here, the
information carried by colors is much reduced. This is an oil-fired, 550 MW steam boiler
in a thermal power plant, where the goal is to reduce the NOx -emissions. The furnace is
7 m · 7 m · 20 m, and in this computation (from 1997) it is subdivided into approximately
250000 sub-volumes.
and perhaps entirely alone. You can even write your own program. There are, however, also very
advanced tools, with detailed documentation, support services, easily accessible user interface and
nice presentations of the results. Some of these programs are general and directed towards nearly
all kinds of flow problems. Then you may have to implement models or expressions for special
phenomena that you work with. Others are more specifically directed towards certain types of flows.
Special versions can be designed for a restricted area of use. Such a program is often called a
simulator.
One side of this is that such program development costs time and money – much time and money,
and you have to pay for this when using the program. Another side is that such computations
put requirements onto the user. You do certain choices and simplifications when you define the
computation; geometric model, physical models, numerical techniques, and so on. Then you have to
evaluate and interpret the results based on the actual conditions and from the choices that you have
made.
At Sintef Energy Research and NTNU a family of such computational tools are developed, with the
name Kameleon. This program is particularly directed towards flow, gas spreading and combustion
(including fire).
Figure 1.7 shows an excerpt from a computer screen where Kameleon with the user interface (pre-
processor) Lizard is in use. With touch keys and mouse clicks one can construct a geometric model
with grid, solid surfaces of different kinds inside the computational domain and various boundary
surfaces, select different physical models and numerical schemes. While the program is running, one
can control the graphical presentation of intermediate results as shown in Fig. 1.8. Examples of two-
dimensional graphical presentations are shown at pages 32, 51, 82 and 158. The three-dimensional
results can also be presented as shown in Fig. 1.9.
1.7 Basic equations

There are several ways and methods for treating turbulent flow and for
mathematical notation. The same equations holds for turbulent flow as for
laminar. I will presume that the reader knows beforehand the equations and
the underlying assumptions. Some of this is included in Appendix A. The
equations are written in Cartesian tensor notation, see Sec. A.4.
For single-phase flow, the equations for continuity and momentum are
∂ρ ∂
+ (ρu j ) = 0, (1.2)
∂t ∂x j
∂ ∂ ∂p ∂τi j
(ρu i ) + (ρu i u j ) = − + + ρ fi . (1.3)
∂t ∂x j ∂ xi ∂x j
The stress tensor for a Newtonian fluid is expressed as

∂u i ∂u j ∂u k
τi j = µ + + (µ B − 32 µ) δi j , (1.4)
∂x j ∂ xi ∂ xk
1.7 Basic equations 25
where µ is the dynamic molecular viscosity and µ B is the bulk viscosity. The Stokes hypothesis:
µ B = 0, see page 201.
The English term momentum comes from Latin “movimentum”: movement amount, of “movere”:
movement. Notice: Moment is something different, both in English and other languages. However,
linguistically, this term has the same origin as momentum.
We often neglect distant forces (for instance gravity) so that the last term of Eq. (1.3) cancels. In
many numerical program systems the continuity equation is replaced by an equation for the pres-
sure correction, which appears from a combination of the continuity equation and the momentum
equation, see Patankar (1980:123-125).
For a mixture of different species, we need an equation for concentration or mass fraction of species,

∂ ∂ ∂ ∂Yk
(ρYk ) + (ρYk u j ) = ρD + Rk . (1.5)
∂t ∂x j ∂x j ∂x j
Here, Fick’s law for mass diffusion is introduced and the diffusion coefficient D is set equal for all
species, see Eq. (A.18).
With N different
P species, we need N − 1 equations. The last mass fraction can be calculated from
the relation k Yk = 1.
The energy equation can be expressed in different ways and with numerous sets of simplifications,
see Sec. A.3. We can use the internal energy or the enthalpy, we can include or exclude kinetic
energy, or we can rewrite the equation to a temperature equation. In many cases, however not
always, an equation for static enthalpy is the most convenient:

∂ ∂ ∂ ∂h
(ρh) + (ρhu j ) = ρα + Sh , (1.6)
where α = λ/(ρC p ) is the thermal diffusivity. This form, with an enthalpy gradient in the diffusion
term, requires an ideal gas or constant pressure so that h = h(T ). Fourier’s law for heat conduction
has a temperature gradient, not an enthalpy gradient, and hence, we have to do some rewriting (cf.
Sec A.3). The content of the source term depends on which simplifications we have adopted. The
temperature can be found from the relation h(T ).
We can also use a temperature
P equation. With combustion, it will at least contain a term for reaction
energy: ST = − k h ◦f,k Rk + . . ., where Rk is positive for formation of species k. h ◦f,k is the
enthalpy of formation for species k. When we are using an equation for enthalpy, the enthalpy of
formation will be included in the variable and we can avoid this reaction term. The temperature
equation is applicable mainly for flows without chemical reactions.
Finally, we need an equation of state for the pressure; p = p(ρ, T ) or ρ = ρ( p, T ), and an
expression for the reaction term Rk , see Sec. 1.8. We can then, in principle, solve the system of
equations for chemical reactions in single-phase flow.
Compressibility and variable density
It will be useful to distinguish between flow at constant density and incompressible flow. Or be-
tween variable density and compressible flow. The term compressible flow is used when the volume
(density) changes with the pressure. Compressibility is a physical quantity and is defined propor-
tional (alternatively, reciprocal) to (∂ p/∂ρ). Thus, in an incompressible flow, there are no changes
in density due to pressure changes. This is the case at a low Mach number (M < 0.2).
With combustion we may indeed have large density differences, with a factor of 2–4 or larger, with-
out any noticeable pressure differences. Then, δρ is large, whereas δp is small. The density varies,
but the flow is incompressible.
For a homogeneous mixture in isothermal flow at a low Mach number, the distinction between
constant density and incompressibility have nearly nothing to say. In the literature, we often find that
the term incompressible is used to mean constant density. Much of the theory and models are in the
first place developed for flow with constant density. And the models are tested against experimental
data of isothermal flows at low Mach numbers.
In this book, we will mainly focus on flows with a constant density. This simplifies the equations
to some extent. Moreover, as mentioned, the theory is often developed with this presumption. In
practice, we will in many cases have to consider variable density in calculations, especially in com-
bustion problems. Those who are going to work with combustion, will have to treat flows with a
density that varies and fluctuates, see Appendix B. But this is not a big and demanding leap from
what is shown here.
The continuity equation and the momentum equation for flows with constant density and viscosity
are shown in the Appendix, Eqs. (A.3) and (A.28)–(A.29).
1.8 Chemical reactions

We will not go closely into chemistry here. This book requires a certain
basis in thermodynamics. (It may pay off to repeat some stuff from there
– see examples of basic books in Sec. 1.1.) In addition, there are many
comprehensive books on combustion theory, references will follow eventually.
Turbulence and combustion
The most important message here is this: Chemistry is chemistry, regardless of whether the flow is
turbulent or laminar.
This means that the chemical reactions follow the same laws in turbulent combustion as in laminar.
The distinction lies in the the flow pattern that is different and accordingly, all the time and length
scales associated with the flow. The mixing conditions and residence times are very different. Due to
this, we will in this book not dwell much upon the chemistry of combustion but more on mechanisms
for mixing and transport – thus turbulence.
1.8 Chemical reactions 27
We will return to modeling of turbulent combustion in Chapters 10 and 11.
Reaction equations
Usually, combustion is oxidation of a fuel with air as the oxidizer. Most simple, we can write this as
an overall reaction,
fuel + λ · amount of air → product + (λ − 1) · amount of air, (1.7)
for example
CH4 + λ · (2O2 + 2 · 3.76N2)

→ CO2 + 2H2 O + 2 · 3.76N2 + (λ − 1) · (2O2 + 2 · 3.76N2) , (1.8)
| {z } | {z }
product excess air
with excess air ratio λ ≥ 1. The numerical value 3.76 (=79/21) appears when we regard air as 21 %
O2 and 79 % N2 . The reaction can also be written with the equivalence ratio (fuel excess number)
8 = λ−1 ,
8CH4 + 2O2 + 2 · 3.76N2

→ 8CO2 + 28H2 O + 28 · 3,76N2 + (1 − 8) · (2O2 + 2 · 3.76N2) . (1.9)
| {z } | {z }
product excess air
Since we regard air as the oxidizer, we include some of the nitrogen as ”product”. If we regard O2
as the oxidizer, N2 is accounted as “inert”.
Many simultaneous reactions
Indeed, it is not so that the fuel reacts directly with O2 . The overall reaction is a sum of a large
number of intermediate steps or elementary reactions. They are linked together like a chain or a
web, since one species can take part in several reactions. We can regard the equations above as a
kind of book-keeping. The equations are written as one-way reactions – however, there is always a
balance between forward and backward reactions. In combustion reactions (i.e. the overall reaction),
the equilibrium is often shifted so far over to the product side that we write as shown above. We
have also neglected species with a low concentration. At the right-hand side, there will always be
some CO, H2 , un-burnt fuel, free radicals etc. If the reactions are incomplete and/or we have deficit
on air (λ ≤ 1, 8 ≥ 1), we have to include such species in the overall reaction.
For a system of NR chemical reactions with NS species Ak , the reaction equations can be written on
a general form as
XNS NS
X
′ ′′
νkl Ak → νkl Ak , l = 1, . . . , NR , (1.10)
k=1 k=1
′ and ν ′′ are stoichiometric coefficients for the species A in the reaction l.

where νkl kl k
Example: Equation (1.10) can best be explained by an example. We consider a stoichiometric

balance between hydrogen, oxygen and water:
2H2 + O2 ⇄ 2H2 O. (1.11)
The balance can be written as two reactions, forward (→) and backward (←), NR = 2; with three
species in each reaction, NS = 3. The symbols in Eq. (1.10) get the values:
k=1 k=2 k=3
Reaction Ak ′
νkl ′′
νkl Ak ′
νkl ′′
νkl Ak ′
νkl ′′
νkl
l=1 2H2 + O2 → 2H2 O H2 2 0 O2 1 0 H2 O 0 2
l=2 2H2 O → 2H2 + O2 H2 0 2 O2 0 1 H2 O 2 0
′ and ν ′′ need not be integers.
For more complex overall reactions, the νkl kl
In order to calculate details of a reaction, we have to consider each elementary reaction. These are
the reactions that occurs at the molecular level.
The reaction balance, Eq. (1.11), is not an elementary reaction but an H/O-system that can be split
into a large number of reactions. Warnatz (1983) use 37 different elementary reactions for the
modeling of an H/O-system. A CO/H/O-system contains 30 more reactions, altogether 67. Warnatz
puts up a system of 231 elementary reactions for combustion of hydrocarbons up to C3 (propane).
For the formation of NO, he puts up 82 new elementary reactions with N and C/H/O. Details can be
found in, for instance, Warnatz (1983,1990) and Warnatz, Maas and Dibble (1999:69,244).
Chemical production term
The production rate (reaction rate) for species k, Rk [(kg)k /(m3 s)], in a system as shown in Eq. (1.10)
is expressed as  
XNR  NS
Y ′ 
ν
Rk = Mk ′
(ν ′′ − νkl ) · kl ci il , (1.12)
 kl 
l=1 i=1
where Mk is molar mass [(kg/mol)k ] for species k, ci is the concentration [(mol)i /m3 ] of species i .
(SI or ISO-standard is mol, not kmol, for amount of substance, and kg, not g, for mass; ISO 31-
0:1996.) See in Sec. A.1.2 on the
Qrelation between mass, amount of substance, molar concentration
and mass fraction. The symbol denotes multiplication. The reaction coefficient kl for reaction l
can be given by the Arrhenius expression,
E al
kl = kl (T ) = Al T βl exp(− ). (1.13)
Ru T
The dimension of kl differs from reaction to reaction. Al , βl and E al are constants for each reaction
l. E al is called activation energy. Ru is the universal gas constant. See more on this in a book on
combustion theory, for instance Kuo (1986:131) or Warnatz, Maas and Dibble (1999:73).
1.8 Chemical reactions 29
In order to make the use of Eq. (1.12) clearer, we can look more at the simple system of Eq. (1.10).
We get
n o
2 2
k = 1 (H2 ) RH2 = MH2 −2k1cH 2
c O 2 + 2k 2 c H 2 O , (1.14)
n o
2 2
k = 2 (O2 ) RO2 = MO2 −k1 cH c + k 2 cH
2 O2 2O
, (1.15)
n o
2 2
k = 3 (H2 O) RH2 O = MH2 O 2k1 cH c − 2k2 cH
2 O2 2O
. (1.16)
Here, we can also see that since MH2 O = MH2 + 21 MO2 , then RH2 O = −(RH2 + RO2 ).
Each of the many elementary reactions (see above) has its own set of constants ( Al , βl , E al ) in the
reaction coefficient kl . Some reactions are fast, others are slow. The difference in chemical time
scale can be many orders of magnitude. In a numerical solution, the smallest time scale has to be
resolved. This means that a complete solution of a combustion reaction demands huge resources,
even for simple laminar flames. This is still an area of research, and such computations were not
made until the 1980s. One have to keep in mind that the systems of elementary reactions and reaction
coefficients are models or images of the reality.
An important message here, is that laminar flames are far more complex than laminar flows
without reactions. In combustion one have to reduce, simplify and model, even when the flow
is not turbulent.
Simple overall reaction
We will return to looking at the combustion reaction as a simple, one-step reaction: It can be written
stoichiometrically on mass basis as
1kg fuel + r kg oxidizer → (1 + r )kg product . (1.17)
Here, r is the stoichiometric amount of oxidizer for the relevant fuel: r kg oxidizer for the burning
of 1 kg fuel. As oxidizer we can consider air or only oxygen.
In the equation for the mass fraction Yk we have the source term Rk . This is the mass of species k
produced per volume and time [(kg)k /(m3 s)]. The left-hand side of Eq. (1.17) can also be expressed
as
Rox = r · Rfu (< 0) . (1.18)
These source terms are negative as the species are consumed. In a reaction, the mass is unchanged,
so that for the product we obtain

1
Rprod = − (Rfu + Rox ) = −(1 + r )Rfu = − + 1 Rox . (1.19)
r
This is primarily a mass balance. What goes in must come out as product. Therefore the first equality
of Eq. (1.19) is valid. What is consumed, and what is produced, have to have opposite signs; thus
minus. Equation (1.18) define the quantity r for a given fuel.
For species that do not react, the production is, of course, Rk = 0. Usually, we use air (not pure
oxygen) in combustion processes. The nitrogen can be included in the oxidizer. When the air
“reacts” (that is, the oxygen in the air reacts), the nitrogen will follow over to the product side (see
Eq. 1.8). The same holds when we, for instance, include argon, CO2 and other gases in the air.
Why so much on turbulence in a book on combustion?
Find a simple gas burner; for instance a burner for ski wax, a welding burner or simi-
lar;
(1) turn up the gas without igniting,
(2) turn up and ignite as usual,
(3) close up the air inlets on the burner neck, turn up the gas and ignite.
Now you have observed a turbulent flow: 1) without reaction, 2) with (partially) premixed
combustion, and 3) with non-premixed combustion. You did probably notice that the sound
was different and that the flames were different. In the last case, you most likely got a flame
that was bigger and that was sooting more. You did not add any other chemical substances, just
changed the flow. This is a simple example where the flow has huge impact on the combustion
and on the formation of pollutants – and on the noise.
Chapter 2
Turbulence modeling
2.1 Modeling
What is a model?
The word model means image or figure. It is used in many ways. A model is something that rep-
resents something; it depicts what we regard as important, still without pretending to be a complete
reproduction. (Aris, 1978:1ff, gives a more comprehensive explaination of what a model is.)
Within this subject we meet wordings like scale-model, mathematical model and physical model.
The expression scale-model nearly explains itself. We find such items both in the laboratory and
in the child-room. A mathematical model is a set of equations that gives a numerical or relational
picture of something. What a physical model is, depends on who is using the word. It can be
an item in the laboratory, for instance a car model for use in a wind tunnel. It can be a thought
model, a simplified image of a physical phenomenon. One example is the sun-and-planet model for
atoms. Another example is Prandtl’s model with «fluid balls», Sec. 2.6 and 3.3. These are strongly
simplified images with simple physical phenomena – thought objects if you like – made to explain
more complex problems.
A physical model can also be the approximation that lies behind a mathematical model of a physical
phenomenon. Heat conduction (thermal diffusion) is a physical phenomenon. A model for this is,
hence, a physical model. The model known as «Fourier’s law» states that the heat flux is proportional
to the temperature gradient, and with a material parameter or coefficient as the proportionality factor.
This can be expressed with mathematical symbols as q = −λ(d T /d x). When q, λ, T and x are
defined, this is a mathematical model.
The mathematical expressions can be put into computer programs in order to simulate (“to make
like”) the specific phenomena. Many people use the word model, then meaning both the mathemat-
ical expressions and the way they are implemented into the program.
Here, I will use the term model mostly in the meaning mathematical model and the simplified image
of a physical phenomenon that lies behind.
32 2 Turbulence modeling
(a) Car model, toy car (11 cm) (b) Molecule model, model of molecular structure (ball-
and-stick model of trans-2-butene)
Temperatur (K)
dT Plane: Y= 3.689 Vel_max= 144.8
q = −λ
Z (m)
dx Max
(c) Model for heat conduction, Fourier’s law 2500.0
2300.0
C + O2 → CO2 2250.0
(e) Reaction model 2200.0
2150.0
20 2100.0
2050.0
2000.0
1900.0
1800.0
1700.0
1600.0
1500.0
15
1400.0
1300.0
1200.0
1100.0
1000.0
875.0
750.0
10
625.0
500.0
386.5
273.0
313.0
Min
(f) Numerical model of a platform

5
Figure 2.1: Examples of models. Each model visualizes something;

2 4
it depicts
6
what is impor-
8
tant, but without pretending to be a complete representation. What
Time= 450.0 aspects that are important,
X (m)
depends on the context.
(d) Numerical model of combustion in a furnace (by Bård
Grimsmo, with Kameleon)
2.1 Modeling 33
Always an approximation
It is important to grasp that a model always involves approximations and underlying assumptions
for these approximations. Therefore, a model is not “right” or “wrong”. A model is just a good or
less good approximation. Whether the model is “sufficiently good” depends on the situation, who is
using it, and for what. A model is developed under certain presumptions. This puts restrictions on
the validity of the model.
Models, ideally and in practice1
Ideally, we can say: A good model is based on correct theory, has clear assumptions and gives results
with all variables close to the actual values. A good model is easy to implement into numerical
program systems and it gives stable numerical solutions.
In practice, we experience often otherwise:
• A model can be good without correct theory behind.

• A model is not necessarily good, even though the underlying theory is correct.
• The assumptions of a model are rarely clear – and do not have to be so either.
• Even if the model gives good approximations for the most important variables, these can be
calculated from other quantities that each by themselves and alone are poor approximations.
• A good model does not have to be complicated.
• A complicated model does not have to be good, it may give poorer approximations than a
simple model.
• A complicated model does not have to take more time than a simpler one.
• No model is also a model – and it may be better.
• A model can give useful results outside the domain it, according to the development, should
be valid for.
• A model can give useless results also within the domain it is developed for.
Whether a model is “good” or not, depends on what you are going to use it for and how accurate your
results need to be. Example: In an engineering design work the model gives results that surely can
be expected to be within the order of magnitude of the actual values. This is “good” if the alternative
methods give deficiencies of more than an order of magnitude. However, if the requirements are
stricter, and the alternatives better, the model is not sufficiently good.
A model can be more or less general. Models are developed to be valid within certain presumptions.
A simple model is as a rule usable within a restricted area but, on the other hand, on this area it is
more accurate than a more general and complex model.
Models for turbulence and combustion are developed over time, with trial and error in small steps.
For the most common models, we find a great number of proposals for changes and additions.
1 Someone may think that this is adapted from Williams (1992). It is not – the content is different. I got however, some
thoughts from there. Actually, I am not sure that Williams would agree on what is written here.
Those who use such models aquire experience over time and can more easily evaluate the results
that appear. New variants of a model can be better. This we do not know, however, before some
experience is gained. It is better to know the weaknesses of a well tested model than hoping for
better results with an unknown model.
2.2 Reynolds-decomposition and averaging

Here and in the next section we will look at a practical way to handle the
basic equations so that they can be solved. We will divide pressure, velocity
(momentum), energy and mass fractions of substances into a mean value
(average) and a fluctuation, and then solve the equations for the mean values.
For practical purposes, we can not solve the basic equations for turbulent flows. As we noted in
Sec. 1.5, we can not resolve the motions down to the smallest eddy of the flow, except for very
simple flows. We have to do simplifications. One approach is to separate mean values (average
values) and then develop equations for these mean values. Then, when leaving out smaller details,
we can use a much coarser subdivision of the volume to be investigated.
Reynolds (1895) divides the field into a mean value (average value) and a fluctuation (departure from
mean value):
u i = u i + u ′i , p = p + p′ , ϕ = ϕ + ϕ ′ , . . . , etc. (2.1)
ϕ
The sketch shows a quantity that fluctuate about a

t mean value.
Here, we have to throw in a remark to those who are going to work with turbulent combustion and
other flows of variable density (cf. page 26): You have to learn about Favre-decomposition. This is
mass-weighted averaging, see Appendix B.
Mean values
There are several ways of defining mean values (see e.g. Hinze, 1975:5):
Time mean: Average taken over a time interval 1t,
Z t + 21 1t
1
ϕ= ϕ(t)dt. (2.2)
1t t − 21 1t
2.2 Reynolds-decomposition and averaging 35
Here, the averaging time interval 1t has to be much larger than the time scale for the fluctuations,
but much less than the time scale of the variations in mean value. Notice that the mean value also
can vary with time.
Ensemble mean: Average of a large number N of events or experiments
N
1 X
ϕ= ϕn . (2.3)
N
n=1
The notation with overbar, ϕ, is common, but also angle parentheses, hϕi, are used to denote a mean
value, particularly for ensemble means.
A mean value can also be defined as the mathematical expectation from a probability density function
f (ϕ) (see page 111), Z ∞
ϕ= ϕ(c) f (c)dc. (2.4)
−∞
From the viewpoint of a statistician, time and ensemble averages are approximations for a mathe-
matical expectation value (see e.g. Pope, 1985:126).
One can easily get caught up in questions about what kind of mean values we are operating with.
When you present data and discuss measurements, it is important to clearify what you are doing. If
you are doing numerical simulations, it can be important in the development and adaption of models
and in the interpretation of the results. However, the program does not ask what kind of variable you
have in mind. It computes a quantity and says nothing about what this is.
The various definitions give the same expressions in the equations, and we have no information about
what kind of average that is used. One have to keep this in mind before using time for discussion of
(or quarrel about) the interpretation of values from numerical simulations.
What you as a user have to think about when the mean value varies with time, is that the time scale
of the turbulence should be much smaller than the time scale of the variation of the mean value. If
these parts slides into each other, you will have difficulties with a strict interpretation of your result.
– What is turbulence, and what is a swaying mean value? (See also Sec. 9.4.)
Fluctuations
A fluctuation is the deviation from the mean value,

ϕ ′ = ϕ − ϕ. (2.5)
The mean value of a fluctuation is zero, regardless of how the mean is defined:
u ′i ≡ 0 , p′ ≡ 0 , ϕ ′ ≡ 0 , . . . , etc. (2.6)
Rules for algebra of ordinary averaging (Reynolds-averaging)
Rules for averaging can be developed from the definition of the mean. This is made in Appendix B,
page 222.
2.3 Turbulence equations, Reynolds equations

We introduce the Reyolds decomposition into the basic equations from Sec. 1.7. We then average
the entire equation, term by term. Here, we keep the density ρ constant (cf. page 26).
Continuity equation:
∂u j ∂u ′j
=0 and = 0. (2.7)
∂x j ∂x j
Momentum:
∂ ∂ ∂p ∂
(ρu i ) + (ρu i u j ) = − + τ i j − ρu ′i u ′j + ρ f i . (2.8)
Mass fraction:
!
∂ ∂ ∂ ∂Y k
(ρY k ) + (ρY k u j ) = ρD − ρYk′ u ′j + Rk . (2.9)
Energy: !
∂ ∂ ∂ ∂h
(ρh) + (ρh u j ) = ρα − ρh ′ u ′j + Sh . (2.10)
This are the Reynolds-averaged equations or Reynolds equations. They are equations for the av-
eraged quantities, and these we can solve numerically. Now, however, we have got new unknown
variables: −u ′i u ′j , −u ′j Yk′ and −u ′j h ′ .
These quantities are parts of the mean convective fluxes, which are transport with the turbulent mo-
tions. They are usually called Reynolds stresses or turbulence stresses2 (−u ′i u ′j ) and -fluxes (−u ′j Yk′ ,
−u ′j h ′ ). In the words of statistics they are denoted correlations (see more about statistical func-
tions in Sec. 7.1). In the mean-value equations, they can be regarded as diffusive terms (turbulence
diffusion), and they are put together with the molecular diffusion terms as shown in the equations.
Generally, in three dimensions, we have 6 Reynolds stresses in the momentum equation and 3
Reynolds fluxes in each scalar equation. These quantities are new unknowns and the system of
equations is no longer “closed” – it has more unknowns than equations. We have to face the “the
closure problem”. Moreover, we have to cope with source terms. In the energy equation, the source
2 The names Reynolds and turbulence stress and -flux are used about these quantities both with and without ρ included.
When it matters, we can distinguish between dynamic and kinematic quantities. The names are also used for the quantities
both with and without the minus in front.
2.4 Turbulent transport and exchange 37
Example: Development of an equation for mean mass fraction

The starting point is Eq. 1.5.
Here, we introduce the Reynolds decomposition Yk = Y k + Yk′ , u j = u j + u ′j :
!
∂ ∂ ∂ ∂(Y k + Yk′ )
ρ(Y k + Yk′ ) + ρ(Y k u j + Y k u ′j + Yk′ u j + Yk′ u ′j ) = ρD + Rk .
Now we average each term according to the rules in Sec. B.1. The terms with only one fluctua-
tion will have mean value zero. The remaining terms are retained:
!
∂ ∂ ′ ′
∂ ∂Y k
(ρY k ) + ρY k u j + ρYk u j = ρD + Rk .
The third term we often put on the right-hand side, as seen in Eq. 2.9.
In this development we have ignored fluctuations in the density ρ and the diffusivity D. We
could also say that ρ and D fluctuate, and that we neglect correlations where they take part.
That is, we neglect terms like e.g. ρ ′ Yk′ and ρ ′ u ′j . When such correlations are important, we
can use Favre-equations, see Appendix B.
term will contain new correlations, depending on the form of the equation we have chosen. In many
cases, the source term of the enthalpy equation can be neglected.
Chemical production term
The production term R k in the Y k -equation is a problem by itself. This term, Eqs.(1.12)–(1.13),
include an exponential function of the temperature. It can not be averaged in the same way as the
other terms above. This means that models for turbulent combustion have to be thought along paths
different from those that lead to models for turbulence transport. We will come back to this in
Chapter 10.
2.4 Turbulent transport and exchange

In order to visualize transport and exchange, we will look at two railroad trains. We imagine that
they are running on two tracks side by side. The trains have open goods cars loaded with sand. Train
1 has a somewhat higher speed than Train 2. We further imagine that a man on each train shuffles
sand over to the other train.
Train 2
✻
Train 1 ❄
We let Man 1 shuffle sand from Train 1 to Train 2. The sand from Train 1 has a larger velocity
and hence, a larger momentum (literally “amount of movement”) per unit of mass than Train 2.
Therefore, we add momentum to Train 2, which will increase its velocity. Opposedly, sand from
Train 2 has less momentum and will slow down Train 1.
This is an example of exchange of momentum. We could extend the picture and imagine that the
sand of Train 1 has higher temperature, and that it is of a different kind (substance) than the sand of
Train 2. Then we will have an exchange of energy and substance.
The exchange between the trains is a picture of diffusion; molecular diffusion is transport with
molecules that move among each other. Turbulence transport («turbulence diffusion») is transport
with «elements» («lumps», «balls», eddies) that move among each other.
In the picture above, we also have transport with the trains along the rails, with the velocity of trains.
This corresponds to convective transport with the flow.
2.5 Turbulence viscosity

The turbulence enhances the transport in the flow. In the momentum equation, we can look at
transport with the turbulence as an additional stress, or as an increase of the viscosity, such that we
calculate with an effective stress and an effective viscosity:
τeff = τ + τturb ; µeff = µ + µturb (2.11)
We talk about turbulence stresses and turbulence viscosity. Other names are eddy viscosity (English
eddy is from Norwegian “ide”) and Boussinesq viscosity. The latter term is named after the one who
in 1877 (Boussinesq, 1877:46f) suggested such an additional viscosity. (Notice that this was before
Reynolds introduced the decomposition and defined the turbulence stresses.) The simplest approach
is to give this turbulence viscosity a constant value. This is often done, and quite a few users are
satisfied with and have benefit of the results from such calculations. One can choose a value of, say,
100–500 times the molecular viscosity. Such a model can be useful but is in no way general.
The molecular viscosity of a gas can be regarded as a product and a length scale (mean free path, ℓ)
and a velocity scale (mean molecular velocity, v̄): µ ∼ ρℓv̄. (See in a book on the kinetic theory of
gases.)
In the same manner, the turbulence viscosity can be interpreted as a product of a characteristic length
scale and a characteristic velocity scale for the turbulence: µt = ρℓ′ u ′ .
2.5 Turbulence viscosity 39
It is important to keep in mind that the molecular viscosity is a thermal property of the fluid, whereas
the turbulence viscosity is a function of the flow.
Reynolds stresses
For a simple, two-dimensional case with only one gradient (du 1 /d x 2 ), the viscous (molecular) stress
is
du 1
τ = τ12 = µ . (2.12)
d x2
This follows from Eq. (1.4). By introducing a turbulence viscosity as in Eq. (2.11), the turbulence
stress can be modeled according to the same pattern:
du 1
τ12,t = −ρu ′1 u ′2 = µt . (2.13)
d x2
These expressions are valid for simple flows such as boundary layers. (Boundary layers will be
described further in Sec. 4.3). However, flows are often more intricate, with gradients in several
directions. The general model becomes

∂u i ∂u j 2 ∂u l
−ρu ′i u ′j = µt + − ρk + µt δi j . (2.14)
∂x j ∂ xi 3 ∂ xl
Here we see an analogy to the model for viscous stress, Eq. (1.4). The final term, with the turbulence
energy k, have to be included in order to make the expression general. Notice that for flows with
constant density, ∂u l /∂ xl = 0 in the final term.
Turbulence diffusivity
In the same way as in the momentum equation, scalar transport with the turbulence can be modeled
by an addition to the diffusion coefficient:
Ŵeff = Ŵ + Ŵt , (2.15)
where Ŵ is the molecular diffusivity and Ŵt is the turbulence diffusivity. Often the turbulence dif-
fusivity is modeled by a Prandtl/Schmidt number for turbulence, see more in Sec. 5.2.
A note on simple models
Here we have to take some knowledge and understanding from later chapters
in advance.
In the next section we will look at Prandtl’s mixing length model. The turbulence viscosity appears
as a function of the mean velocity gradient and an external length of the flow, for instance a pipe
diameter or the distance to a wall. Such models can be suitable in simple flows where measured data
are available for adjusting the model. There are quite a few simple models that are made for and
adapted to certain flow configurations. One of these is the Baldwin-Lomax model (see e.g. Wilcox,
1993:52), which has been widely used for turbines and compressors. In this model, the turbulence
viscosity is a function of mean velocity gradients, a characteristic mean velocity and two lengths of
the flow.
In the next and later chapters, we will look at models that solve one or more equations for turbulence
quantities. It is then obvious to attempt to solve an equation for the turbulence viscosity itself. Such
models are proposed, and one of the newer and most popular is by Spalart and Allmaras (1994).
This model has proved to be good for flows near walls, such as wings, airplane bodies, turbines.
Giddy with joy for a good result, and without further reflection, one can readily transfer the model
to other cases. This can turn out to be really bad: The equation in the Spalart-Allmaras model has
destruction terms that go to zero as the distance to walls increases. The production is finite and,
hence, the turbulence increases without limitations. In flows in open air, open volumes or wide
channels, the model is inadequate. This is also the case for many other simple models (zero or
one-equation models) where the distance to a wall is an important parameter.
2.6 Prandtl’s mixing length model

We consider a turbulent flow (boundary layer) with a simple velocity gradient du/d y > 0. At a
location yo in the profile, the velocity along the flow is on average u o . Prandtl (1925) adopts a picture
from the kinetic gas theory and imagine “balls” of fluid that move crosswise in the flow through the
plane yo from above and from below. Like molecules in kinetic gas theory these turbulence balls
have a velocity and a length of motion. In kinetic gas theory, one looks at an averaged molecular
velocity and the mean free path and expresses the viscosity from this.
y u 1✲
✻
✻
ℓ v′
yo ❄ ❄u o✲
✻ ✻
ℓ v′
❄ u 2✲
A turbulence ball from above carries the velocity u 1 = u o + 1u from where it comes from. In
the location yo , this ball has a deviation from the surrounding local value equal to u 1 − u o =
u o + 1u − u o = 1u. This deviation is the fluctuation u ′ (> 0). At the original position, the ball had
a velocity u 1 , and the distance ℓ from yo is given by u ′ = 1u = ℓ(du/d y). The ball moves with a
2.6 Prandtl’s mixing length model 41
larger velocity and this increases the momentum in location yo . The ball is carried with velocity v ′
downwards crosswise of the flow (v ′ < 0), and the flux of momentum becomes ∼ v ′ (ρu ′ ).
The same lines of thought is used for balls from below. They carry a lower velocity u 2 = u o − 1u
(u ′ < 0) and have a velocity v ′ upward (v ′ > 0).
v′
✻
v′
❄ uo + u′ u − u′
✲ ✲o ✲ ✲
✻
v′
v′
❄
The crosswise velocity v ′ , upwards or downwards, must be of the same magnitude as u ′ . Prandtl
illustrates this by looking at two balls in yo , one that came from above with velocity u o + u ′ and one
from below with velocity u o − u ′ . The difference between them becomes 2u ′ . If the faster is behind,
it approaches the slower. The fluid in between is pushed as a ball to both sides with velocity ±v ′
(difference 2v ′ ). This velocity have to be about as large as u ′ , thus v ′ ≈ ℓ(du/d y).
Flux of momentum is called stress. The turbulence stress is the flux of momentum (ρu ′ ) with the
v-fluctuation and can be expressed as
2
du
τturb ∼ v ′ (ρu ′ ) ∼ ρℓ2 . (2.16)
dy
The constants of proportionality can be included in the (still unknown) length scale ℓ, and we have
du du
τturb = −ρu ′ v ′ = ρ ℓ2 . (2.17)
dy dy
| {z }
νt
The absolute-value symbol is introduced to ensure the correct sign.

The mixing length have to be determined from some function of external quantities, for instance a
pipe diameter or a distance to a wall. Rodi (1980) gives an overview over some such functions.
A mixing length model for energy or mass flux can be worked out in the same manner. In place of a
deviation in velocity, we then look at a deviation in temperature or mass fraction.
Schlichting (1979:579f) and Tennekes and Lumley (1972:42) give a more detailed and mathemat-
ically funded development of this model. We should, however, not put too much emphasis on the
mathematical aspects of this model. Objections can readily be raised against the details (see e.g.
Tennekes and Lumley, 1972:45). What we should maintain is that Prandtl achieved the model by
dimensional arguments and a very simple image of the turbulence of the flow. The model has proved
usable in a wide range of cases. The strict mathematical development has appeared later. Prandtl’s
development hardly use any mathematical expression before the final result. It is interesting to no-
tice that Tennekes and Lumley (1972:47–49) themselves find the strict mathematical development
useless, but rather explain the model from dimensional arguments – just as Prandtl did.
2.7 More on turbulent transport and length scales

Turbulence is quite complicated, and it is easy to loose oneself in lengthy (and intricate) mathemati-
cal developments. We shall use mathematics as a tool. But we must also maintain the simple images
of thought like that used by Prandtl in the mixing length model.
Turbulence is small and large eddies in the flow. The large eddies move fluid crosswise of the flow.
We can, like Prandtl, think of a “package” or “ball” of fluid that is carried across in the flow. The
packages are moved a length that corresponds to the size of the large eddies, ℓ′ .
ℓ′
This is transport with the turbulence, ”turbulent diffusion”. If temperature or species concentration
varies in the field, these eddies will dilute these variations. Packages of one substance are carried to
a region with another substance.
Smaller eddies do the same, although over a shorter distance. For the transport, they have a smaller
effect. They are, however, important for the local mixing and, hence, for the chemical reactions.
These lines of thought we will return to later on, particularly in Chapter 8.
Imagination gives the best pictures.
In large flames, we can often see turbulence and other eddy motions. It is difficult to find good
photographs of this, with appropriate contrast and resolution. Each time you see large flames
on film, TV or in reality, you can look for large and smaller eddies. Then you can imagine that
even smaller eddies behave similarly to and in-between those that you can see.
2.8 Characteristic scales: – Into the fog? 43
2.8 Characteristic scales: – Into the fog?

In relation to turbulence, we often talk about characteristic scales and order
of magnitude. For many students (and researchers), this gives a feeling of
something unclear and floating. It is about as easy to take in as it is to catch the
fog. The purpose of this section is to show way in the fog, and may be to lift the
fog as well.
For many readers it may be more useful to return to and read this section more
closely later on. Actually, one do not need to read this section at all.
Experience tells that many feel a little uncertain on what is meant by characteristic scale and order
of magnitude. In fluid mechanics, and especially in turbulence, we often talk about characteristic or
typical scales. It may be length, velocity or time. In rarer cases, it can be other quantities, perhaps
temperature.
One explanation can be this: A characteristic scale is a measure that gives a feeling of how large
something is.
Another explanation can be that this scale is representing or telling something about a phenomenon
that is important for the process we are interested in.
Order of magnitude
It is nice to have tangible and accurate quantities to work with. This feels the safest. But reality is
not always easy and plain. We do not avoid some “perhaps”, “approximately”, “about”, “usually,
but can also” and so on.
Order of magnitude is a term that many already know. We are confident with the phenomenon, even
without using the word. An automobile is approximately one ton, the driver 80 kg and his wallet
may be 1 kg. Here we have the orders of magnitude O(103 kg), O(102 kg), O(100 kg).
It becomes a bit more unclear when we are to put a distinction between two orders of magnitude. –
Where is the limit between O(10−1 ) and O(100 )? – Which order of magnitude are 0.3, 0.4 and 0.5?
The answer – as foggy it may be felt – is that the limit is so unclear that these numerical values can
be put into O(10−1 ) just as well as into O(100 ). When we talk about order of magnitude, we often
do not know any exact numerical value.
Attempt on clarification – Different yardsticks
Here, it will be convenient to talk about length. But there may be characteristic scales for all kinds
of quantities. Most common are length, time and velocity. We meet them in our daily life, however,
without the label characteristic scale.
The daily life is full of characteristic scales. If we think on a wheelbarrow, an automobile and an
airplane, we have also considered different scales of velocity, length and time. If something is to be
moved a few meters, we do not use an airplane or a car. And we do not use a wheelbarrow from
Trondheim to Oslo.
The words cleaning cupboard, office, lecture room and hangar tell something about the use of a
room. But, without thinking on it, we also have an idea of a characteristic length scale of each
of these rooms. There are, surely, offices that are smaller than a large cleaning cupboard, while
others are larger than some lecture rooms. However, a typical office is larger than a typical cleaning
cupboard. Or: The characteristic length scale of an office is larger than the characteristic length scale
of a cleaning cupboard and less than that of a lecture room. But what if we compare the cleaning
cupboard and the lecture room with the distance from Trondheim to London – or with the wave
length of light? Then we might find that we can use one common length scale for all rooms and halls
at campus.
Here, length is used as an example. Alternatively, we could talk about time and compared human
reaction time (seconds), lectures (hours) and period of study (years) with historical and geological
time spans, and with the time for laser pulses.
Thus, we look at a group of different items. It might also be a situation where something varies, for
instance the flow pattern in a turbulent flow. The point is that even when there is a variation within
a group, the group can be represented or characterized by a common scale, and similarly when the
state varies.
Time scale for distance?
Above, we have looked at length scales and lengths. But in many cases a length is represented by a
time scale. Distance between cities can be measured in length. However, also in time of travel. A
time scale for the distance Trondheim – London can be 2 hours, as most travelers use an airplane.
The time scale for the distance Trondheim – Steinkjer can also be 2 hours. Here, most travelers use
car or rail. These are examples of characteristic time scales for lengths.
Chapter 3
The k-ε turbulence model

The main purpose of this chapter is to present the so-called k-ε model. The outlines are also valid for
many other one and two-equation models.
A k-ε model, Eqs. (3.20)–(3.25), with a certain set of constants, has got status as a “standard”.
The use of this model has primarily got legitimacy by the fact that very many workers have used
exactly the same model in a wide range of different flows. In this manner, much knowledge and
a lot of experience is gained on weak and strong sides of the model.
If the model is used with other constants, or with significant adaption, the collected experience
has limited validity. Such use should be conducted with great care – and be met by sound
scepticism. All deviations have to be documented and the reasons well specified.
3.1 A note on turbulence models1

This section is an introduction and can be read accordingly. The last part,
however, discusses items that will be more easy to grasp when you have read
more of the book.
The starting point here is the equations for averaged quantities, the Reynolds equations, Eqs. (2.7)–
(2.10). We want to find a suitable model for the Reynolds stresses. This closure problem has been
worked on for a century. Simple models with a constant turbulence viscosity, or with simple alge-
braic expressions, are in daily service. Other methods, where mean values are not used, we will
return to in Sec. 9.4.
Models are often classified by counting equations: Zero, one, two, many. It is then meant equations
for turbulence quantities, for instance Reynolds stresses, turbulence energy or length scale. The
1 This is a rewritten and shortened version of my dissertation, Ertesvåg (1991). More details and references are given
there.
46 3 The k -εε turbulence model
counting does not include the equations for momentum, energy and mass.
Prandtl’s mixing length model (Sec. 2.6) is an example of a zero-equation model or an algebraic
model. There exist several other such models. Common for all these is that they are not of much
help when we need turbulence quantities in a combustion model.
Reynolds (1895) puts up a transport equation for the turbulence energy (see next section). About
1940 some researchers formulated models for the unknown terms of such equations. Kolmogorov
(1942) has a model with an equation for the turbulence energy and for a characteristic frequency.
Prandtl (1945) has a model equation for turbulence energy and Chou (1940) has equations for the
Reynolds stresses, triple correlations and (Chou, 1945) for dissipation of turbulence energy. Rotta
(1951a,b) also presents a model with equations for the Reynolds stresses and for a length scale.
Several others came with proposals in the 1950s and 60s. These researchers did not have access to
electronic computers, and their work was based on theory and simple calculations.
At the end of the 1960s, computers were available. Then the models could be effectively tested
numerically. From about 1970, a series of publications appeared. At that time, two-equation models
were developed and put into application. These are models with an equation for the turbulence
energy and an equation for a length-scale variable. They were primarily k-ε models: Jones and
Launder (1972), Launder and Spalding (1974), but others as well. With small or no changes, these
models are applied in research and industry today. The paper of Launder and Spalding has showed to
be an important milestone on the road. Their conclusions can very well be debated but are important
because they settled a way to do it. Their model, and closely related models, are the most popular as
of today. The terms and constants in the ε equations are usually the same as in 1974.
Launder and Spalding (1974) summarize work with two-equation models at Imperial College in
London: an ε-equation, a (k/ℓ2 )-equation and a (kℓ)-equation. They give reasons for choosing an
ε-equation and not the others. The main argument is that equations for quantities with a dimension
other than ε (∼ k 3/2/ℓ) have to have additional terms (diffusion terms), which would be functions
of the wall distance in a boundary layer. Certain questions can be raised to this arguments, e.g.
Ertesvåg (1991:45). Nevertheless, the article by Launder and Spalding (1974) marks a watershed
since nearly everyone else either chooses their model, or have to consider their conclusion.
Models with equations for the Reynolds stresses
In the 1970s and forward, models are developed that solve the Reynolds-stress equations, as sug-
gested by Chou and Rotta: Hanjalić and Launder (1972,1976), Launder, Reece and Rodi (1975) and
others with an ε equation; Wilcox and Rubesin (1980) (see also Wilcox, 1993) with an ω2 -equation
(where ω ∼ ε/k). For such models Launder (1989a,b) made up a kind of status for the development
so far. We will come back to these in Chapter 6.
Models of higher order
It is possible to go on to develop model equations for quantities of higher orders. Chou (1940,1945),
and others after him, models the equation for triple correlations (u ′i u ′j u ′k ). Testing of such models
is barely found in the literature. Chou (Zhou in newer transcription) and his group is an exception.
3.1 A note on turbulence models 47
They went on and tested models with equations for triple correlations and quadruple correlations
(u ′i u ′j u ′k u l′ ), Zhou (Chou) and Chen (1987).
Time or length scale for turbulence.
Turbulent flows have characteristic time and length scales that are continously distributed over a
spectrum. The smallest, the Kolmogorov scales (Chapter 8, page 122 and 130), are determined by
viscous forces while the largest are restricted by the external physical dimensions of the flow. In the
type of models that we treat here, the different time scales and length scales are represented by one,
single characteristic scale at each point of location. For good reasons, it has become√ customarily
to determine a characteristic velocity scale (length divided by time), for instance k. In addition,
another characteristic scale of turbulence is needed, either for length or for time.
Algebraic models, for instance mixing-length models, determine both the velocity and length scales
from the mean flow. One-equation models (Prandtl, 1945; see Sec. 3.3) determine a turbulence ve-
locity scale from a transport equation, whereas the external dimensions are used to determine the
length scale. Two-equation models and models with Reynolds-stress equations contain (usually) a
transport equation for a time-scale or length-scale variable. In principle, this equation is the same, in-
dependent of whether the model has an equation for turbulence energy or equations for the Reynolds
stresses.
The development of such model equations has been stepwise. It seems that√the first to be published
was by Kolmogorov (1942). He proposes an equation for a frequency, ω ∼ k/ℓ, to be used together
with an equation for turbulence energy. This ω equation has both diffusion term and dissipation term
but no production term.
A literature review, Ertesvåg (1991:42f), shows a true multitude of equations to determine a length
scale or time scale for the turbulence. When it comes to practical usage of the various models, doc-
umentation is much harder to find. It seems clear that Wilcox’ models with an ω2 equation (Wilcox
and Traci, 1976; Wilcox and Rubesin, 1980; Wilcox, 1988,1993), and a group of models with a (kℓ)
equation are or have been used for practical flow problems. It is, nevertheless, primarily models with
an ε equation that dominates. Influential groups chose in their time to go in for a “standard”-version
(Launder and Spalding, 1974) of this equation, with virtually fixed constants. This equation was
then equipped with additional terms to enable computations of different phenomena as flow near
walls, buoyancy, streamline curvature and others.
Low Reynolds numbers
To enable simulation of regions where viscous forces affect the flow, so-called low-Reynolds-number
models are developed. Rotta (1951a) has a viscous dissipation term, while Jones and Launder
(1972) introduce empirical functions of the turbulence Reynolds number to adjust the constants
in the model. Then, one can calculate turbulence at low Reynolds numbers, for instance in a flow
close to a wall. With a minor change, Launder and Sharma (1974), this model is often regarded as
a “standard”. The approach with empirical functions is followed by others, see the review article
by Patel, Rodi and Scheuerer (1985), where also the models are presented. There it appears that
the models of Launder and Sharma (1974) (and Jones and Launder, 1972) and Wilcox and Rubesin
(1980) are the only low-Reynolds-number two-equation models that are used to any extent. Hanjalić
and Launder (1976) introduced the same in a model with Reynolds stress equations.
What is ε ?
For high Reynolds numbers, most of the dissipation occurs in the small eddies. The mechanical
energy is transferred from the mean flow to the larger turbulence eddies, and then further to smaller
and smaller eddies. This can be sketched as a cascade or a waterfall, see Sec. 8.1, page 123.
Basically, ε is defined as a function of the local viscous stresses, ν∂u ′i /∂ x j , see Eqs. (3.4) and
(6.4). From the model equation, Eq. (3.23), we see that ε becomes a function of mean quantities.
The viscosity enters only the viscous diffusion term, and this can be neglected at high Reynolds
numbers. This means that ε is determined by quantities that characterize the mean flow and the large
turbulence eddies. The quantity ε in the model is not the dissipation directly, but expresses rather the
energy transfer from large to smaller eddies. The model-ε can be compared to w′′ in Magnussen’s
model (see Fig. 11.1, page 173). As ε results from large-scale arguments, it becomes a characteristic
scale for the large eddies.
Time scale k / ε
These models have only one time scale or length scale for turbulence. The time scale k/ε, with k
and ε from the “exact” k equation (Eq. 3.4), is characteristic for smaller scales. It expresses the “life
time” of a turbulence eddy, or transfer time from large to smaller scales, see page 132. But since ε
and k are solved from large-scale arguments, the k/ε from the model is more like a transport time
scale.
3.2 Turbulence energy: Definition and development

The kinetic energy per unit of mass for a fluid in motion is 12 u i u i . When we subtract the mean energy
from the mean of this quantity, we obtain the kinetic energy of the turbulent fluctuations 21 u ′i u ′i . The
mean value of this is the mean kinetic turbulence energy or simply the turbulence energy (somewhat
misleading also called turbulent kinetic energy)2:
k = 21 u ′i u ′i = 21 (u ′1 2 + u ′2 2 + u ′32 ). (3.1)
From the basic equations, we can find an equation for the turbulence energy k:
(1) We start out from the equation for momentum, (ρu i ), and assume the density to be constant.
(2) Introduce the decomposition u i = u i + u ′i , and deduce an equation for the mean value u i .
2 On the use of the words turbulent and turbulence, see Sec. D.3.1 or Ertesvåg (1991:7). As an attributive noun (noun
adjunct), turbulence tells that the following (modified) noun has something to do with turbulence, while the adjective turbulent
tells that something is uneasy, fluctuates. Averaged quantities do not fluctuate, hence turbulence energy, turbulence viscosity,
and so on. The term kinetic is redundant – how could turbulence have any other energy than energy of motion?
3.2 Turbulence energy: Definition and development 49
(3) Subtract the latter equation from the former and obtain an equation for the fluctuation u ′i =
u i − u i . Thus,
∂ ′
∂
∂

∂t (ρu i ) + · · · = · · · = ∂t (ρu i ) + · · · = · · · − ∂t (ρu i ) + · · · = · · · . (3.2)
(4) Multiply the equation for the fluctuation u ′i with the fluctuation itself; and obtain an equation for
1 ′ ′ ′
2 u i u i (now and then denoted as k , that is, fluctuating turbulence energy).
(5) Average this equation to obtain an equation for the mean turbulence energy, k = 12 u ′i u ′i :
average
u ′i · ∂t∂ (ρu ′i ) + · · · = · · · = ∂t∂ (ρ 12 u ′i u ′i ) + · · · = · · · −→ ∂t∂ (ρ 21 u ′i u ′i ) + · · · = · · · . (3.3)
Other procedures can also be followed, we have for instance the relation 21 u ′i u ′i = 1
2 ui ui − 12 u i u i
that can be used.
The result, with constant ρ, can be written

∂ ∂ ∂u i ∂ ∂k
(ρk) + (ρku j ) = −ρu ′i u ′j + µ
∂t ∂x j ∂x j ∂x j ∂x j
| {z } | {z } | {z }
ρCk ρ Pk ρ Dk,v
∂ 1 ′ ′ ′ ∂u ′ ∂u ′i
+ − 2 ρu i u i u j − p′ u ′j − µ i . (3.4)
∂x j ∂x j ∂x j
| {z } | {z }
ρ Dk,t ρε
This is a so-called “exact” equation for the turbulence energy k. The term exact is used because
the equation is developed from the basic equations by mathematical operations. No new models
are introduced. (One might well say that the basic equations themselves are “models”.) However,
information on the instantaneous motions is discarded by the decomposition and statistical filtering
(averaging). Although the representation is incomplete, what comes out of the operation is as correct
as what was put into it.
In the equation above, each term is assigned a label. We will come back to these in relation to
the equation for the Reynolds stresses. The first term is the transient term (storage term) and the
convective term (“float with the stream”). Pk is the production term, Dk,t is turbulence diffusion or
mean convective transport with the turbulent motions. The viscous term is split into a gradient term
Dk,v and a dissipation term ε. The dissipation is transfer of mechanical energy to thermal energy
and for the turbulence, this is
!
∂u ′i ∂u ′i ∂u ′j ∂u ′i
′
φ = τi j =µ + . (3.5)
∂x j ∂x j ∂ xi ∂ x j
We re-find the term as a source term in the equation for internal energy (alternatively, enthalpy). It
is a part of the dissipation function 8, see Eq. (A.44) or Eq. (A.46) in the appendix. In Eq. (3.4),
the last term of the sum in Eq. (3.5) is drawn into the gradient term Dk,v . This is a mathematical
operation that is done in order to obtain a gradient diffusion term and the equation is still exact. The
first part of the sum in Eq. (3.5) we find again as as the last term of Eq. (3.4). This term, which
is labeled ε, contains most of the dissipation and the names viscous diffusion and dissipation are
usually adopted for the two terms as they appear in Eq. (3.4).
For a further interpretation of the terms, see Sec. 6.3 on the Reynolds stress equations or, e.g., Hinze
(1975:68–78,323f).
When we introduced the Reynolds decomposition u i = u i + u ′i , p = p + p′ , into the momentum
equation, we got the Reynolds stresses as new unknowns in the system of equations. Here, we have
got even more new correlations: The triple correlation u ′i u ′i u ′j and the pressure-velocity correlation
u ′j p′ . These terms act diffusively. They are unknown and Dk,t has to be modeled. The viscous
gradient term Dk,v (“diffusion”) can be calculated as it includes no new unknowns. At high Reynolds
numbers, it is small in comparison to the turbulence diffusion. The production term Pk contains
only quantities that we know from before. It can be calculated when we have modeled the Reynolds
stresses. The dissipation term ε have to be modeled.
3.3 Modeling of the turbulence energy equation

This section is on modeling of the k equation. The approach is to present
models that have only this equation; k-equation models. However, we bring
this equation with us to the k-ε model. Hence, this section will be part of the
development of the k-ε model in the next section.
Turbulence diffusion: Gradient model
The diffusion term Dk,t contains two quantities that are unknowns and that have to be modeled. In
Chapter 5 we will see how turbulence diffusion can be modeled with a gradient model. This becomes
an analogy to Fick’s law, which is a model for species mass flux, see Eq. (A.18) at page 200. For the
flux of turbulence energy we can use such a gradient model:
µt ∂k
. (3.6)
σk ∂ x j
This is interpreted in two ways:

1) as a model for (− 12 ρu ′j u ′i u ′i − p′ u ′j ), or
2) as a model for − 12 ρu ′j u ′i u ′i , then p′ u ′j is neglected.
The background for and consequences of the two interpretations will be left out here. For the prac-
tical purposes, they give the same result. The diffusion model thus becomes

∂ µt ∂k
ρ Dk,t = . (3.7)
∂ x j σk ∂ x j
We will return to the gradient model for turbulence diffusion in Sec. 5.2.
3.3 Modeling of the turbulence energy equation 51
KAMELEON 96
Turb. kin. energy

Plane: X= 146.4
120
Z (m)
Max
9.82E+01
6.00E+01
100
4.50E+01
3.50E+01
2.50E+01
2.00E+01
80
1.50E+01
1.00E+01
5.00E+00
1.00E+00
60
1.00E−10
Min
320 340 360 380
Time= 800.0 Y (m)
Turb. kin. energy

Plane: Y= 338.9
Z (m)
Max
100
93.761
60.000
45.000
80
35.000
25.000
20.000
60
15.000
10.000
5.000
40
1.000
0.000
Min
50 100 150
Time= 800.0 X (m)
Figure 3.1: Turbulence energy from a numerical simulation of flow around an oil platform.
Behind the tower, the wind (25 m/s) sets up a wake over the heli-deck. Large velocity gra-
dients give much turbulence and this affects the inflight conditions for the helicopter. The
graphs show the turbulence energy in vertical planes across and along the heli-deck. Fig-
ure 2.1(f) at page 32 shows a perspective of the three-dimensional platform model.
Dissipation
Prandtl (1945) imagines a “turbulence ball” in a flow. He assumes a high Reynolds number for the
flow. The ball has a length scale L (mixing length) and a velocity scale u ′ (velocity relative to the
surroundings; fluctuation).
✻ u′ ✲
L
❄
The resistance against the motion has to be F ∼ ρu ′ 2 · A ∼ ρu ′ 2 · L 2 . (Confer the expression for
drag.) The resistance work per unit of time, effect, is F · u ′ ∼ ρu ′ 3 · L 2 . This is loss of energy for
the “ball” that we are looking at. Divided by the volume V ∼ L 3 , this becomes
u′ 3
ρε ∼ ρ , (3.8)
L
√
and with u ′ ∼ k Prandtl obtains
k 3/2
ε = CD . (3.9)
L
Here, CD is a constant numerical value
√ that has to be determined from experiments. A dimensional
analysis, with velocity scale u ′ ∼ k and length scale L, also gives ε ∼ u ′ 3 /L ∼ k 3/2 /L.
This length scale, L, is a mixing length or a characteristic length scale for the larger eddies. It is
not identical to the length scale of Prandtl’s mixing length model, although it is of the same order of
magnitude.
Turbulence viscosity
We can regard the turbulence viscosity as a product of a characteristic velocity scale and a charac-
teristic length scale for the turbulence: νt ∼ u ′ · ℓ′ . This, we can think of as an analogy
√ to the model
for molecular viscosity in kinetic gas theory, cf. pages 40 and 38. With u ′ ∼ k and ℓ′ = L, we
have √
νt = C L k L, (3.10)
where C L is a constant numerical value. This gives us the Reynolds stresses with Eq. (2.14).
The two constants CD and C L are not independent. From experimental data one can set CD · C L
to approximately 0.08–0.11. The value 0.09 is mostly used. We can choose one of the constants to
unity.
Modeled equation for k

∂ ∂ ∂ µt ∂k
(ρk) + (ρku j ) = µ+ + ρ Pk − ρε, (3.11)
∂t ∂x j ∂x j σk ∂ x j
3.4 k -εε model 53

∂u i ∂u j ∂u i
ρ Pk = µt + , (3.12)
∂x j ∂ xi ∂x j
√
µt = ρνt = ρ k L, (3.13)
k 3/2
ε = CD , (3.14)
L
L = · · · [ algebraic expression ]. (3.15)
Models with one equation for a turbulence quantity are called one-equation models. Such k-equation
models are somewhat used for special cases. Because of the empirical expression for L, it is often
claimed that a one-equation model is not significantly better than a mixing length model. This is
valid in some cases, but not in general. The alternative can be to solve an equation to determine the
length scale. This is a more general procedure. In certain cases, however, a special function for L
can give better results than a more general equation.
3.4 k -εε model

The k-ε model has one equation for the turbulence energy k and one equation
for the dissipation ε. The model equation for k is explained in the preceding
section, which should be read first.
In place of an algebraic expression, we can solve a transport equation to determine a length scale.
This equation can be for the length scale itself, or for a quantity k m ℓn (n 6= 0). There are many
model proposals with different values for m and n. The most common and nearly dominating is a
model with an equation for the dissipation term ε, thus m = 3/2 and n = −1.
Modeled equation for ε
An equation for ε can be developed from the basic equations (momentum and continuity equations).
This “exact” equation is, however, of little help except that we can identify terms for diffusion,
production and destruction. We can write
∂ ∂
(ρε) + (ρεu j ) = ρ Dε + ρ Pε − ρ Q ε . (3.16)
∂t ∂x j
For the diffusion term, we can use a gradient model, cf. Eq. (3.7),

∂ µt ∂ε
ρ Dε = µ+ . (3.17)
∂x j σε ∂ x j
Production and destruction are set proportional to the production and destruction of k. The terms
have to be multiplied with ε/k in order to get the right dimension:
ε
Pε = Cε1 Pk (3.18)
k
and ε
Q ε = Cε2 ε. (3.19)
k
The thought behind such a relation is that the more is available the more will be drawn. When the
amount of turbulence energy increases, also the destruction of it will increase; ε have to increase
when k increases. Correspondingly, when there is less to drain, less will be drained. When Pk
increases, ε should increase; otherwise k might grow beyond any limit. An increase in ε is obtained
by increasing Pε – hence, it is reasonable to let Pε depend on Pk . The same argument can be
forwarded for the relation between the destruction Q ε and the dissipation (destruction) ε.
Standard k -εε model:
Turbulence viscosity:
k2
µt = ρνt = Cµ ρ . (3.20)
ε
Reynolds stresses:
∂u i ∂u j 2
−ρu ′i u ′j = µt + − ρkδi j . (3.21)
∂x j ∂ xi 3
Two modeled equations are solved:

∂ ∂ ∂ µt ∂k
(ρk) + (ρku j ) = µ+ + ρ Pk − ρε, (3.22)
∂t ∂x j ∂x j σk ∂ x j

∂ ∂ ∂ µt ∂ε ε ε
(ρε) + (ρεu j ) = µ+ + Cε1 ρ Pk − Cε2 ρε, (3.23)
∂t ∂x j ∂x j σε ∂ x j k k
where
∂u i ∂u j ∂u i
ρ Pk = µt + . (3.24)
∂x j ∂ xi ∂ x j
The constants in the model (from Launder and Spalding, 1974):
σk = 1.0 σε = 1.3 Cε1 = 1.44 Cε2 = 1.92 Cµ = 0.09 (3.25)
The first publication of a k-ε model was by Jones and Launder (1972). This was a model that also
could be used for flows at low Reynolds numbers. The numerical values of the constants were
somewhat adjusted, and the most used version is published by Launder and Spalding (1974). The
model is developed for “incompressible flow”, which here means flow with constant density (cf.
page 26).
3.4 k -εε model 55
This model version, Eqs. (3.20)–(3.25) with the constants shown here, has got a status as a “stan-
dard” k-ε model for high Reynolds numbers. See also page 45 and the discussion of the constants
and the ε equation in Ertesvåg (1991:45). We will come back to the values of the constants in
Sec. 4.5.
Notice that we here consider the density constant. Therefore, a term has dropped out of the model
for the Reynolds stresses, Eq. (3.21). If we let ρ vary, for instance with the temperature, we have to
use the expression in Eq. (2.14). Since the Reynolds stresses enter the production term Pk , there will
also be an extra term in Eq. (3.24) and Eq. (3.12). With varying density, there will also be a term
with a pressure–strain correlation in the k equation. This term is zero with constant density, and is
neglected otherwise as well.
k -εε model for combustion
The density ρ is not constant in combustion – on the contrary, the variations can be very large. We
can then use mass-weighted (mass-averaged) quantities, see Appendix B. The equations for mass,
momentum and energy are largely the same, with some deviations (that usually are neglected), Sec.
B.2–B.5.
A k-ε model can still look like Eqs. (3.20)–(3.25) above. In order to denote that the quantities
are mass-weighted, we can write k̃ and ε̃. There will be a couple of important additions: In the
expression for the Reynolds stresses, Eq. (2.14), a term drops out when the density ρ is constant, as
in Eq. (3.21). This term has to be included and we obtain:

∂ ũ i ∂ ũ j 2 ∂ ũ l
−ρu ′i u ′j = −ρ̄ ug
′ u′ = µ
i j t + − ρ̄ k̃ + µ t δi j . (3.26)
∂x j ∂ xi 3 ∂ xl
This expression also enters the production term so that

∂ ũ i ∂ ũ j ∂ ũ i 2 ∂ ũ l ∂ ũ i
ρ̄ Pk = µt + − ρ̄ k̃ + µt . (3.27)
∂x j ∂ xi ∂ x j 3 ∂ xl ∂ x i
Furthermore, we have a pressure term and a correlation between pressure and strain rate in the
k̃ equation, see Eq. (B.42). These have to be modeled, but no model has appeared that has got
widespread usage. Usually, these terms are neglected. Finally, we can add that the viscous diffusion
term, strictly, cannot be expressed as a gradient term, although we model it that way anyhow.
Boundary values
For boundaries at solid walls, we often use wall functions. We will return to this in Sec. 4.4.
At symmetry lines or -planes, both k and ε have zero gradients and are symmetric about the line or
plane.
Towards free boundaries, such as the outer edge of a boundary layer, a jet or similar (see Sec. 4.3),
the gradients of k and ε will go to zero and the values of k and ε towards the values in the outer
flow. In a practical case, the boundary might not be sufficiently far out, and we have to find some
suitable expression for the boundary value. This can, for instance, come from a simplification of
the equations. In many instances, the outer flow has weak turbulence (low Reynolds number). If
one uses known or estimated low values as boundary values for k and ε, this may cause problems
for the usual (“standard”) k-ε models for high Reynolds numbers. The boundary values might give
Reynolds numbers that are too low for the model to maintain the turbulence. The remedy might be
to use somewhat larger values, or to use a model for low Reynolds numbers (see below).
The values for inflows you have to specify or calculate on the basis of other knowledge of the inflow.
You can, for instance, make estimates for the relevant quantities (see Sec. 9.5).
If you are uncertain of how to specify the boundary values, it will be wise to investigate the sensitivity
of the solution. You must then vary the boundary values within the limits that are regarded as
relevant, and see how this affects the result.
Low Reynolds numbers
The model above has to be modified in order to calculate flows with low Reynolds numbers. Patel,
Rodi and Scheuerer (1985) gives an overview of a number of such models. Most of them have
empirical functions that adjust the constants Cµ , Cε1 and Cε2 . These functions are unity at high
Reynolds numbers. The models have also a term that enhances the production in the ε equation near
a wall.
The model of Jones and Launder (1972) was the first of this kind. The model is slightly modified
by Launder and Sharma (1974) and widely used in that form. It is reproduced by Patel, Rodi and
Scheuerer (1985). At a solid wall k = 0, while ε has a finite value. In some models, among these
the most common, an equation is solved for a modified dissipation variable that has zero value at the
wall.
Such models are developed for flows with nearly constant density. Some of them may seem to
be designed primarily for wall flows. With combustion, with a very large variation in density and
viscosity, it is not certain that these adaptions work as intended. In some cases they perform rather
poor, see Gran, Ertesvåg and Magnussen (1997).
Other adaptions
The standard version of the k-ε model can get additional terms for low Reynolds numbers (see
above) and/or for buoyancy in the flow (next section). We might also find other adaptions of the
model, for instance for compressible flows and other flows with variable density, or with directional
turbulence viscosity, cf. Sec. 6.5. A k-ε model from the so called RNG theory (see page 142)
looks like the model above, except with other values for the model constants. The development is,
however, somewhat different.
3.5 Buoyancy and turbulence 57
3.5 Buoyancy and turbulence

Buoyancy acts on the flow for instance in buildings (ventilation, especially in
large halls), atmospheric flows, fires and smoke plumes. For the standard
model presented above, additional terms are developed to capture the effects
of buoyancy on turbulence.
Buoyancy terms in the k equation
If one uses the Boussinesq-approximation for buoyancy in the momentum equation, an additional
source term appears in the k equation (see Sec. B.6):
G k = α(−u ′i T ′ ) f i . (3.28)
(Notice that the subscript k in G k denotes that it is a source term in the k equation, not a directional
index.) Here, f i is the acceleration of gravity in x i direction, cf. Sec. A.2.4. The Reynolds flux u ′i T ′
can be modeled in the usual way with a gradient model (more on this in Sec. 5.2):
νt ∂ T
−u ′i T ′ = , (3.29)
σT ∂ x i
νt ∂ T
Gk = α fi . (3.30)
σT ∂ x i
The model constant σT is the turbulence Prandtl number, which is often set to 0.9 (see Sec. 5.2).
Stability
Stable stratification occurs when the density ρ decrease upwards. For a homogeneous mixture, ρ is
reduced with increased temperature. In this case, the turbulence and the flow are damped. With x 3
directed upwards ( f i = −gδ3i ), this gives
∂ρ ∂T
< 0; > 0; G k < 0. (3.31)
∂ x3 ∂ x3
In the opposite, for unstable stratification, the density increase upwards and the source term G k is
positive:
∂ρ ∂T
> 0; < 0; G k > 0. (3.32)
∂ x3 ∂ x3
The unstable state will increase the flow and turbulence.

Buoyancy terms in the ε equation
The production term in the ε equation can be extended as

ε
Pε = Cε1 (Pk + Cε3 G k ) . (3.33)
k
For the model constant Cε3 , a variety of values is used for stable and unstable stratification: For
unstable stratification Cε3 = 1, whereas for stable stratification one sets the value to Cε3 = 0,
alternatively in the range 0–0.2 (Rodi, 1980:28,31; 1991). In the latter instance, the source term G k
in the k equation is negative and the term acts as a destruction term.
Notice: This model can be written in different ways, however, with exactly the same content. In
some formulations, (1 − Cε3 ) will correspond to Cε3 above.
The criterion for «stable» or «unstable» can be one of the expressions in Eqs. (3.31)–(3.33), usually
whether G k is negative or positive. For stable stratification, the flow will predominantly be directed
normally to the gravity; that is |u 1 | ≫ |u 3 |. For unstable stratification, the flow will tend more to
follow the direction of gravity; |u 3 | ≫ |u 1 |. This can be utilized to design a function that changes
the value of Cε3 smoothly from 1 to 0.
Buoyancy and combustion
The model above is based on the Boussinesq approximation, which among other things assumes
small variations in the density (cf. Sec. A.2.4). In a flow with combustion, the density will show a
strong variation with temperature and composition.
As opposed to the Reynolds-averaged k equation, buoyancy and other distant forces do not enter
directly into a mass-weighted (Favre-averaged) equation for turbulence energy or Reynolds stress,
see Sec. B.6–B.7. These forces act on the turbulence through the momentum equation. They will
affect the mean flow, pressure field and density field. How this affect the turbulence – and how it
should be modeled – seems not to be much investigated. Jones(1994:317f,322) says more on this,
but states that much remains to be done.
3.6 What can we expect from a k -εε model?

You may read this section in the context, although the full benefit you will get
later, after reading Chapters 6 and 8 and Sec. 9.4. The k-ε is of main relevance,
although what is written here is generally valid for all two-equation models with
a turbulence viscosity.
In Chapter 6 we will look at models where one solves an equation for each of the 6 independent
Reynolds stresses. Such models should give better predictions than a simpler model with only one
equation for the turbulence energy. (Both types of model have an equation for the dissipation or
some other turbulence quantity.) The cost of this is more complex modeling, stronger demand on
computers, and – not least – higher requirements on the user of the model.
3.6 What can we expect from a k -εε model? 59
Here, we will look at the differences between two- and multi-equation models. The difference lies in
the expression for the Reynolds stresses. A model with a turbulence viscosity uses Eq. (2.14), which
is a linear relation between stress and strain rate. We can write (with constant density):

2 ∂u 1 ∂u 1 ∂u 2
ρu ′1 2 = ρ k − 2µt −ρu ′1 u ′2 = µt + (3.34)
3 ∂ x1 ∂ x2 ∂ x1

2 ∂u 2 ∂u 1 ∂u 3
ρu ′2 2 = ρ k − 2µt −ρu ′1 u ′3 = µt + (3.35)
3 ∂ x2 ∂ x3 ∂ x1

2 ∂u 3 ∂u 2 ∂u 3
ρu ′3 2 = ρ k − 2µt −ρu ′2 u ′3 = µt + (3.36)
3 ∂ x3 ∂ x3 ∂ x2
Here we see, first, that the relation between strain rate (velocity gradient) and stress is the same,
regardless of direction. Second, we see that only the strain rates have direction. The turbulence
viscosity is the same in all directions. Forces with a certain direction can not act directly onto the
turbulence through this model. Third, we see that each Reynolds stress can be expressed individually.
Now and then, we can read that the k-ε model gives, or presumes, isotropic turbulence (Sec. 4.1) or
an isotropic state of stress. The meaning may be that the normal stresses are equal. However, as we
see, this is wrong.
Some effects that may cause difficulties for k-ε and other two-equation models, are (following Laun-
der, 1989b; Hanjalić, 1994):
– strong curvature in the flow (recirculation zones, swirl, bends, etc.);
– strong anisotropy in the Reynolds stresses;
– directional forces that influence the turbulence, for instance buoyancy and Coriolis force (but
see Sec. 3.5 on buoyancy and combustion);
– large deviation from equilibrium of turbulence production (that is, where the production Pk de-
viates strongly from the dissipation ε), as in sudden boundary changes, unsteady and periodic
flows – and also when the turbulence is delayed compared to the mean flow.
A boundary layer (see Sec. 4.3) along a wall has one dominating velocity gradient. Quite a few
practical flows have mean strain rates in more directions. For instance, the flow out of a burner in a
furnace can both rotate about its own axis (swirl) and have recirculation zones in planes along this
axis. In such a case, all 9 velocity gradients will have a value, and all of them will influence the
momentum equations. Then, the linear model is not likely to give fully satisfactory representation
of the flow.
A simpler example is the recirculation zone in a plane, two-dimensional flow. This will have two
velocity gradients, only one shear stress and two normal stresses that enters the momentum equa-
tions. (The third normal stress also has a value but cancels from the momentum equations.) In such
cases, one (often) finds that the recirculation zone becomes somewhat shorter, and stronger, with a
k-ε model compared to a model with Reynolds stress equations. Apart from this, the flow pattern is
mainly in accordance with measurements.
Figure 3.2 (from Lai, 1996) shows computed streamlines for the flow in a pipe with sudden expansion
and swirl. For the (upper) recirculation zone behind the step, we see that the length is close to the
experimental data (x/ h = 3.2) for all three models. But the circulation zone (the lower) in the
middle of the pipe behind the step, nearly disappears with a k-ε model. Here, the model with
Reynolds stress equations gives results that are closer to the experimental data (x/ h = 4.2). Also
Figure 3.2: Computed streamlines for a pipe flow with sudden expansion,
from Lai (1996). The lower boundary is the pipe axis. The expansion from
3h to 4h lies on the left side of the figure. The models are a) a model
with Reynolds-stress equations (Gibson and Launder, 1978), see Sec. 6.4;
b) standard k-ε model (Launder and Spalding, 1974), see Sec. 3.4; c) k-ε model based on
RNG theory (see page 142). The graphs to the right show the tangential velocity. Solid
curves are the model in a); dashed lines show the two k-ε models (close resemblance).
for the tangential velocity, the Reynolds-stress equations give considerably better results than the k-ε
models in these simulations. At some distance downstream, the k-ε models give an approximately
straight line, as if the fluid rotates as a stiff body.
For many flow cases, small and weak recirculation zones are of less importance. However, if the
flow in Fig. 3.2 had been a combustion chamber with a burner, the central circulation zone would
have been important for the stabilization of the flame. It could then be decisive for the overall result
that we were able to provide a good representation of this circulation zone.
A more general summary could be that the more velocity gradients, the more the flow is twisted
and bended, the poorer will the agreement be between a two-equation model and measurements.
Furthermore, a two-equation model gives a somewhat coarser picture than a model with Reynolds-
stress equations. If one is interested in small secondary flows, e.g. weak rotation normal to the
main flow or weak recirculation zones in corners or similar, one have to solve the equations for the
Reynolds stresses.
Having said all this, it also has to be mentioned that the current models with Reynolds-stress equa-
tions do not fully remedy the problems listed above. Moreover, both two- and multi-equation mean
3.7 Formal requirements to turbulence models 61
value models (cf. Sec. 9.4, page 141) have some common weaknesses, such as
– scalar length scale, that is, not directionally dependent;
– only one time scale for the turbulence, which is supposed to represent all eddy sizes (cf.
Chapter 8);
– difficulties in modeling effects of walls and other separating interfaces;
– difficulties in modeling effects of variable density, e.g. in combustion.
Eventually, also a k-ε model will usually give a result. The user, considering purpose and practical
circumstances, have to evaluate what is “good enough” or “important” for his or her case.
3.7 Formal requirements to turbulence models

Independent of coordinate system
The turbulence energy k = 21 u ′i u ′i is unchanged, whatever is done to the coordinate system. However,
the distribution among the normal stresses will change. Also the shear stresses will change. The
physical quantities will be the same, although they are expressed in a different coordinate system.
One can construct functions that are independent of the direction. A simple example is the function
(u ′12 )−1 + (u ′22 )−1 + (u ′32 )−1 . It will change when the coordinate system is twisted or skewed.
Therefore, models can not contain such terms.
Realizability
A turbulence model is a set of equations with approximations for some of the terms. From these
equations, we obtain a solution which is a set of values for different quantities. We can put up a
series of requirements that can be applied to these values. The most simple is that the turbulence
energy k and the normal stresses have to be positive. The same holds for the dissipation ε, which is
also a squared value, see Eq. (3.4). From statistics (see Sec. 7.1) one can show that for a covariance
(shear stress) r
q
u i u j ≤ u i · u ′j 2 .
′ ′ ′ 2 (3.37)
From Eq. (2.14) or Eq. (3.34), we find that for constant ρ,

∂u 1 1
νt ≤ k, (3.38)
∂ x1 3
or otherwise the normal stress u ′12 becomes negative.

For mass fractions Yk and mixing fractions ξ (Sec. 5.4), a requirement is that they should be between
zero and unity. One can also show that

ξ′ 2 ≤ ξ 1 − ξ . (3.39)
Virtually all models are made such that the turbulence energy and the dissipation remain positive.
However, quite a few models are not constructed such that we beforehand can be sure that the
requirements of Eqs. (3.37) and (3.38) are met. Among these are the most common of all, the
standard k-ε model (Sec. 3.4). There are, however, also model proposals where the turbulence
viscosity νt behaves according to the requirement of Eq. (3.38).
Several such requirements for the relations between quantities can be developed. Schumann (1977)
and Lumley (1978:31) raise this discussion for turbulence modeling. It is called realizability. The
name indicates that the solution should not be in conflict with the reality. For instance, the turbulence
energy should not be negative. A realizable model is constructed such that all these requirements are
satisfied. This does not necessarily imply that the solution is right – only that it is less wrong. Even
though the solution is possible, it is not sure that it is a good approximation for our case. (Actually
the term should perhaps be non-irrealizable, but the terminology has become as it is.)
The most common and somewhat early models do not pay attention to this. Hence, they can give
non-realizable (unrealistic) results. Newer models account for realizability, although not fully. It has
been seen that such models give solutions with fewer iterations. During the numerical procedure of
trial and error, the models stay at a more narrow and more straight path to the result and away from
solutions that are contrary to reality.
Chapter 4
Some simple flows

4.1 Homogeneous and isotropic turbulence
Homogeneous turbulence: Statistically independent of the location in space, that is, independent
of x i . Averaged turbulence quantities, for instance correlations, have no spatial gradients. Thus,
∂ ′ ′
u u = 0. (4.1)
∂ xk i j
The averaged quantities are unchanged when one moves the coordinate system (translation).
Isotropic turbulence: Statistically independent of direction. For instance is u ′1 2 = u ′2 2 = u ′3 2 . The

averaged quantities are unchanged if one rotates the coordinate system. In practice, the flow has to
be homogeneous. See more in Sec. 7.2, page 115f.
From this follows that shear stresses u ′i u ′j = 0 (i 6= j ) and turbulence fluxes ϕ ′ u ′j = 0 in isotropic
turbulence. This can most simply be explained by Fig. 4.1.
Notice: The terms homogeneous turbulence and isotropic turbulence are used at various levels – not
always made clear by the context. It is then often used in the meaning that the Reynolds stresses, or
even only the normal stresses, are homogeneous/isotropic. Then, for instance, the tripple correlations
can have spatial gradients and the turbulence diffusion is non-zero.
Warning: Now and then, we can read that the k-ε model presumes that the turbulence is “isotropic”.
Often, it is then meant that the normal stresses are equal. This is, however, not correct. We can define
the turbulence energy as the sum k = 12 (u ′1 2 + u ′2 2 + u ′3 2 ), independently of whether they are equal
or not. Also the k-ε model distributes the energy among the components, cf. page 59.
64 4 Some simple flows
x2 x2
Ej Ej
x1
x1
(a) j1 = 0, j2 = j = | Ej| > 0 (b) Rotate 20◦ : j1 < 0, j2 > 0
Ej
Ej
x2
x2
x1 x1
(c) Rotate 90◦ : j1 = − j < 0, j2 = 0 (d) Rotate 110◦ : j1 < 0, j2 < 0
Figure 4.1: Isotropic turbulence requires turbulence fluxes and turbulence shear stresses (flux
of momentum) to be zero. The figures show a flux Ej through an area element. We rotate the
coordinate system and see that the flux components j1 and j2 (which may, e.g., be ϕ ′ u ′1 and
ϕ ′ u ′2 ) change. The turbulence is only isotropic if j1 and j2 are unchanged with rotation. This
requirement can only be satisfied by j = j1 = j2 = 0.
4.2 Isotropic decaying turbulence 65
4.2 Isotropic decaying turbulence

When the turbulence is isotropic, the normal stresses are equal, shear stresses are zero and spatial
gradients of averaged values are zero. The Reynolds-stress equation becomes an equation for k =
3 ′2
2u when u ′1 2 = u ′2 2 = u ′3 2 = u ′ 2 . Without spatial gradients, the production term, diffusive terms
and convective terms become zero. The equation for k is then simplified to
dk
= −ε. (4.2)
dt
In the same way, the modeled ε equation becomes
dε ε2
= −Cε2 . (4.3)
dt k
In this very simple system, one can study the dissipation alone, without influence from any other
process. The system is sufficiently simple to be solved analytically.
Experimentally, one can achieve a good approximation: A wind tunnel is devised such that the mean
velocity is constant both crosswise and along the flow: u 1 = U = constant, u 2 = u 3 = 0. This is a
flat velocity profile that does not change downstream. A grid is placed in the wind tunnel to produce
turbulence, see Fig. 4.2. At some distance downstream the grid, the flow has been smoothed and
all mean velocity gradients are zero. This gives a zero production of turbulence energy. If we let
the coordinate system move with the mean flow (constant velocity), the mean flow is zero.1 We can
transform a spatial gradient to a time gradient by ∂/∂t = U ∂/∂ x 1 (Taylor hypothesis; see e.g. Hinze
1975:46f).
grid wake
❄region isotropic turbulence
U ✲ U ✲
Figure 4.2: Wind tunnel with grid
In such an experiment, all the diffusive terms are not equal to zero in reality. There will be a spatial
gradient along the wind tunnel. The terms are, however, so small that we can neglect them. In
reality, the flow in the wind tunnel is not stationary either. But the changes occur slowly compared
1 This is a simple example of a Galileo transformation, see page 266.
to the time scale for the smaller eddies, and therefore, we can regard the flow as isotropic (Hinze,
1975:176).
Experiments (summarized by Ferziger, 1980) show k ∼ t −n , where n ≃ 1.25 ± 0.06.
The system of Eqs. (4.2)–(4.3) can be solved as
1 1
k ∼ t 1−Cε2 or Cε2 = 1 + . (4.4)
n
This is utilized to determine the value of the constant Cε2 in the k-ε model (cf. page 77) and
correspondingly in other models (see for instance Ertesvåg, 1991:46f).
4.3 Boundary layers

The theory for boundary layers gives the basis for the boundary conditions we
apply towards solid surfaces. Furthermore, testing against experimental data
for such flows is important for the development of turbulence models.
Those feeling that this matter is completely new and unknown, will most likely
benefit from studying a more introductorial textbook on fluid mechanics or heat
and mass transfer.
The boundary layer analysis originates back to Prandtl (1904). He observes a liquid at low viscosity
(e.g. water) that flows by a wall or a solid body (Fig. 4.3). He presumes that the liquid sticks to
the solid surface, so that a boundary layer is formed between the flow and the surface. With a small
viscosity, this connecting layer will be thin compared to the length scale of the flow direction. Then
he can simplify the equations of motion as
Figure 4.3: Apparatus for studying boundary layers, drawing from Prandtl (1904)
4.3 Boundary layers 67
∂u 1 ∂u 2
+ = 0, (4.5)
∂ x1 ∂ x2

∂u 1 ∂u 1 1 ∂p ∂ ∂u 1
u1 + u2 =− + ν , (4.6)
∂ x1 ∂ x2 ρ ∂ x1 ∂ x2 ∂ x2
where he assumes the pressure gradient to be given. The flow is regarded stationary and two-
dimensional.
For a turbulent boundary layer, the averaged momentum equation becomes

∂u 1 ∂u 1 1 ∂p ∂ ∂u 1 ′ ′
u1 + u2 =− + ν − u 1u 2 . (4.7)
∂ x1 ∂ x2 ρ ∂ x1 ∂ x2 ∂ x2
Equations for other quantities are simplified similarly; the crosswise gradient dominates over the
streamwise gradient. One exception is the pressure: The crosswise gradient is very small compared
to the streamwise gradient. Figure 4.5 shows the experimental data and computed results for a
boundary layer along a long, plane surface. Notice the ratio of the streamwise length (x 1 = 4.33 m)
and the thickness (δ = 0.076 m).
Compared to the entire flow, the boundary layers at a solid boundary is often a tiny and thin part of
the flow. It is however, an important part as forces and heat is transferred through this layer.
The boundary-layer approximations can be used for some other flows that are not influenced by
solid surfaces, for instance jet flows and mixing layers (half jets). The name boundary layer is
occasionally used for such flows as well. Thin shear layers is another generic term for such flows.
Figure 4.4 shows sketches of some of them. Boundary layers, channel flows, jet flows and other
similar flows are test cases for turbulence models since the equations are relatively easy to solve
numerically. Launder and Spalding (1972) discuss various models for turbulence and compare with
experimental data for such flows.
There is an extensive literature on boundary layers, from the time of Prandtl till today. Most books
on mass and heat transfer or fluid mechanics have a shorter or longer section on the issue. More
comprehensive expositions can be found in the books of Schlichting (1979), Cebeci and Bradshaw
(1984), and several others. In thousands of papers, boundary layers and other similar flows are
studied and discussed from a great diversity of viewpoints, approaches and aims.
Here, the discussion will be restricted to the inner layer, roughly spoken to x 2 < 0.2δ. In a numerical
simulation, a suitable modeling of this region is important. The outer layer or wake region will be
left out here, as it is modeled by the ordinary turbulence models.
Dimensionless quantities near a wall
In the boundary√layer, we can define dimensionless quantities by the viscosity ν and the friction
velocity u τ ≡ τw /ρ, where τw is the shear stress at the wall. The dimensionless mean velocity
and distance from the wall then become
u1 x2u τ
u+1 = and x 2+ = . (4.8)
uτ ν
u∞ u1
δ
δ u2
(a) Boundary layer along a flat plate (b) Mixing layer between two parallel flows
u∞
u∞
u∞
um δ δ
u jet b
(c) Wall jet along a flat plate (d) Wake
A plane jet has an outlet with a larger

extension in the third direction (normal to
the paper plane).
A round jet has a round outlet, e.g. a pipe.
1
A radial jet is symmetric about a vertical
b 2 us axis upstream the outlet (i.e. about a line to
us the left of the sketch).
u jet
b 1 b is the half width. This is the distance
2 us
from the center-line to where the velocity is
equal to half the centerline velocity, u s
(e) Jet (free)
Figure 4.4: Sketches of different types of boundary-layer flows or thin shear layers. Notice
that in the sketches, the crosswise length is exaggerated compared to the streamwise length;
cf. data for δ and x 1 in Fig. 4.5. Near the starting point they can often not be regarded as thin
shear layers. Especially, this applies for a wake but also to some degree for the other flow
types.
4.3 Boundary layers 69
u 1 /u s
1
0.8
0.6 Klebanoff (1955): Measurements in a bound-

ary layer with zero pressure gradient (flat plate)
0.4 k-ε model Free flow: u s = 15.2 m/s
Klebanoff Thickness: δ = 0.076 m after x 1 = 4.33 m
0.2
0
0 0.02 0.04 0.06 0.08
y (m)
−u ′1 u ′2 /u 2s k/u 2s
0.0015 0.01
0.008
0.001
0.006
0.004
0.0005
0.002
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
y/δ y/δ
Figure 4.5: Boundary layer. Computations (Ertesvåg, 1991) with a k-ε model for low
Reynolds numbers (Launder and Sharma, 1974) and experimental data (Klebanoff, 1955).
This is a “classical” data series that virtually all turbulence models are tested against.
These are now and then called inner quantities or inner scales. One can also use outer quantities;
the free stream velocity u ∞ and the boundary-layer thickness δ. This is customary when looking at
the entire boundary layer, not only the layer near the wall.
A layer very close to the wall
In the region x 2+ < 100, the flow is strongly intermittent, with intense whirls between more quiet
zones. The fluctuations about the mean value can be very large. This is described by, among others,
Hinze (1975:656f) and Landahl and Mollo-Christensen (1986:111f). This region is characterized by
large gradients in mean quantities as u 1 , k and u ′1 u ′2 . Substantial fractions of the production and
dissipation take place at x 2+ < 70. The maximum value for production of turbulence energy is found
about x 2+ = 10, while the dissipation has its maximum value at or very close to the wall (Hinze,
1975:656; Kim, Moin and Moser, 1987; Spalart, 1988). In spite of the very complex flow pattern,
we assume that the same averaging procedures can be used throughout the entire boundary layer.
When approaching the wall (x 2 → 0), all velocity fluctuations go to zero. The ratios of the fluctu-
ations to mean values, however, go towards finite values. Kreplin and Eckelmann (1979) find that
u ′1 /u 1 ≈ 0.25 and u ′3 /u 1 ≈ 0.065 when x 2 = 0. This means that the fluctuations from the mean
values still are intense. The turbulent transport near the wall is, however, not large in comparison to
the molecular. From direct simulations, Kim, Moin and Moser (1987) (see also Mansour, Kim and
Moin, 1988), Spalart (1988), it is found that for momentum, Reynolds stresses and dissipation of tur-
bulence energy, the molecular transport is of the magnitude of or larger than the turbulent transport
when x 2+ < 10–12. This region is customarily called the viscous sublayer (or the laminar sublayer).
This does not, indeed, mean that it is free of strong whirls and fluctuations but that the viscous forces
are strong.
The variation of fluctuations and of statistical quantities as k, ε and Reynolds stresses near the wall
can be analyzed by series expansion, see Ertesvåg (1991:11) and Patel, Rodi and Scheuerer (1985).
4.4 A logarithmic region in boundary layers – wall functions

In this section, we will go through some of the most important wall functions
(wall laws) that are used as boundary conditions for turbulence models.
Section 5.6 on thermal boundary layers is closely related to this section, but
will come after the treatment of mass and heat transfer.
The region near a wall is characterized by large gradients in the mean quantities. In order to simulate
a flow numerically, we have to assign a fine grid to this region. In practice, this requires too much
computer resources. A usual way to remedy this is to compute the outer field down to a point (node)
near the wall, and then use wall functions as boundary conditions, bridging from the wall to the
near-wall point. This point should preferably lie in the logarithmic region of the boundary layer. For
this reason, we will have a closer look at the log-layer.
The boundary conditions can be implemented in a variety of ways. To describe all the solutions
4.4 A logarithmic region in boundary layers – wall functions 71
30
u+
1
25 u+ +
1 = x2
20
15
,0
+ +5
10 1 ln x 2
+ = 0,41 k-ε-modell
u1
Klebanoff (1955)
5
0
1 10 100 1000
x 2+
Figure 4.6: Boundary layer. Velocity profile in a semi-logarithmic diagram. Computations

and experimental data as in Fig. 4.5. In this case, the boundary layer thickness is δ + = 2774
or δ = 76 mm. The limit of the linear region, x 2+ ≈ 10, corresponds to x 2 = 0.3 mm, and
the outer limit of the log-layer is approximately x 2+ = 450 or x 2 = 12 mm. The outer region,
outside the log-layer, is called the wake region.
adopted by program developers would be far too extensive. I believe that it will be better to give a
theoretical foundation and then leave the reader to evaluate each instance by himself.
The wall functions are developed for boundary layers with no pressure gradient, without mass trans-
fer through the wall and for smooth walls. Eventually we will shortly see how the wall functions are
affected by pressure gradients along the wall, mass flow through the wall and roughness.
Wall function for velocity
Quite close to the wall, at x 2+ < 5 (or 5–10, the value shows some variation in literature), a turbulent
boundary layer can be regarded as laminar, thus u + +
1 = x 2 . This curve is shown in Fig. 4.6, and the
region is called the laminar or viscous sublayer.
For a region slightly farther from the wall, experiments show that the gradient is
du 1 uτ
= , (4.9)
d x2 κ x2
where κ is known as von Kármán’s constant. The integral of Eq. (4.9) is

1 1
u+
1 = ln x 2+ + D = ln (E x 2+ ), (4.10)
κ κ
where κ, D and E = exp (κ D) are constants. In a semi-logarithmic diagram, this becomes a straight
line, see Fig. 4.6. An expression like Eq. (4.10) is called a wall function or a wall law.
Schlichting (1979:601f) shows experimental data by Nikuradse and Reichardt that give κ a value of
0.37 – 0.41 and D a value of 4.5 – 6.0. Schlichting states κ = 0.4 and D = 5.5, while Coles and Hirst
(1968) choose κ = 0.41 and D = 5.0. Others have also used κ = 0.42 – 0.43. For E a value of 9.0 is
often applied, although we can find values used up to 10 (Launder, 1988).
It is assumed that Eq. (4.10) is normally valid from x 2+ ≈ 30 and outwards to x 2 ≈ 0.15δ. The latter
figure may correspond to a x 2+ in the range 1000–3000. This region is often called the logarithmic
layer or the log-region. The region between x 2+ ≈ 5 and x 2+ ≈ 30 is a transition or buffer layer. The
diagram in Fig. 4.6 shows u + +
1 as a function of ln x 2 .
One can discuss whether Eqs. (4.9)–(4.10) is curve-fitting to experimental data or whether they
are first developed from theory and then confirmed by experiments. It has also been doubted that
experiments actually give very much support to these curves. Here, we will leave out the discussion
on these matters.
Numerical computer programs often use the wall function

+ + 1 +
u 1 = min x 2 , ln x 2 + D . (4.11)
κ
Here, one neglects the deviation from experimental data in the transition between the two expres-
sions.
Approximately constant shear stress
According to theory, the forces from the wall will be transferred outwards through the nearest layer.
Experiments show that
τ = τvisk + τturb ≈ constant = τw (4.12)
is a reasonable approximation for the log-region and nearer to the wall.

We can also make this the starting point of our development; that is, we can presume constant shear
stress. Then we can talk about the region near the wall as approximated to a Couette flow (which
is a flow of constant shear stress). This implies that we regard Eq. (4.12) as a simplification of the
momentum equation. For constant (turbulence) shear stress, the dimensional analysis determines that
outside the viscous sublayer, the velocity gradient has to be proportional to u τ /x 2 , as in Eq. (4.9).
In the log-region, τvisk ≪ τturb so that
τturb = −ρu ′1 u ′2 ≈ ρu 2τ = τw . (4.13)
This gives the turbulence viscosity
−ρu ′1 u ′2 ρu 2
νt = ≈ u τ = κ x2u τ . (4.14)
∂u 1 ρ
τ
ρ κ x
∂ x2 2
In the transformation, the approximation in Eq. (4.13) is used above the fraction line and the velocity
gradient of Eq. (4.9) under the line. The last equality is simply a rearrangement of the quantities.
So far, nothing is said about modeling. With Prandtl’s mixing length model, νt = ℓ2 |∂u 1 /∂ x 2|, and
the u 1 -gradient from Eq. (4.9), we have
uτ
νt = ℓ2 . (4.15)
κ x2
When this is set equal to the last expression of Eq. (4.14), we obtain
ℓ = κ x2. (4.16)
This expression for the mixing length is a consequence of the measurements that provide the log-law,
Eq. (4.10), and the approximation of Eq. (4.12).
Approximate equilibrium for turbulence energy
In the logarithmic layer, we can assume that convection and diffusion of turbulence energy (k) are
small. This means that the production of k is approximately equal to, or in equilibrium with, the
dissipation:

∂u 1 2
ε = Pk = νt . (4.17)
∂ x2
This is in fair accordance with experimental data. The approximation is not valid in the layer near-
most the wall. The region where the approximation is valid is called the equilibrium region. With νt
from Eq. (4.14) and the gradient ∂u 1 /∂ x 2 from Eq. (4.9), we obtain
u 3τ
ε= . (4.18)
κ x2
This is used as a wall function for ε.
If we use the turbulence viscosity from the k-ε model together with Eqs. (4.17), (4.9) and (4.18), we
have
u 3τ 2 κ x2 uτ 2
ε= = Cµ k 3 = Pk . (4.19)
κ x2 u τ κ x2
This can be rearranged to

2 !2
u 2τ −u ′1 u ′2
Cµ = ≈ . (4.20)
k k
The latter approximation comes from Eq. (4.13). The expression can also be rearranged so that
p
Cµ k = u 2τ . (4.21)
This is equal to τw /ρ, which is a constant for a given value for x 1 .

In the k-ε model, Cµ is a constant equal to 0.09 = (0.3)2. Experiments show values of (−u ′1 u ′2 /k)
ranging from 0.2 to 0.3 in the log-region (see Ertesvåg 1991:24). The value of Cµ is determined on
the basis of these values.
The approximations for the velocity gradient, Eq. (4.9), for the shear stress, Eq. (4.13), and equilib-
rium, Eq. (4.17), are each by themselves reasonably good. Together, they imply that the turbulence
energy k should be constant in the log-region. This is, at best, a rough approximation.
Actually, not all boundary layers are in equilibrium, rather, on the contrary. The approximation in
Eq. (4.21) can be as good as anything else. The expression is used in many simulation programs,
either for determining the u τ at the wall or to find k in the node close to the wall.
Pragmatic approach
The theory above is valid for simple boundary layers at long, plane walls. In practice, we would like
to have a wall function that can be used through many kinds of wall layers. At walls, we often have
recirculation zones and stagnation points. In the stagnation point, the shear stress equals zero. The
wall functions, as in Eqs. (4.18) and (4.21) then give zero turbulence energy and zero dissipation.
Moreover, Eq. (5.31) gives zero heat transfer, whereas it right here perhaps should have a maximum
value, cf. Launder (1988).
One can safely state that the wall function is unphysical here. However, we need a result and then
we have to find a solution. The following alternative is widely used. The theory behind is exactly
as described above – although the result can√ turn out quite differently. The trick – as it indeed is
– consists of using Eq. (4.21) to introduce k in place of u τ in some of the expressions (Launder,
1988; Launder and Spalding, 1974).
For the logarithmic layer – or more correct, for a region where we assume the turbulence energy to
be constant – we can put up
τw ∂u 1
≈ −u ′1 u ′2 = κ ∗ k 1/2 x 2 , (4.22)
ρ ∂ x2
1
u ∗1 = ∗ ln (E ∗ x 2∗ ), (4.23)
κ
u 1 k 1/2
u ∗1 = , (4.24)
τw /ρ
x 2 k 1/2
x 2∗ = , (4.25)
ν
1
E∗ = ∗ ∗
∗ exp κ x 2,v . (4.26)
x 2,v
For the viscous region, as before

τw
u1 = x2, (4.27)
ρν
and the velocity u 1 is assigned the smallest of the values from Eqs. (4.27) and (4.23). The two
curves meet at the distance x 2 = x 2,v , where the turbulence energy becomes k = kv .
No new theory or new knowledge is introduced here – itpis the same stuff, except formulated some-
what differently. The starting point is Eq. (4.21), u 2τ = Cµ k. This we put into Eq. (4.14) for the
1/4
turbulence viscosity, νt = κ x 2u τ , and with κ ∗ = κCµ we obtain Eq. (4.22).
Now we integrate Eq. (4.22) and get u ∗ (defined in Eq. 4.24) expressed by a logarithmic function of
x 2 , plus an integration function. The latter we determine by using Eq. (4.27) as the limit at x 2 = x 2,v.
The rest is simply a rearrangement of quantities and definition of symbols.
If we introduce Eq. (4.21) for u τ into Eq. (4.18), we obtain an alternative wall function for the
dissipation:
k 3/2
ε = Cµ3/4 . (4.28)
κ x2
We can once again find that
u ∗1 = (Cµ )−1/4 u +
1 and x 2∗ = (Cµ )−1/4 x 2+ . (4.29)
If the turbulence energy is constant, the expressions of Eqs. (4.22)–(4.28) are identical to the wall
functions above. It can be mentioned that Launder (1988) keeps k constant in the development of
the wall function, although still allowing k to vary in the simulations. This appears to underline the
point that it is a matter of a pragmatic approach to the problem.
Wall functions for k and ε
For the dissipation, the expression of either Eq. (4.18) or (4.28) is usually used as a wall function.
In boundary layers near equilibrium, Eq. (4.21) can be used as a wall function for the turbulence
energy. However, this works less good for flows with stagnation points, recirculation zones or other
conditions where the near-wall transport is important. One can then resolve the k equation for the
cell nearest to the wall (the wall-function cell of the grid). The boundary condition will be zero
transport through the wall. One then has to put up a model for production and dissipation for the
near-wall cell. This has to be made in two parts, with one expression for the viscous sublayer and
one for the log-layer. One then has to integrate through the entire cell. How this can be done is
shown by among others Launder (1988).
In some cases it may happen that the entire wall-function cell is put inside the viscous sublayer. Also
this we have to account for to avoid strange results. One way of doing it is to assign the wall-function
values for k and ε to the values at the edge of the sublayer. The practical effect is that the equations
get suitable values for the boundary.
To compute or not to compute – That is the question
It is not hard to find researchers who criticize wall laws and point out “errors” or problems in the use
of such functions. If we include wall functions of mass and heat transfer (thermal boundary layers,
Sec. 5.6), the reading can soon become quite miserable. Bradshaw and Huang (1995) summarize
and discuss some of these matters and what they regard as consequences for the Reynolds-averaged
turbulence model we use.
Launder (1988,1984) also develops and discusses the wall functions. He primarily advocates models
that can resolve the entire boundary layer with no wall functions, what we call low-Reynolds-number
models. For certain purposes – where the walls are not too numerous or not too complex while still
not quite simple – this is a suitable and feasible approach. It requires a finer grid and hence, more
resources – but this does not need to be a big problem. It becomes worse when there is a multitude
of solid surfaces in complex constructions. For instance, it is nearly un-thinkable to resolve all the
wall layers around and inside an oil platform.
On the first hand, we need a result. In spite of doubt and uncertainty, this is the best we can provide
in many occasions. On the other hand, we should not forget that the problems indeed exist. This
means that we have to interpret the results with all we have of knowledge, wisdom and experience.
Boundary layer with pressure gradient: influence on wall functions
The influence of pressure gradients on the wall functions is discussed by, among others, Kays and
Crawford (1993:206). Here, I include only a short summary:
– Accelerating, i.e. forward pressure (∂ p/∂ x 1 < 0), gives a thinner wake layer (the outer region)
– and hence, a larger log-layer. The wall function (the log-law) is valid farther out from the
wall.
– Retarding, i.e. adverse pressure (∂ p/∂ x 1 > 0), enhances the wake layer and the log-layer
4.5 Constants in the k -εε model 77
becomes thinner. If the pressure gradient is sufficiently large, the log-layer may disappear.
In both cases, the log-law falls along the same line as for the zero pressure gradient.
Effects of flow through the wall
In some cases, fluid flows out of or into perforated or porous walls, that is u 2 6= 0 for x 2 = 0.
The effects on the wall functions compare to the effect of pressure gradients. The log-curve will,
however, not follow the same line as without through-flow.
– Outflow from the wall causes a thinner log-layer (cf. adverse pressure/retarding), and the
log-curve becomes steeper.
– Suction/inflow towards the wall gives a thicker log-layer, cf. forward pressure/acceleration,
and the log-curve becomes less steep.
See more, for instance in Kays and Crawford (1993:227).
Effects of wall roughness
The wall functions above are developed for smooth, plane surfaces. For many boundary surfaces
this is a good, or at least usable, approximation. However, many flows are bounded by surfaces that
certainly are not smooth, and the modeling has to account for this. Walls can be uneven in many
ways. The inside of a concrete pipe will be uneven, although it will look much the same all the way;
the roughness is evenly distributed. The same holds for a grain field, and to some extent for a forest.
In other landscapes the roughness can vary a lot; from lakes and waving grass fields to scattered
woods in broken terrain or urban areas.
Different types of surfaces are modeled with roughness elements with a characteristic length scale.
The roughness elements used in experiments are often sand (sand paper) or a variety of small bodies
(tops, cones, blocks, etc.) on a plane. When this is to be transferred to real cases, one has to do the
approximation that the roughness is evenly distributed. Eventually, one has to discuss whether this
is a (sufficiently) good approximation for the actual case.
Near a smooth wall, we imagine that there is a thin sublayer, which we regard as laminar (see above).
A rough wall will disturb this sublayer and may fully disrupt it. The flow will be retarded so that the
near-wall velocity becomes somewhat reduced compared to a similar flow along a smooth wall.
Figure 4.6 shows the log-region for a smooth wall. Starting from this, the curve for a rough wall will
be parallel, although lower. This means that the profile still (nearly) follows a logarithmic profile
but that the linear layer is more or less gone and the flow is retarded. One of several methods for
modeling rough layers is shown by Kays and Crawford (1993:230f).
4.5 Constants in the k -εε model

In the k-ε model (page 54), we have the constants Cµ , Cε1 , Cε2 , σε and σk .
For isotropic turbulence we simplified the model as much as only one single constant was left, Cε2 .
From experiments, we found a value, see Eq. (4.4).

√
In the logarithmic region of the boundary layer, we found the constant Cµ as the ratio u τ / k,
Eq. (4.21). Similarly, we can put up the ε equation for the log-layer. We neglect the net convection
and are left with diffusion, production and destruction. In the equilibrium layer, Pk = ε and we use
the expressions
u3 u2
νt = κ x 2u τ , ; ε = τ and k = p τ , (4.30)
κ x2 Cµ
and develop the relation

κ2
Cε2 − Cε1 = p . (4.31)
σε Cµ
With these relations, plus (and not least) trial and error (“computer optimalization”) Launder an co-
workers found a set of recommended values for the constants in the k-ε model. See Launder and
Spalding (1974) and Eq. (3.25).
The constant values of Eq. (3.25) give a reasonably correct development of the boundary-layer thick-
ness along a plane wall, for the width of a plane mixing layer, for the half width of a plane jet and
a radial jet (see Fig. 4.4). However, for a round jet, the spreading becomes too large. Experimental
data show b1/2/x 1 ≈ 0.105 for plane jets and b1/2/x 1 ≈ 0.085 for round jets. The standard model
gives approximately 0.10 and 0.12 for the two cases; thus larger spreading for a round jet, not less.
This discrepancy is called the round-jet/plane-jet anomaly (“anomaly” means inequality, deviation,
irregularity.)
If we set Cε2 ≈ 1.80, the spreading will be approximately correct for a round jet (although too low
for the others). This can be a suitable approach in cases where you only need a turbulence field in a
simple round jet. It may be that you are, for instance, investigating a combustion model. But if the
jet confluence with other flows, for instance a cross wind, the change can not be recommended.
Bradshaw and Huang (1995) discuss wall functions and their use in the development of turbulence
models. They point out that virtually all Reynolds-averaged turbulence models are calibrated against
wall functions in simple boundary layers. One question they rise is whether such models are valid
much beyond the validity of the wall functions themselves. Seen from their point of view, the answer
is not positive.
Chapter 5
Energy and mass transfer

(scalar transport)
5.1 Mixing by turbulence
When different substances are brought together in a flow, the turbulent motions are the most im-
portant mechanism for mixing. The experiment of Reynolds (1883) shows this very clearly, see
Figs. 5.1 and 1.6. In a laminar flow, the dye remains in a thin streak in the tube. Diffusion – trans-
port and mixing with molecular motions crosswise of the flow – has only a weak effect on the flow
pattern. Where the flow becomes turbulent, we see that the color streak is immediately mixed into
the surrounding clear water.
Turbulent motions are spread over a wide spectrum of small and large eddies. We can imagine a
droplet of colored water inserted into water with evenly distributed small-scale turbulence (homo-
geneous turbulence). The droplet will be distorted as shown in the sketch of Fig. 5.2. The sketch
also shows the effects of molecular mixing. With some distortion, the molecular diffusion becomes
considerably more efficient. This is because a much larger part of the flow has gradients in the
concentration.
When the droplet is much less than the eddies of the flow, it will be stretched inside an eddy. The
sketch of Fig. 5.3 shows how this can be imagined.
Figure 5.1: Mixing in tube flow, Reynolds (1883). A thin streak of colored water is led into
a pipe with plain water. First the flow is laminar (to the left) and then changes and becomes
turbulent.
80 5 Energy and mass transfer (scalar transport)
Figure 5.2: Sketch of turbulent mixing of dye, the eddies are less than the droplet. The curve
shows the concentration along a cross section, without molecular diffusion (above) and with
molecular diffusion (below). From Corrsin (1961).
Figure 5.3: Sketch of stretching of droplet of dye in eddies larger than the droplet. From
Corrsin (1961).
5.2 Model for turbulence fluxes 81
5.2 Model for turbulence fluxes

When we introduced the Reynolds decomposition in the basic equations, Sec. 2.2, new unknown
variables appeared: The Reynolds stresses u ′j u ′i and the Reynolds fluxes u ′j Yk′ and u ′j h ′ . These are
transport (flux) of momentum, mass and energy with the turbulent motions.
Gradient model for turbulence fluxes
In the momentum equation we introduced a turbulence viscosity to model the Reynolds stresses, see
Eq. (3.21). This is an analogy to the molecular viscosity in the model for molecular flux (Sec. 2.5).
In the same way, we can model turbulent scalar fluxes by turbulence diffusivities:
∂h νt ∂h
−u ′j h ′ = αt = , (5.1)
∂x j σh ∂ x j
∂Y k νt ∂Y k
−u ′j Yk′ = Dt = . (5.2)
∂x j σY ∂ x j
Here, σh is the turbulence Prandtl number and σY turbulence Schmidt number. Both are often set
equal to 0.9, although other values are also in use.
Figure 5.5 shows a computation of spreading of warm exhaust over a helicopter deck. Here, the
diffusion models in Eqs. (5.1) and (5.2) are used together with the standard k-ε model (Sec. 3.4).
The turbulence diffusivities are set proportional to the tur-
bulence viscosity. This becomes an approximation to the
Reynolds analogy, which strictly says that ν = α and νt =
αt , or that the molecular Prandtl number and the turbulence
ℓ′ Prandtl number both are unity. We can do this approxima-
tion because turbulent transport of momentum, mass and en-
ergy are related to the same motions. It is the same “ball”, to
use Prandtl’s picture, Sec. 2.6. In the mixing length model, a
”ball” has a velocity, a species composition and a temperature
Figure 5.4: The same
that it carries in the motion.
whirling motions transport
momentum, mass and en- We can write this as a general gradient model for turbulence
ergy. flux with turbulence viscosity,
νt ∂ϕ
−u ′j ϕ ′ = , (5.3)
σϕ ∂ x j
or as a diffusion model,
∂ νt ∂ϕ
Dϕ,t = . (5.4)
∂x j σϕ ∂ x j
We call σϕ a turbulence Prandtl/Schmidt number.

KAMELEON 96
Temperature (K)
Plane: X= 267.6
Z (m) Max
600 351.51
353.00
343.00
323.00
580
313.00
308.00
303.00
560
300.00
297.00
294.00
540
291.00
288.00
50 100 150 Min
Time= 700.00 Y (m)
Temperature (K)
Plane: Y= 92.36
Z (m)
Max
600
342.80
353.00
343.00
323.00
580
313.00
308.00
303.00
560
300.00
297.00
294.00
540
291.00
288.00
250 300 350 Min
Time= 700.00 X (m)
Figure 5.5: Mean value simulation and exhaust dispersion over a heli-deck (by Mette Bugge).
Here, a k-ε model and gradient models for turbulence diffusivities are used. The density
is approximately reciprocal of the temperature so that for each 3 ◦ C (1 % of the absolute
temperature) the density, and hence the lift, is reduced by 1 %. Less lift for the helicopter has
to be compensated with less payload. Therefore one wishes to position the flue gas outlets
so that the warm gas rarely flows over the heli-deck. Here, we see that the isotherm of 3 ◦ C
above ambient lies right down to the heli-deck, and in an area over the deck, the temperature
is 25–45 ◦ C higher. In this case, the exhaust prohibits the helicopter inflight.
5.3 Diffusion model with u ′i u ′j equation 83
Model equations for turbulence fluxes
In the next chapter, we will look at equations for the Reynolds stresses:
∂ ′ ′ ∂ ′ ′
ρu i u j + ρu i u j u ℓ = · · · [terms that have to be modeled] . (5.5)
∂t ∂ xℓ
The next step will be to construct equations for the turbulence fluxes:
∂ ′ ′ ∂ ′ ′
ρu j T + ρu j T u ℓ = · · · [terms that have to be modeled] . (5.6)
∂t ∂ xℓ
∂ ′ ′ ∂ ′ ′
ρu j Yk + ρu j Yk u ℓ = · · · [terms that have to be modeled] . (5.7)
∂t ∂ xℓ
!
∂ ′ ′
′
∂T ′ ∂u j
These equations can be developed using the relation u ′j + T′ = ujT .
∂t ∂t ∂t
Such equation are somewhat used, although not very common. It has been said that with so many
modeled terms, the entire calculation becomes uncertain. With development and testing, the validity
of this argument has faded over time. A large number of equations requires huge computer resources,
although this is an argument that has become weaker. For pure heat-transfer problems, equations for
turbulence heat fluxes are sometimes used together with the equations for the Reynolds stresses.
This might be, e.g., for cooling in the nuclear industry. However, in combustion problems, they are
farther away. The reason is that so many other things are uncertain, e.g., models for radiation, soot,
reaction mechanisms.
What is of relevance, is to solve the transport equations for the Reynolds stresses. In the three-
dimensional case, there are 3 equations for shear stresses and 3 equations for normal stresses. In the
literature, Launder (1989b), can be a suitable starting point.
5.3 Diffusion model with u i′ u ′j equation

When we solve the equations for the Reynolds stresses, we can use these instead of a turbulence
viscosity in the gradient model above, Eqs. (5.1)–(5.4).
The general-gradient flux model becomes
∂ϕ
−u ′j ϕ ′ = Cϕ · θ · u ′j u ′k , (5.8)
∂ xk
or as a diffusion model,
∂ ′ ′ ∂ϕ
Dϕ,t = Cϕ · θ · u j u k . (5.9)
∂x j ∂ xk
Here, θ is a time scale for the turbulence, for instance k/ε. Cϕ is a model constant. We can compare
the two diffusion models, Eq. (5.3) and (5.8), in a boundary layer with x 2 normal to the wall. Then
we find that the ratio between the diffusion coefficients is proportional with u ′22 /k. We can regard
the normal stress normal to the wall as more representative than k for the diffusion towards the wall.
Also when solving algebraic simplifications of the Reynolds stress equations (Sec. 6.5), we can use
Eq. (5.9) as the diffusion model for scalar variables.
5.4 Mixture fraction, conserved scalar

Definition
If the diffusion coefficient D is equal for all species, we can put up a transport equation for the
elemental mass fraction, Z i :

∂ ∂ ∂ ∂ Zi
(ρ Z i ) + (ρ Z i u j ) = ρD . (5.10)
The equation has no source term since the reaction does not affect the elements.
1
✲
mix
✲
✲
2
Figure 5.6: Mixing chamber; definition of mixture fraction
If we let two streams of fluid mix, we can define a mixture fraction. We can look at a mixing
chamber with two steady inflows: Inlet 1 gives ξ kg of fluid with the property ϕ1 and Inlet 2 gives
(1 − ξ ) kg fluid with the property ϕ2 . We then get 1 kg mixture, and this mixture gets the property
ϕmix = ξ ϕ1 + (1 − ξ )ϕ2 . This can be written as
ϕmix − ϕ2
ξ= . (5.11)
ϕ1 − ϕ2
This is the definition of the mixture fraction. All properties that have no source or sink, can be
written on this form. We call them conserved scalars and all get the same mixture fraction ξ .
The elemental mass fraction Z i is such a property, and also mass fractions of non-reacting species.
From Eqs. (1.18), (1.19) and (1.5), we can show that the functions (Yfu − 1r Yox ),
5.4 Mixture fraction, conserved scalar 85
1 r
(Yfu + 1+r Yprod ) and (Yox + 1+r Yprod ) have no source term. Hence, they are properties that satisfy
Eq. (5.11). See more on this in a combustion book, e.g. Kuo (1986:35f).
With certain simplifications, we can neglect the source term of the energy equation. Then, we
assume the Lewis number to be unity (i.e. Pr = Sc or α = D), neglect radiation heat and friction
heat (dissipation), and we neglect effects of pressure gradients. We can then write
h − h2
ξ= , (5.12)
h1 − h2
where h is the enthalpy of the mixture, and h 1 and h 2 are the enthalpies of the two inlets.
The idea is that now, we can put up an equation for the mixing fraction,

∂ ∂ ∂ ∂ξ
(ρξ ) + (ρξ u j ) = ρD . (5.13)
If we solve this equation, we can determine the various conserved scalars, for instance
h = ξ · h 1 + (1 − ξ ) · h 2 , (5.14)
1 1 1
(Yfu − Yox ) = ξ · (Yfu − Yox )1 + (1 − ξ ) · (Yfu − Yox )2 . (5.15)
r r r
Remark: I said above that with N species in the mixture, we need (N − 1) equations for mass
fractions in addition to the energy equation. If we account the reaction as three species (fuel, oxidizer,
product), then N = 3. If we regard enthalpy as a conserved scalar with the same diffusion coefficient,
it is sufficient to solve two equations for the four quantities. We couple them through a mixture
fraction.
Example: Mixing of fuel and air
A stream (1) of methane and a stream (2) of air are brought together. We imagine that the fuel does
not burn but is mixed with the air. With no reaction there is no source term, and all mass fractions
are conserved scalars. Equation (5.11) gives (Yk )mix = ξ · (Yk )1 + (1 − ξ ) · (Yk )2 . In stream (1),
Yfu = 1 and all other are zero; in stream (2), Yair = 1 and Yfu = 0. We can also split the air into
nitrogen (79% volume, 77% mass) and oxygen; YN2 = 0.77 · Yair and YO2 = 0.23 · Yair . Figure 5.7(a)
shows the composition as a function of mixture fraction.
Now we let the fuel react with air. We imagine (as an approximation) that the reaction is immediate
and complete. A location can then have either fuel or air (oxygen), but not both. At stoichiomet-
ric mixing neither air nor fuel are present, only product, Yprod = 1, and the conserved scalar is
(Yfu − r1 Yox ) = 0. The mixture is stochiometric at ξ = ξst :
(Yfu − 1r Yox )st − (Yfu − r1 Yox )2 0 − (−r −1 ) 1

ξst = = = . (5.16)
(Yfu − 1r Yox )1 − (Yfu − r1 Yox )2 (1) − (−r −1 ) r +1
For methane the stochiometric mass of air is r = 17.16, thus ξst = 0.055.
1 1
Yprod
Yair
YN2 Yfu YN2 Yfu
Yair
YO2
YO2 YCO2
YH2 O
0 0
0 ξ 1 0 ξ 1
(a) Mixing without reaction (b) Combustion
Figure 5.7: Composition as a function of mixture fraction. Mixing of fuel (methane) from
Inlet (1) and air from Inlet (2).
For ξ < ξst is Yfu = 0 and for ξ > ξst is Yair = 0. From the conserved scalars (Yfu − r1 Yox ) and
1
(Yfu + 1+r Yprod ), we can resolve
Yair = 1 − (1 + r )ξ for ξ ≤ ξst , (5.17)
1
Yfu = ((1 + r )ξ − 1) for ξ ≥ ξst , (5.18)
r
Yprod = (1 + r ) (ξ − Yfu ) . (5.19)
When methane reacts with air, the product will contain 15% CO2 , 12% H2 O and 73% N2 by mass.
We can write YCO2 = 0.15 · Yprod and YH2 O = 0.12 · Yprod . Nitrogen exists in both air and product
so that YN2 = 0.77 · Yair + 0.73 · Yprod , while YO2 = 0.23 · Yair . Figure 5.7(b) shows the composition
as a function of mixture fraction.
The enthalpy and temperature can also be calculated as functions of mixture fraction. For Inlet (1),
we have h 1 = h f,fu + 1h 1 , where h f is the enthalpy of formation and 1h is the thermal enthalpy, cf.
Sec. A.3.1. For the air, Inlet (2), h 2 = 1h 2 . From Eq. (5.14), we have h = ξ(h f,fu + 1h 1 ) + (1 −
ξ )1h 2 , that is, a straight
P line in a h-ξ diagram.
P This enthalpy for the mixture of air and fuel we also
can write as h = k Yk h k = Yfu (h f,fu + k Yk 1h k . Here, we can introduce the expression for Yfu
and other mass fractions as functions of the mixture fraction. From these expressions and enthalpy
tables, we can determine the temperature. As a simplification, we can assume constant specific heats
and resolve the temperature as a function of the mixture fraction. We will get a maximum value at
stoichiometric mixing. From T1 and T2 of the inlets, the temperature will be linear up to the peak at
ξ = ξst .
Also the density can be calculated as a function of the mixture fraction. First then, we have to say
something more about the flow, e.g. specify the pressure. Then we can use the equation of state and
the temperature just found. The function ρ = ρ(ξ ) is not linear in general.
5.5 Scalar variance 87
Turbulence equation
The Reynolds equation for the mixture fraction becomes

!
∂ ∂ ∂ ∂ξ
(ρξ ) + (ρξ u j ) = ρD − ρξ ′ u ′j . (5.20)
For the Reynolds flux, we can use a gradient model:
νt ∂ξ
−u ′j ξ ′ = , (5.21)
σξ ∂ x j
where σξ is a model constant (σξ = σh = σY ).
The equation for mixture fraction is used in some combustion models. We will return to this in
Chapter 10.
5.5 Scalar variance

A variance is a measure of the deviation from an average value: ϕ ′ 2 = (ϕ − ϕ)2 . The square root
of the variance is the standard deviation or the root mean square (“rms”). In many occasions, this is
more convenient to use since it has the dimension of the quantity itself.
Normal stresses are variances of the velocity components. The equation for the variance of a scalar
is not very different from the equation for turbulence energy, and the development follows the same
pattern:
∂ ∂ ∂ ′ (5.22)
∂t (ρϕ) + · · · = · · · − ∂t (ρϕ) + · · · = · · · = ∂t (ρϕ ) + · · · = · · · .
Multiply the ϕ ′ -equation with 2ϕ ′ , average the equation and obtain an equation for the variance ϕ ′ 2 :
average
2ϕ ′ · ∂t∂ (ρϕ ′ ) + · · · = · · · = ∂t∂ (ρϕ ′ 2 ) + · · · = · · · −→ ∂t∂ (ρϕ ′ 2 ) + · · · = · · · . (5.23)
For the temperature variance, the result with constant density and constant specific heat capacity
becomes
!
∂ ′ 2 ∂ ′2 ∂ λ ∂T ′2 ′ ′ 2
ρT + ρT u j = − ρu j T
|∂t {z } |∂ x j {z } |
∂x j Cp ∂x j
{z }
transient convection
term molecular turbulence
diffusion
∂T λ ∂T ′ ∂T ′
− 2ρu ′j T ′ − 2 + 2ρT ′ ST′ . (5.24)
∂x j Cp ∂x j ∂x j | {z }
| {z } | {z }
production destruction
For flows without reaction, the last term can often be canceled. (It does not have its own name,
except that it is a correlation.) The diffusion of Eq. (5.24) can be modeled by a gradient model. The
turbulence flux in the production term has to be modeled as described earlier in this chapter. The
modeling of the destruction term we will return to later in Sec. 8.5, page 131.
The equation for mixture fraction, Eq. (5.13), has no source term, and the equation for its variance
ξ ′ 2 looks like the T ′ 2 equation above, although without the last term. Such an equation is used in
some combustion models, Chapter 10.
The variance for mass fraction can be expressed with a similar equation. If the species takes part in
chemical reactions, also this equation has a source term, Rk′ Yk′ , which is hard to model.
The variance can, for instance, be used to determine the probability of a concentration to be over (or
under) a certain limit. This can be the lowest concentration that can give ignition in air (flammability
limit), a limit for a poisonous substance or a limit specified by agreement for the concentration of a
species in a gas mixture (e.g. natural gas) for sale.
The mean value we determine by solving the equation for the mean mass fraction. If we in addition
resolve the variance of the concentration and presume a distribution function (e.g. normal distribu-
tion), we can say something of how often the concentration comes over (or under) a certain limit.
When this is said, we also have to include a warning: Turbulence modeling gives – as we have
seen previously – no exact or absolute answers. Therefore, one has to use this with great care – in
particular in cases with potential harm for humans or the environment.
5.6 Thermal boundary layer – wall functions

This section has to be read in relation to section 4.4 on wall functions for
momentum (velocity). Also here, we will primarily look a the wall functions that
we use as boundary conditions for turbulence models.
Wall boundary conditions for temperature and heat transfer can roughly be divided into three groups:
– wall with heat transfer through the wall to/from the fluid.
– isolated wall with surface temperature equal to the fluid temperature; no heat transfer.
– isolated wall with surface temperature that is equal to neither the fluid temperature nor the temper-
ature inside the wall. Radiation towards a wall gives such a case. The heat transfer through the wall
is then zero, although there is still a heat transfer from the surface to the fluid. This heat transfer
balances the net radiation heat transfer.
In the second case, it does not matter much which expression we use, except that it gives zero tem-
perature gradient and zero heat transfer. In the other two cases, it is important to find an expression
for the temperature and the temperature gradient. In the third case it might be that the surface tem-
perature is primarily determined by the radiation heat balance; the wall function is then of somewhat
less importance.
Thermal boundary layers are treated in many books on heat and mass transfer. Kays and Craw-
5.6 Thermal boundary layer – wall functions 89
ford (1993:269f) give a clear introduction. Kader (1991) gives a good presentation of temperature
functions for boundary layers under different conditions. Bradshaw and Huang (1995) discuss wall
functions for velocity and temperature, and provide some critical viewpoints both on their use as
models, and as a basis for developing and calibrating turbulence models. Launder (1988) summa-
rizes different approaches to determine conductive heat transfer in complex turbulent flows.
In this section, we will look at wall functions for temperature. If one wants wall functions for
enthalpy, these can be developed in the same manner. The relations between temperature and various
enthalpies are shown in Sec. A.3. In Sec. 4.4, we looked at wall functions for velocity (momentum).
That is, relations that give a velocity profile in the region near the wall. Very close to the wall, the
profile is linear, and a little further out (but still close the the wall) a logarithmic profile, Eq. (4.11).
In the same way, the temperature or heat transfer can be made dimensionless and the temperature
profile can be expressed by
T + = σ x 2+ (5.25)
quite near the wall, and a little further out (still near the wall) by
1 1
T+ = ln x 2+ + DT = ln (E T x 2+ ), (5.26)
κT κT
or also as
T + = σT (u +
1 + P). (5.27)
Here is κT = κ/σT and P = κ1 ln (E T /E) is a function of the molecular Prandtl number σ . P

is actually also a function of the turbulence Prandtl number σT but this is regarded constant in the
context. For the function P, several expressions exist, all empirical. The most widely used seems to
be a formula by Jayatillaka
3/4 !
σ σ
P = 9.24 −1 1 + 0.28 exp −0.007 , (5.28)
σT σT
which is said to be based on a large collection of experiments. (Here after Patankar and Spalding
(1967:70) who refer to a Russian conference paper by Jayatillaka from 1964.)
The expressions above can, at least to some extent, be explained by a boundary-layer analysis of
the heat transfer: If there is no source term in the energy equation (temperature equation) and we
can neglect transport with the mean flow, the heat transfer in the x 2 -direction will be approximately
constant within some distance from the wall.
We then model the heat transfer as

ν νt dT
q2 = −ρC p + = qw . (5.29)
σ σT d x2
Here, we have a conductive or diffusive sublayer where the molecular transport dominates (cf. vis-
cous sublayer in Sec. 4.4), and a layer where the turbulent transport dominates. We let the distance
x 2 = x 2,d be the limit between the two layers.
From Eq. (4.14), we use the expression νt = κu τ x 2 and rearrange the equation to
ρC p u τ d T d x 2+
− = . (5.30)
qw 1 κ +
+ x
σ σT 2
This equation we integrate from the wall to some location in the boundary layer. On the left-hand
side, we obtain a dimensionless quantity that we define as T + ,
Z T
ρC p u τ d T ρC p (Tw − T )u τ
T+ = − = . (5.31)
Tw qw qw
The integral of the right-hand side of Eq. 5.30 has to be split in two, one part in the diffusive sublayer
and one part outside it. Within the sublayer (x 2+ < x 2,d +
) we neglect turbulence transport, while
outside, we neglect the molecular transport:
Z x+ Z x+ +
2,d 2 σT d x
+ 2
r.h.s. = σ d x2 + . (5.32)
0 +
x 2,d κ x 2+
+
The integral gives Eq. (5.26), where DT (or E T ) becomes a function of σ , σT and κ (and of x 2,d ,
which is also a function of these quantities).
If we integrate to a certain location inside the diffusive sublayer (x 2 < x 2,d ), we have only the first
part of the integral of Eq. (5.32) and we get the linear expression of Eq. (5.25).
It can be worth noting the presumptions for this development:
– steady flow of constant density (often said “incompressible”);
– no source terms due to reactions, pressure work, etc.;
– heat capacity, viscosity and diffusivity do not vary;
– that the Reynolds analogy (page 81) is approximately valid; and
– that the turbulence Prandtl number is constant through the entire boundary layer.
In particular, the two first items are doubtful for a flow with combustion. That is, the question is not
whether they are right – they are certainly not. The question is whether it matters in practice that
we do these approximations, that is, how good the approximations are. Han and Reitz (1997) give
a development without these presumptions. With the ideal-gas equation of state, ρ = p/(RT ), (but
still with constant C p ), the integration of Eq. (5.31) gives
ρC p u τ T ln (Tw /T )
T+ = . (5.33)
qw
The right-hand side of Eq. (5.26) will have an additional term that is a function of the reaction heat.
Pragmatism here as well (cf. page 74)
When we stretch the domain of utilization for wall functions – and so we do – we can get difficulties
with, for instance, stagnation points where the shear stress is zero. Above, page 75, we saw how
5.6 Thermal boundary layer – wall functions 91
√
this problem is overcome by introducing k in place of u τ in some expressions for momentum wall
functions, see Eqs. (4.22)–(4.26). For thermal boundary layers, the problem becomes even bigger,
and we use the same trick here. Equation (5.31) and (5.26) can be reformulated to
ρC p (Tw − T )k 1/2 σT σT
T∗ ≡ = ∗ ln x 2∗ + DT∗ = ∗ ln (E T∗ x 2∗ ). (5.34)
qw κ κ
The expression on the left-hand side defines T ∗ . Then we have also for the linear layer
T ∗ = σ x 2∗ . (5.35)
The distance x 2∗ is defined equal to x 2 k 1/2/ν, Eq. (4.25). The integration function DT∗ or E T∗ is found
by setting Eqs. (5.34) and (5.35) equal at the distance x 2 = x 2,c . (Notice that x 2,c do not need to be
equal to the corresponding distance x 2,v in the momentum boundary layer, Eq. (4.26).) This gives
1
E T∗ = ∗ ∗
∗ exp κ x 2,c σ/σT . (5.36)
x 2,c
Here, as previously, we can express the temperature wall function by the velocity wall function:
T ∗ = σT (u ∗1 + P ∗ ), (5.37)
1
where P ∗ = κ∗ ln (E T∗ /E ∗ ).
Since the (∗)-laws should give the same results as the (+)-laws at local equilibrium and constant
turbulence energy, E T∗ /E ∗ = E T /E, and therefore must P ∗ κ ∗ = Pκ, or P ∗ = (Cµ )−1/4 P.
Effects of pressure gradient
The effects of a pressure gradient on wall functions are discussed by, among others, Kays and Craw-
ford (1993:274) and Kader (1991).
For temperature it applies as for velocity:
– Acceleration, i.e. forward pressure (∂ p/∂ x 1 < 0), gives a thinner wake layer (the outer region)
and, hence, a thicker log-layer. The wall function (the log-law) is valid farther out from the
wall.
– Retarding, i.e. adverse pressure (∂ p/∂ x 1 > 0), enhances the wake layer and the log-layer
becomes thinner. If the pressure gradient is sufficiently large, the log-layer may disappear.
An important difference is that these laws do not follow the same line as for zero pressure gradient.
Acceleration gives a steeper log-curve, whereas retardation gives a less steep curve.
5.7 Wall function for mass transfer

This section follows the preceeding. Wall functions for mass transport seems
not to be much discussed in literature. We can develop them in the same way
as wall functions for heat transfer. As long as we use equal turbulence Prandtl
and Schmidt numbers, the wall functions ought to look quite similar.
In most cases, we do not need wall functions for the concentration equations. It is sufficient to set
the gradient and the mass transfer (net flux of a species) to zero. If the net flux is non-zero, we
need a better boundary condition. One example is a substance that is added through a porous wall.
Another example is when chemical reactions occur on a solid surface – for instance in a catalyzer,
or oxidation of solid carbon (char, graphite). We then get a gradient in the concentration and a net
flux of different species.
Wall functions for concentration can be set up in the same way as wall functions for temperature.
We assume the net flux of species k to be constant some distance outward from the wall, jk,2 = jk,w .
Then we apply the other presumptions as in the development of the temperature wall function. We
get a dimensionless mass fraction,
ρ(Yk,w − Y k )u τ
Yk+ = . (5.38)
jk,w
Near-most the wall, we get a linear relation in a diffusive sublayer, like the viscous and conductive
sublayers. For the region where turbulent transport dominates, we get a logarithmic expression,
1 1
Yk+ = ln x 2+ + DY = ln (E Y x 2+ ) = σY (u +
1 + PY ). (5.39)
κY κY
Here, σY is the turbulence Schmidt number and the other parameters correspond to those of the
temperature wall function. The constants in the functions have to be found from experimental data
– and there will surely be a shortage on such data. It may be that you get away by using the same
constants and functions as in the temperature wall functions, only replacing the Prandtl number with
the Schmidt number. If doing this, you can argue that the same motions are making the boundary
layers for both temperature and concentration.
Chapter 6
Equations for the Reynolds stresses

In preceeding chapters, we have looked at simpler, algebraic expressions for the Reynolds
stresses. In this chapter, we will see how to make model equations for the Reynolds
stresses.
6.1 Models for Reynolds stresses – a summary

In Chapter 2 we saw that several new variables appeared in the turbulence equations (Reynolds
equations); among these the Reynolds stresses u ′i u ′j . Viscous stresses are modeled as the product of
a viscosity and a velocity gradient, Eq. (2.12) page 39 (or more generally in Eq. (A.25), page 201).
An early and much used attempt is to model the Reynolds stresses in the same way, Eq. (2.13) or
Eq. (2.14). Mathematically seen, the models looks the same. The important difference is that the
molecular viscosity µ is a substance or material property of the fluid itself, whereas the turbulence
viscosity µt is a function of the flow.
Above, we looked at how the turbulence viscosity can be expressed from quantities of the flow. First
by mean quantities and external measures of the flow, and then (Chapter 3) by turbulence quantities.
With transport equations for such quantities, e.g. k and ε, we can include effects from other parts of
the flow and changes over time.
The question remaining is whether the relation between Reynolds stress, turbulence viscosity and
velocity gradient is the same as in the model for the viscous stresses. Or more correctly: The question
is whether this is a sufficiently good approximation. In some instances the model is good enough,
in other cases it is not. Equation (2.14) gives a linear relation between stress and rate of strain. The
limitation of this model lies in the presumption that one and the same turbulence quantity, µt , should
represent the stresses in different directions. The turbulence viscosity that gives a good result for
u ′1 u ′2 is not necessarily the same that give good results for u ′12 or u ′2 2 . This we have been looking at
in Sec. 3.6 (page 58).
94 6 Equations for the Reynolds stresses
r/d r/d
7 7
6 6
5 5
4 4
3 3
2 2
1 1
0 0
0 5 10 15 x/d 0 5 10 15 x/d
(a) Isothermal mixing, k-ε model (b) Isothermal mixing, Reynolds stress equations
r/d r/d
7 7
6 6
5 5
4 4
3 3
2 2
1 1
0 0
0 5 10 15 x/d 0 5 10 15 x/d
(c) Combustion, k-ε model (d) Combustion, Reynolds stress equations

r/d r/d
7 7
6 T (K) 6 T (K)
400
5 5
500
4 4
600
3 3
700 800
2 400 2 1100
500 800 1100 1400 1700 1400 1700
1 1
0 0
0 5 10 15 x/d 0 5 10 15 x/d
(e) Combustion, k-ε model (f) Combustion, Reynolds stress equations
Figure 6.1: Recirculation zone behid an axisymmetric body with a centered fuel jet (r <
d/2) with velocity 80 m/s and air outside (r > 6d) with velocity 6.5 m/s. Computations for
isothermal mixing without reaction and for combustion with Magnussen’s combustion model
(EDC) and fast reactions (see Chapter 11). Both cases are computed with a k-ε model and
with Reynolds stress equations. The streamlines, (a)-(d), have the same values in all cases.
We see that the recirculation zone has about the same length with both turbulence models, but
that the k-ε model gives a stronger circulating motion (denser and more streamlines). Weaker
circulation gives mass and energy from the fuel jet and the flame (see temperature plots) a
possibility to intermingle (diffuse) more with the co-flowing air, without being blown away.
This is particularly apparent in the temperature field close to the body. From Gran, Ertesvåg
and Magnussen (1997), see also Gran (1994:87f).
6.2 Development of the equation 95
For two-dimensional boundary layers, only the streamwise velocity u 1 is important. In its equation,
only the shear stress u ′1 u ′2 has to be included, see Eq. (4.7). In a more complex two-dimensional
case, e.g. a recirculation zone, both u 1 and u 2 are important. We have to include terms with u ′1 u ′2 ,
u ′12 and u ′22 in the equations of motion. If we, in additions, let the flow move in the third direction
as well, we can have six important Reynolds stresses. With one and the same turbulence viscosity
for all, some stresses easily get too large and some get too small.
The alternative to such model expressions can be to solve equations for the Reynolds stresses as
well. This is what we will look at in this chapter. Figure 6.1 shows a comparison between k-ε
and Reynolds stress equations. Without combustion, the differences are small. With combustion,
however, there are considerable deviations between the two models.
6.2 Development of the equation

In the same way as the equation for turbulence energy, k, the equation for the Reynolds stresses can
be developed from the basic equations. We use the relation
∂ ′ ′ ∂u ′ ∂u ′j
u i u j = u ′j i + u ′i . (6.1)
∂t ∂t ∂t
For a flow with constant density, the result becomes

!
∂ ′ ′ ∂ ′ ′ ∂u j ∂u i ∂ ∂u ′i u ′j
ρu i u j + ρu i u j u k = − ρu ′i u ′k + ρu ′j u ′k + µ
∂t ∂ xk ∂ xk ∂ xk ∂ xk ∂ xk
| {z } | {z } | {z }
ρCi j ρ Pi j ρ Di j,v
∂
+ −ρu ′i u ′j u ′k − p′ u ′i δ j k + p′ u ′j δik
∂x
| k {z }
ρ Di j,t
∂u ′ ∂u ′j ′
∂u ′ ∂u j
+ p′ ( i + ) − 2µ i , (6.2)
∂x j ∂ xi ∂x ∂x
| {z } | {zk k}
ρ8i j ρεi j
where Ci j is the transient term and the convective transport with the mean flow; Pi j is production,
or work conducted by the Reynolds stresses, transfer of mechanical energy from the mean flow to
the turbulence; Di j,v and Di j,t are viscous diffusion and turbulence diffusion; 8i j is redistribution,
exchange of energy between the components; and εi j is dissipation, transfer of kinetic energy to
thermal energy.
We can sum the three equation for the normal stresses u ′1 2 , u ′2 2 and u ′32 , and divide by two. We then
get the equation for turbulence energy k = 21 u ′i u ′i (see Sec. 3.2, page 48). We can regard the normal
stresses as components of the turbulence energy.
Some authors would like to restrict the terms redistribution and dissipation to i = j , that is, to
normal stresses or components of turbulence energy. When i 6= j , they may use the term loss of
correlation for 8i j and εi j .
6.3 The terms of the equation

The normal stresses can be regarded as components of the turbulence energy. The different equations
can thus be regarded as energy transfer between these energy components and other forms of energy,
Fig. 6.2.
Production
We have put the label “production” on one of the terms. How do we know this? As said, we
regard the normal stresses as components of the turbulence energy. This energy have to come from
somewhere, and it can only come from the mean flow. This means that the term providing the energy
transfer from the mean flow to the turbulence has to be some function of u i or its derivative. As Ci j
is a convective term, Pi j is the only term left that can be the production term. In the equation for
turbulence energy, the production term is Pk = 21 Pii .
Production comes from Latin “producere” which means bring or lead forth/forward, see Sec. D.
It can be shown that Pi j has the same sign as u ′i u ′j . Then, Pii have to be positive. In an equation for
kinetic energy of the mean flow, ( 12 u i u i ), we will get a term equal to − 21 Pii , which is a sink term.
Viscous diffusion
Here, the viscous contributions are split in two. The gradient term Di j,v we call viscous diffusion.
Here, as in the discussion of the k equation (page 49), a smaller part of the dissipation term has
actually been included to give a gradient term.
Dissipation
In a viscous fluid, viscous forces will act against the motion, that is, friction forces. Heat is released;
that is, mechanical energy is transferred to thermal energy. This means that εi j is a sink term.
Dissipation, dissipate comes from Latin and means to spread abroad, scatter, disperse, see Sec. D.1.
In order to compare the processes in production and dissipation, the terms can be written as
∂u
Prod = τt , where τt = −ρu ′i u ′j ; (6.3)
∂x
6.3 The terms of the equation 97
Mean flow
P22
T22
Mean flow
P11 822
−811 ε22
T11
Mean flow
ε11
833
P33
T22
ε33
Figure 6.2: Energy flow for Reynolds stresses (normal stresses) in a general three-
dimensional case
∂u ′ ∂u ′
Diss = τv′ , where τv′ = µ . (6.4)
∂x ∂x
Although the terms are not written completely, this shows that they have the same form: Stress times
velocity gradient.
Production is work done by turbulence stresses against the mean flow, that is, against large-scale
motions. Dissipation is work done by viscous stresses against the turbulence, that is, against small-
scale motions.
The turbulence energy is drained from the mean flow (large scales) and transferred downwards in
scale levels from larger to smaller and smaller turbulence eddies. It is lost in frictional work (heat),
mainly in the smallest eddies. (More on this in Chapter 8.)
Redistribution
There is a term in Eq. (6.2), 8i j , that disappears when we summarize the equations for the normal
stresses. For a flow with constant density, 8ii = 811 + 822 + 833 = 0. This means that the term
expresses neither supply nor loss of energy. It then has to be the transfer from one component to
another, for instance from u ′12 to u ′22 . We call this re-distribution.
x2
xη
uE
xξ
ϕ
x1
Figure 6.3: Example of different coordinate systems
Turbulence diffusion
There are still two more terms in Eq. (6.2). They are similar to each other and are gathered under the
same label, Di j,t . They express stirring/mixing due to turbulence and are called turbulence diffusion.
Shear stresses
What is said so far is valid for the normal stresses (i = j ), which are components of turbulence
energy. For the shear stresses (i 6= j ), we can not talk about energy, only about correlation between
u ′i and u ′j . Nevertheless, we use the same names and labels on the terms:
P12 is production of correlation between u ′1 and u ′2 ;
ε12 is viscous loss of correlation between u ′1 and u ′2 ; and
812 is loss of correlation between u ′1 and u ′2 due to the turbulence.
The convective and diffusive terms are transport terms as for the normal stresses.
There are no fundamental differences between normal stresses and shear stresses. In a given case,
the magnitude and distribution of normal and shear stresses depend on the coordinate directions.
We can show this by using the two coordinate systems in Fig. 6.3. One system has the axes (x 1 ,x 2 )
while the other has the axes (x ξ ,x η ). A vector uE is independent of the coordinate system,. In the first
system the vector has the components (u 1 ,u 2 ). These components can be expresses as functions of
6.4 Modeling of the equation 99
the components (u ξ ,u η ) in the other system:

u 1 = u ξ cos ϕ − u η sin ϕ, (6.5)
u 2 = u ξ sin ϕ + u η cos ϕ, (6.6)
u 21 = u 2ξ cos2 ϕ + u 2η sin2 ϕ − 2u ξ u η cos ϕ sin ϕ, (6.7)
u 22 = u 2ξ sin2 ϕ + u 2η cos2 ϕ + 2u ξ u η cos ϕ sin ϕ, (6.8)
u1u2 = (u 2ξ − u 2η ) 12 sin 2ϕ + 2u ξ u η cos 2ϕ. (6.9)
Using these relations for the components of the velocity fluctuations and averaging the expressions,
we see that a normal stress u ′12 in one coordinate system is a function involving both the normal
stresses u ′ξ 2 , u ′η2 , and the shear stress u ′ξ u ′η in the other coordinate system. This discussion can also
be made for a three-dimensional case.
A model has to be independent of the orientation of the coordinate system. This means that the terms
have to be modeled with the same expressions in the equations for normal and shear stresses. This
requires, for instance, that the dissipation model is such that the sum εii equals 2ε, independently of
the coordinate system. Correspondingly, for constant-density flows, the sum of the redistributions,
8ii , has to be zero, regardless of the choice of coordinate system.
6.4 Modeling of the equation

In Eq. (6.2) we have to model the turbulence diffusion Di j,t , the redistribution 8i j and the dissipation
εi j . The convection Ci j , the production Pi j and the viscous diffusion Di j,v are functions of known
quantities and do not require modeling. In addition, we need a length scale or timescale for the
turbulence. The usual way to provide this, is to solve an equation for the dissipation of turbulence
energy, ε.
Turbulence diffusion
The most common model, at least in the literature, is a gradient model after Daly and Harlow (1970):
′ ′
!
∂ k ′ ′ ∂u i u j
Di j,t = Cs u k u ℓ . (6.10)
∂ xk ε ∂ xℓ
The model constant Cs is often set to 0.22. Here, k/ε is used as the timescale. As in the k-equation,
Eqs. (3.6)–(3.7), there are two interpretations of this model:
1) as a model for the entire term Di j,t ,
2) as a model for the triple-correlation gradient; the pressure term is then neglected.
In practice, the two interpretations give identical expressions.
The expression looks much like the general gradient model for diffusion, Dϕ,t , Eq. (5.9), but does
not quite fit into the pattern.
The simplest model is just to use a turbulence viscosity:

!
∂ k ∂u ′i u ′j
Di j,t = Cµ′ k . (6.11)
∂ xk ε ∂ xk
In order to be in correspondence with the re-distribution model (see below), the constant Cµ′ has to
be a function of the constants of the redistribution model. This means that Cµ′ is not required to equal
Cµ in the k-ε model. This model is much simpler to code than the one in Eq. (6.10). Moreover, it
has showed to be more stable numerically in some cases.
A third model is proposed by Hanjalić and Launder (1972). It is called a “kvasi-Gaussian” model.
This has shown to be better in some cases, poorer in some other cases. In practice, there seems to be
little difference, except that the gradient model is simpler. More proposals can be found in review
articles (which there are several, see later in this section).
Dissipation
Using an isotropic model for the dissipation tensor is the most common. This means that the dissi-
pation is equal for all three normal stresses:
εi j = 32 εδi j . (6.12)
The model presumes high Reynolds numbers, however, not that the normal stresses are equal (isotropic).
The rationale for the model is (Rotta, 1951a; Kolmogorov, 1941a) as follows: The dissipation
occurs in small scales. The smallest eddies are independent of the large eddies and of the main
flow direction. For the small eddies, all directions are equally probable. The dissipation then
has to be distributed evenly between the three energy components, u ′12 , u ′2 2 and u ′3 2 . Hence,
ε11 = ε22 = ε33 = 23 ε. If εi j 6= 0 for (i 6= j ), the model would not be directionally indepen-
dent, therefore εi j = 0 for (i 6= j ).
This reasoning can be discussed, but the practical consequences are small (see Ertesvåg, 1991:14).
All deviations from the isotropic model can be put into the redistribution model.
Re-distribution
First, we consider a two-dimensional boundary layer: The equations for u ′1 u ′3 and u ′2 u ′3 cancel,
and u 3 = 0 (the turbulence is, however, three-dimensional so that u ′3 6= 0). The dominating
gradient for the mean velocity is ∂u 1 /∂ x 2 . The shear stress is (−u ′1 u ′2 ) > 0. Furthermore, we have
P22 = P33 = 0, whereas ε22 6= 0 and ε33 6= 0. This means that u ′1 2 is the only component receiving
energy from the mean flow. The other components, u ′2 2 and u ′32 should then die unless they receive
energy through other terms. The two normal stresses are of the same magnitude as u ′1 2 , and the
energy has to come through the redistribution.
6.4 Modeling of the equation 101
For u ′1 u ′2 , the production term is P12 = −u ′22 (∂u 1 /∂ x 2 ). The sign of P12 is the same as for
u ′1 u ′2 . This means that |u ′1 u ′2 | is produced (increases due to P12 ). For high Reynolds numbers, the
dissipation term ε12 = 0. The redistribution then has to diminish the correlation u ′1 u ′2 , that is, to
reduce u ′1 u ′2 . Otherwise the shear stress would go to infinity.
Rotta (1951a) imagines “turbulence balls” that clash together. He deduces that the net released
energy from one component (normal stress) has to be

−8ii ∼ 23 u ′i 2 − 32 k [no sum over i ] . (6.13)
√
In order to get the correct dimension, he has to multiply this by k/L:
√
k ′2 2
8ii = −C1 ui − 3 k [no sum over i ] . (6.14)
L
We see that if u ′i 2 > 23 k, 8ii < 0 , and that if u ′i 2 < 32 k, then 8ii > 0.
The term take from those that have much and gives to those that have little (a “Robin Hood term”,
if you like). Said in another way: The term evens the energy between components, or isotropify the
turbulence.
For shear stresses, 8i j has to degrade the correlation. Is also has to comply with the requirement
of being independent of the directional orientation of the coordinate system (see above). Rotta uses
this to show that with Eq. (6.14),
√
k ′ ′
8i j = −C1 uu when i 6= j. (6.15)
L i j
the two expressions of Eqs. (6.14) and (6.15) both can be written as
√
k ′ ′
8i j = −C1 u i u j − 32 kδi j . (6.16)
L
Rotta formulated an equation for the length scale L. We can also use L = k 3/2/ε. From experimental
data, Rotta (1962:48) found C1 = 2.8.
The term of Eq. (6.16) is often denoted as 8i j,1 . This is because more terms are required to model
the redistribution. The redistribution term of Eq. (6.2) can be expressed with an integral (Chou,
1945; Rotta, 1951a):
Z
8i j = [ (turbulence) + (mean-velocity gradient) · (turbulence) ]. (6.17)
vol | {z } | {z }
(i) (ii)
The first term, from (i), has to be a function of turbulence quantities, like the Rotta term 8i j,1 above.
The second term, from (ii), has to be a function of both mean-velocity quantities and turbulence
quantities. The simplest model is made by Naot, Shavit and Wolfshtein (1970):

8i j,2 = −C2 Pi j − 31 Pkk δi j . (6.18)
This term is called the rapid term because the model constant C2 was determined by rapid distortion.
That is, isotropic turbulence where a strain rate ∂u 1 /∂ x 2 suddenly is imposed. In this case, the
turbulence equations becomes very simple and can be solved analytically. This gives C2 = 0.6.
The model was originally proposed as an alternative to Rotta’s model. It has appeared, however, that
the sum of these two models gives reasonably good results, thus
ε
8i j = −C1 u ′i u ′j − 23 kδi j − C2 Pi j − 31 Pkk δi j . (6.19)
k
The combination of the two constants C1 = 2.8 and C2 = 0.6 appears to work poorly. Eventually,
two sets of constants have gained status as a kind of standard:
– Gibson and Launder (1978): C1 = 1.8 and C2 = 0.6;
– Gibson and Younis (1986): C1 = 3.0 and C2 = 0.3.
When the flow is adjacent to solid boundaries, the 8i j -model requires an addition that damps the
fluctuations normal to the wall. This term is caused by reflection of pressure fluctuations from the
wall. This is not a “near-wall” or “low-Reynolds-number” term, but has effect far into the flow from
the wall, also outside the boundary layer. The term has to be included in all internal flows.
The redistribution model is regarded as the most uncertain part of the model equation for the Reynolds
stresses. New and better models are in development (see Launder, 1989a,b; Hanjalić, 1994; Speziale,
1998). It seems however, to be far ahead to a “final” model, in particular with combustion in mind.
Jones (1994) discusses the development of turbulence models with respect to combustion.
Now we have modeled all terms of the u ′i u ′j equation. We also need (at least) one length scale or
timescale for the turbulence. The most common is to solve an equation for the dissipation ε.
Launder (1989b) presents a complete version of a model with equations for the Reynolds stresses
and equations for the Reynolds fluxes. This is a model that is established as a model of use for
practical applications.
Equation for dissipation
We can use the same ε-equation as in the k-ε model. For the diffusion term, there is an alternative:
When the u ′i u ′j equation is solved, we can use the diffusion model of Eq. (5.9),

∂ k ′ ′ ∂ε
Dε,t = Cε u j u k . (6.20)
∂x j ε ∂ xk
The model constant Cε is often set to 0.18, although the value show some variation in the literature.
The other two constants in the equation are as in the k-ε model: Cε1 = 1.44 and Cε2 = 1.92.
Boundary values
What is said about boundary values for k and ε at page 55, is to a large extent valid for models with
equations for the Reynolds stresses (and ε) as well.
6.5 Algebraic simplification of the equation 103
At a symmetry boundary, the normal stresses will be symmetric similar to the k and the ε. The shear
stresses, however, will be anti-symmetric (symmetric in absolute value, but with opposite sign).
Hence, they have to be zero at the boundary, and their gradients can (but do not have to) have a finite
value.
At solid boundaries, wall functions can be used as for a k-ε model. The theory is the same as
shown in Sec. 4.4. The normal stresses can be resolved like it is done with the turbulence energy k
(page 76), whereas the shear stresses have to be treated in a particular way at the wall. How this is
done, is described in the literature.
Notice: For practical purposes it is important to note the following:
Those who develop wall functions and numerical techniques can locate the near-wall cell (near-
wall node) in the grid as they find it to be the best. In industrial computations, it is not always
straightforward to do this. Some investigations indicate that with a coarser grid near the wall, the
results for a Reynolds-stress-equation model are worsened considerably more than those of a k-ε
model. The equations and wall laws for the Reynolds stresses require a finer near-wall grid than for
the k and ε. Said in another way: Reynolds stresses and wall laws give poorer results for a finer grid
compared to a k-ε model with wall laws. This is important when you are evaluating which model to
use in a case where practical considerations only allow a coarse grid near the wall. Such a practical
issue may for instance simply be that you have a large number of solid surfaces in your calculation
domain. Because of this, there is a risk of bad/poor usage of such models.
6.5 Algebraic simplification of the equation

This section is on so-called algebraic Reynolds stress models (or algebraic
stress models). The idea is to simplify the partial differential equations for the
Reynolds stresses to algebraic equations, hoping that they will be easier to
solve.
Equation (6.2) for the Reynolds stresses can be written on the form
Ci j − Di j,v − Di j,t = Pi j + 8i j − εi j , (6.21)
| {z }
Ti j
where Ti j is transport of u ′i u ′j .
The equation for turbulence energy can be written on the same form,
Ck − Dk,v − Dk,t = Pk − ε, (6.22)
| {z }
Tk
where Tk is the transport of turbulence energy k.
Rodi (1976) simplifies the set of equations by modeling the transport of Reynolds stresses by
u ′i u ′j
Ti j = · Tk . (6.23)
k
He introduces Tk from Eq. (6.22)

u ′i u ′j ε
Ti j = · 1− Pk , (6.24)
k Pk
and then this into Ti j in Eq. (6.21)

u ′i u ′j ε Pi j 8i j εi j
1− = + − . (6.25)
k Pk Pk Pk Pk
With models for 8i j and εi j , this can be rearranged to algebraic functions of the Reynolds stresses:
∂u p
u ′i u ′j = F ( u ′k u ′ℓ , k, ε, ). (6.26)
∂ xq
The symbol F here represents some algebraic function. In addition, we have to solve equations for
k and ε. We can then use the diffusion model in Eq. (5.9) both for k and ε.
These simplified model equations are called algebraic Reynolds stress models.1 There are only two
equations to be solved, however with more directional information than in the usual k-ε model with
turbulence viscosity. It has to be mentioned that there are other approximations for the relation
between the transport terms Ti j and Tk than the one shown in Eq. (6.23).
Even more simplified equations give a directional turbulence viscosity that can be used in a k-ε
model.
A drawback of the algebraic expressions is that they can give instabilities in the numerical solution,
which requires more computational time. Some investigators say that the gain may be lost by the
extra efforts. Others find that the algebraic models are a useful alternative. Those who already have
(and can use) a code that uses these models, will be likely to have gain from it.
A unique feature of the algebraic equations for the Reynolds stresses is that we have the correct
answer for the approximation made, namely the full transport equation for the Reynolds stresses. The
source terms are, in principle, the same, only the transport terms are simplified. This makes it simple
to evaluate whether it is a “sufficiently good” model. If the program also can solve the transport
equations, one can see the savings of time and the error made by approximating the transport terms.
For combustion problems, these algebraic models are not very relevant. There are so many other
things that require computational time that a few more equations do not make the big difference. On
the other hand, the problems are complex, and instabilities and other additional problems should be
avoided if possible.
1 The term algebraic (Reynolds) stress model is somewhat awkward because also models with a turbulence viscosity
are algebraic models for the turbulence stresses, Eq. (2.14). The models are also termed algebraic-stress-equation models
as opposed to differential-stress-equation models to make the distinction precise. Nevertheless, the interested reader will
understand what is meant, regardless of the terminology used.
6.6 Why (not) use the Reynolds stress equations? 105
6.6 Why (not) use the Reynolds stress equations?

In computational codes for flow, one can often choose between various turbulence models, or one
have to decide whether to spend the work of implementing a new model.
It is not difficult to find flows where the Reynolds-stress equations give better predictions than the
k-ε model (see Sec. 3.6, page 58). Furthermore, it is not difficult to argue against a statement saying
that more equations require so much more computer resources. Seen in this perspective, there is
much that speaks in favor of replacing the k-ε model by the Reynolds-stress equations. A large
number of investigators draw exactly that conclusion.
However, the practice is not that simple. The Reynolds-stress equations are much harder to code.
As opposed to k and ε, a shear stress has a direction. When we do not have a simple turbulence
viscosity, we have to have special treatment of the Reynolds stresses in the momentum equations.
Moreover, the requirements to the boundary and initial conditions are stricter. Solving the Reynolds-
stress equations is more prone to unstabilities. Many practical problems have complex boundaries,
often with separated surfaces and bodies inside the computational volume.
We often have to use wall laws. We then lose some of the benefits of using Reynolds-stress equations
– because wall laws are not as general as the model itself. It also seems that the wall laws for
Reynolds stresses require a finer grid than wall laws for k and ε, see above at page 102.
Another aspect is that simulation of turbulent combustion comprise several other models, for instance
combustion models and radiation models. It is far from sure that the turbulence model is the weakest
link of the chain. On the other hand, we have seen cases where better turbulence transport from a
turbulence model with Reynolds-stress equations gave better combustion results than with the k-ε
model. An example is the case presented in Fig. 6.1. For isothermal mixing, the results were not very
different. With combustion modeled as fast reactions, the differences were apparent. When finite
chemical kinetics was used, the k-ε model gave blow-off whereas the Reynolds-stress equations gave
a stable solution. (Gran, 1994:108).
A third question can be how accurate results you actually need. It can not be denied that the
Reynolds-stress equations require more resources, and man-hours may be the most important. The
question is whether the extra effort pays off.
As a summary, we can safely conclude that there is no simple and general answer to the question of
which model is the best.
Chapter 7
Statistical functions in turbulence

7.1 Statistical functions
The contents of this section should be known to most readers from courses
in statistics. It is included partly for repetition and partly in order to put the
statistical functions into the context. The subject is more extensively treated
by, among others, Pope (1985), Kuo (1986:432f) and Tennekes and Lumley
(1972). In addition, textbooks in statistics will give a deeper understanding.
We look at a stochastic (= random) variable u, (−∞ ≤ u ≤ ∞). The probability for u being less
than a given value c, (−∞ ≤ c ≤ ∞) is
P(u < c) = fraction [0,1] of the outcome where u < c (7.1)
or
number of experiments where u < c
P(u < c) = . (7.2)
total number of experiments
We can regard this as a function of c because P(u < c) varies with c. We define a distribution
function:
F(c) ≡ P(u < c). (7.3)
A realization c is the value of the variable u in a single trial, or in one point of time and space. The
values that c can take, and thus the values that u can have, are called the sample space.
A typical distribution function is shown in Fig. 7.1. Here, the sample space is (−∞, +∞). From
the definition, we see that
P(c1 ≤ u < c2 ) = P(u < c2 ) − P(u < c1 ) = F(c2 ) − F(c1 ). (7.4)

108 7 Statistical functions in turbulence
F(c)
c
0
Figure 7.1: Typical distribution function (sketch)
Properties of the distribution function F(c) are:
• F(c) ≥ 0, because a fraction is not negative

• F(c1 ) ≤ F(c2 ), if c1 ≤ c2
• F(c) ≤ 1
• F(−∞) = 0 and F(∞) = 1
The probability density function (pdf) or probability density, is defined as the function
P(c ≤ u < c + 1c) F(c + 1c) − F(c) d F(c)
f (c) ≡ lim = lim = . (7.5)
1c→0 1c 1c→0 1c dc
Properties of f (c):
• f (c) ≥ 0, because F(c) do not decrease with larger c

Z b
• f (c)dc = F(b) − F(a) = P(u < b) − P(u < a) = P(a ≤ u < b)
a
Z b Z ∞
• f (c)dc = F(b) − F(−∞) = F(b) and f (c)dc = F(∞) − F(−∞) = 1
−∞ −∞
For a normal distribution, the probability density is

1 (c − c)2
f (c) = √ exp − , (7.6)
2πσ 2σ 2
where c is a mean value (average, expectation value) and σ 2 = c′ 2 is the variance (see below).
The velocity fluctuations can have an approximately normal distribution, whereas skew distributions
(see below) are typical for flows with strong gradients in turbulence intensity (see more in Townsend
1976:126f).
7.1 Statistical functions 109
Figure 7.2: Probability density functions (here denoted as P(ξ )) for a conserved scalar in
different types of flow (sketches from Bilger, 1980). Perfect mixing, e.g. far to the right in
the reactor or Fig.(c) would give an impulse function (delta function) at ξ = 0.5. Notice that
Fig. (a) is drawn so that ϕmix = ϕ1 for ξ = 0, thus the opposite of Eq. (5.11) and Fig.(b)
where ϕmix = ϕ1 for ξ = 1.
6
f (ξ )
5
a = b = 25
4 a=b
= 10
3
a=b=5
2
a=b=2
1
a = b = 0,1
0 ξ
0 0.2 0.4 0.6 0.8 1
6
f (ξ )
a = 0,1
5 b=5
a=5
4 b=1
a=2
b = 10
3
a=3 a=b=5
b=5
0 ξ
0 0.2 0.4 0.6 0.8 1
Figure 7.3: Examples of curves for β-probability density functions for different sets of pa-
rameters a,b, Eqs. (10.12)–(10.13). Some statistical quantities are given in Table 7.1
Table 7.1: Mean value ξ , standard deviation σ = (ξ ′ 2 )1/2 , intensity σ /ξ , skewness S =

ξ ′ 3 /σ 3 and flatness K = ξ ′ 4 /σ 4 calculated for β-probability density functions with different
sets of parameters a,b, Eqs. (10.12)–(10.13). See also Fig. 7.3.
a b ξ σ σ /ξ S K
0.1 0.1 0.5 0.46 0.91 0 1.19
2 2 0.5 0.22 0.44 0 2.14
5 5 0.5 0.15 0.30 0 2.54
10 10 0.5 0.11 0.22 0 2.74
25 25 0.5 0.07 0.14 0 2.89
0.1 5 0.020 0.056 2.81 3.5 18.5
5 1 0.83 0.14 0.17 −1.2 4.2
3 5 0.375 0.16 0.43 0.31 2.6
2 10 0.17 0.10 0.62 0.92 3.8
Another probability density is the β-probability density, which is shown in Eq. (10.12). This func-
tion, with various values of the parameters a and b, is shown in Figure 7.3. We see that the function
can represent states with, e.g., peaks at both edges (small values for both a and b); peak at one edge
(small a, large b, or vice versa); or a peak somewhere in between.
More examples on such function are found in Kuo (1986:440ff). Probability density functions are
used in some combustion models. This we will come back to in Chapter 10.
Moments
For a function G = G(u), where u is a random (stochastic) variable with c as realization and a
probability density f (c), the mean value is
Z ∞
G= G(c) f (c)dc. (7.7)
−∞
This is called the first moment or expectation value. From this, the mean value of u can be expressed
by setting G(u) = u (and G(c) = c):
Z ∞
u= c f (c)dc. (7.8)
−∞
Here the moment of a stochastic variable is expressed by the distribution of the stochastic variable
itself (or by a realization of it, here denoted c). In Eq. (7.7), the stochastic function G and the
probability density f are expressed as functions of another stochastic variable or function, u (i.e. the
realization of it, c). This we will see again in Sec. 10.7, where the temperature, mass fraction and
other quantities are expressed as functions of the mixtures fraction.
The second moment about zero, or the mean-square value, is

Z ∞
u2 = c2 f (c)dc, (7.9)
−∞
and correspondingly with u n and cn for higher moments.
The central moment is the moment about the expectation value (mean value). The first central
moment is Z ∞
(u − u) = (c − u) f (c)dc = 0. (7.10)
−∞
This is the mean value of what we call the fluctuation, u ′ = u − u.

The second central moment is called variance and gives a measure of the departure from the mean
value: Z ∞
(u − u)2 = (c − u)2 f (c)dc (7.11)
−∞
or Z ∞
u′ 2 = c′ 2 f (c)dc. (7.12)
−∞
q p
The standard deviation is the square root of the variance: σ = (u − u)2 = u ′ 2 . This is a
root-mean-square value (rms value).
If u is the velocity,
p we obtain the turbulence intensity by dividing the standard deviation by the mean
value: σ/u = u ′ 2 /u. This quantity is often reported from measurements.
Two higher-order moments we can come across are the skewness factor, S, and the flatness factor,
K.
The skewness factor for a variable u is defined as
u′ 3 u′ 3
S= = . (7.13)
σ3 (u ′ 2 )3/2
This gives a measure for how skewed the probability density f (c) is about the mean value, that is
for the deviation from symmetry. For a symmetric distribution, for instance the normal distribution,
S = 0. A skewed probability density (S 6= 0) can have a maximum value (most probable value) that
is not equal to the mean value.
The flatness factor for a variable u is defined as
u′ 4 u′ 4
K = 4
= . (7.14)
σ (u ′ 2 )2
This quantity is also known as the kurtosis (Greek: curvature). It gives a measure for how flat the
probability density f (c) is about the mean value. Table 7.1 shows flatness factors calculated for the
functions of Fig. 7.3. For a normal distribution, the flatness factor equals 3.0. For a function that is a
triangle with two equal legs (isosceles), the factor is 2.4, and for a square function 1.8. These three
values are independent of the width of the function. A large value for the flatness factor will indicate
intermittency in the flow (see page 119). Interpretation of the flatness factor for the flow velocity is
discussed by Hinze (1975:242) and McComb (1990:103).
Two-dimensional distributions
For two stochastic variables (u,v), the distribution function is expressed as
Fu,v (c1 , c2 ) ≡ P(u < c1 , v < c2 ) = F(c1 , c2 ), (7.15)

where c1 and c2 are specified values. We can regard this as a function of c1 and c2 , because P and
F vary with c1 and c2 .
The probability density for (u,v) is
P(c1 ≤ u < c1 + 1c1 , c2 ≤ v < c2 + 1c2 ) ∂ 2 F(c1 , c2 )

f u,v (c1 , c2 ) ≡ lim = = f (c1 , c2 ).
1c1 →0 1c1 · 1c2 ∂c1 ∂c2
1c2 →0
(7.16)
As for one-dimensional distributions, we can determine mean values or moments. For a function
G = G(u, v), the expectation value or mean value is
Z ∞Z ∞
G= G(c1 , c2 ) f (c1 , c2 )dc1 dc2 . (7.17)
−∞ −∞
The covariance or correlation between u and v is expressed as

Z ∞Z ∞
(u − u)(v − v) = u ′ v ′ = (c1 − u)(c2 − v) f (c1 , c2 )dc1 dc2 . (7.18)
−∞ −∞
If u and v are independent, the covariance u ′ v ′ = 0. (But we can not conclude the other way round.)
The correlation coefficient for u and v is defined as
Covar(u, v) u′v ′
Ru,v = √ √ = p p . (7.19)
Var(u) · Var(v) u′ 2 · v ′ 2
Here, |Ru,v | ≤ 1.
The functions F and f are also functions of space and time, thus f (u, v) is actually f (u, v, xE, t).
Two-point correlations
We can define a correlation between quantities in two points: u ′ (E x + rE). This two-point corre-
x )v ′ (E
lation says something about the joint variation between u ′ in position xE and v ′ in position (Ex + rE),
cf. Fig. 7.4
x2 v′
✻
rE ✿
✲ u′
✒
xE
x1
x3
Figure 7.4: Fluctuations in two locations
More general, we can define the correlation
x , rE, t) = u ′i (E
Ri j (E x , t)u ′j (E
x , rE, t). (7.20)
Here, it holds that Ri j (E

x , rE, t) → 0 when |Er | → ∞, because the variables in two points infinitely far
from each other have to be independent.
The expression can be simplified with different presumptions:
Statistically stationary turbulence is independent of time: Ri j = Ri j (E
x , rE).
Homogeneous turbulence (see page 63) is independent of location in space: Ri j = Ri j (Er , t).
Isotropic turbulence (see page 63) is independent of orientation in space and must in practice also be
homogeneous: Ri j = Ri j (r, t), where r = |Er |. For stationary isotropic turbulence is Ri j = Ri j (r ).
7.2 Isotropic turbulence 115
7.2 Isotropic turbulence

There are hardly any flows that really are isotropic. Nevertheless, many have
pondered much about isotropic turbulence. With this simplification, the system
can be treated analytically. Through the theory of isotropic turbulence, one can
gain understanding that is also valid for turbulence more in general. One can
also develop relations that can be used as approximations for real flows. We
can find extensive sections on isotropic turbulence e.g. in Hinze (1975) and
Lesieur (1990).
We touched this subject in Sec. 4.1 and 4.2. When we regard the turbulence as homogeneous,
stationary and isotropic, the two-point correlation is a function of the distance between two points:
Ri j = Ri j (r ).
We consider two points with distance r , and the velocity fluctuation along the line through the two
points:
u ′ℓ r u ′ℓ (r )
✲ ✲
The correlation is
Rℓℓ (r ) = u ′ℓ · u ′ℓ (r ) = f (r ) · u ′ 2 . (7.21)
Here, we have introduced and defined the longitudinal correlation function (“longitudinal”: along
with):
u ′ u ′ (r )
f (r ) ≡ ℓ ℓ . (7.22)
u′ 2
We continue looking at two points with distance r and at the velocity fluctuation across the line
through the two points:
✻′ ✻′
un u n (r )
r
The correlation is
Rnn (r ) = u ′n · u ′n (r ) = g(r ) · u ′ 2 . (7.23)
The lateral correlation function (“lateral”: side-, sideways) is
u ′n u ′n (r )
g(r ) ≡ . (7.24)
u′ 2
These two functions are independent of direction:
f (−r ) = f (r ) and g(−r ) = g(r ). (7.25)

f (r ) g(r )
1 1
0 r 0 r
0 0
Figure 7.5: Typical curves (sketches) for f (r ) and g(r )
If the two points fall together (r = 0), we look at the one and same fluctuation (autocorrelation).
The correlation is then complete, or
f (0) = 1 and g(0) = 1. (7.26)
If there is an infinite distance between the two points (r → ∞), there is no relation or influence from
one point to the other, hence
f (∞) = g(∞) = 0. (7.27)
Typical curves for f (r ) and g(r ) are sketched in Fig. 7.5. The value of f (r ) can be negative for a
range of r . The value of g(r ) have to be negative for a range of r . Because of continuity, a fluctuation
across the line have to be balanced by fluctuations in the opposite direction.
Length scales in isotropic turbulence
f (r ) f (r )
1 1
0 r 0 r
0 Lf 0 λf
Figure 7.6: Length scales from f (r )

7.2 Isotropic turbulence 117
We define a length scale L f such that the area of the rectangle f (0) · L f is equal to the area under
the curve f (r ) for r ≥ 0, cf. Fig. 7.6. Since f (0) = 1, we can write
Z ∞
Lf = f (r )dr. (7.28)
0
This is known as the longitudinal macroscale or longitudinal integral scale.

Similarly, from the lateral correlation function g(r ), we can define the lateral macroscale or lateral
integral scale: Z ∞
Lg = g(r )dr. (7.29)
0
It can be shown that L f = 2L g .

These length scales tell something about how distant fluctuations affect each other, or how far an
influence is “felt”.
A Taylor expansion of f (r ) about r = 0 gives

1
f (r ) = f (0) + f ′ (0) · r + 2 f ′′ (0) · r 2 + · · · . (7.30)
From before we have f (0) = 1 and f ′ (0) = 0 (symmetry).

When the series expansion is truncated after the r 2 -term, this gives a parabola about r = 0, see
Fig. 7.6. We define a length scale λ f as the r value where this parabola crosses the r -axis, thus
f (r = λ f ) = 0 in Eq. (7.30):
2
λ2f = − ′′ . (7.31)
f (0)
This is known as the longitudinal microscale, or longitudinal Taylor scale.

Similarly, from the lateral correlation function g(r ) we can define a lateral microscale or lateral
Taylor scale:
2
λ2g = − ′′ . (7.32)
g (0)
√
It can be shown (see Hinze, 1975:188) that λ f = 2 λg .
λ) and dissipation
A relation between microscale (λ
Dissipation of turbulence energy is

∂u ′i ∂u ′i
ε=ν . (7.33)
∂x j ∂x j
For isotropic turbulence must

2 2 2
∂u ′1 ∂u ′2 ∂u ′3
= = (7.34)
∂ x1 ∂ x2 ∂ x3
and
2 2 2 2 2 2
∂u ′1 ∂u ′1 ∂u ′2 ∂u ′2 ∂u ′3 ∂u ′3
= = = = = . (7.35)
∂ x2 ∂ x3 ∂ x1 ∂ x3 ∂ x1 ∂ x2
Thus is  
2 2
∂u ′1 ∂u ′1
ε = ν 3 +6 . (7.36)
∂ x1 ∂ x2
It can be shown (Hinze, 1975:189) that

′ 2
∂u 1 2u ′ 2
= −u ′ 2 · f ′′ (0) = 2 (7.37)
∂ x1 λf
and
2 ′ 2
∂u ′1 4u ′ 2 2u ′ 2 ∂u 1
= −u ′ 2 · 2 f ′′ (0) = = = −u ′ 2 · g ′′ (0) = 2 . (7.38)
∂ x2 λ2f λ2g ∂ x1
What you show first and last of these equalities depends on how you proceed. Anyhow, we obtain
that
′ 2
∂u 1 u′ 2 u′ 2
ε = 15ν = 30ν 2 = 15ν 2 . (7.39)
∂ x1 λf λg
The procedure to obtain these relations (or similar relations) varies in the literature. For instance
Tennekes and Lumley (1972:66) define the microscale λ by the expression
′ 2
∂u 1 u′ 2
≡ 2 (7.40)
∂ x1 λ
in isotropic turbulence. Then, ε = 15νu ′ 2 /λ2 .
For anisotropic turbulence, we can define length scales λi j so that for instance
2
∂u ′1 2u ′12
= . (7.41)
∂ x2 λ212
This λi j is not a tensor, but nine different length scales. Among other Tavoularis and Corrsin
(1981:330) have used (and measured) such length scales. It can be convenient to define one sin-
gle length scale λ so that
u′ 2 k
ε = 15ν 2 = 10ν 2 . (7.42)
λ λ
7.3 Intermittency 119
Then we still have one length scale, as in isotropic turbulence. λ corresponds to λg . With this length
scale, we can define a turbulence Reynolds number,
u′λ
Reλ = . (7.43)
ν
This is what is often reported from experiments as the turbulence Reynolds number. With k and ε
from a simulation, we find u ′ = ( 32 k)1/2 and λ = (10νk/ε)1/2, and hence
21
u′λ 20 k 2
Reλ = = . (7.44)
ν 3 νε
The latter expression contains the Reynolds number ReT = k 2 /(νε), cf. also Eq. (9.17).
Scalar correlations
Here we have only looked at correlations for velocity, that is mean values of the product of two
velocity fluctuations. In the same way, we could put up correlations for scalars. We can define
correlation functions, integral scales and Taylor micro-scales. In a mixture with many chemical
species (e.g. combustion), there could be a wide range of length scales. We will come back to this
in Sec. 8.6 and 10.10, but I will mainly leave the reader to himself and the literature of the subject.
Hinze (1975:278ff) can be a place to start.
7.3 Intermittency
In the outer part of a boundary layer, there will be a region that now and then has intense turbulence
and that is calm in between. The sketch shows a location that for the moment is quiet but will get
intense turbulence when one of the large vortices reaches it. The curve of Fig. 7.7 shows the time
lapse of such a point of measurement.
This jerking shift between turbulent and laminar flow is called intermittency (from Latin “inter”:
between, “mittere” of “miss-”: let go, send.) We can define an intermittency factor by
time turbulent
I = . (7.45)
total time
The intermittency factor can also be a volume fraction,
volume turbulent
I = . (7.46)
total volume
Byggstøyl and Kollmann (1981,1986a,b) present turbulence models that account for this alternation
between intense and weak turbulence at a boundary between two states of flow.
u∞
Undisturbed
flow
Turbulent
flow
Figure 7.7: Intermittency: Alternation between turbulent and laminar motions at the bound-
ary of a flow (sketch). The curve is an imagined signal from the fixed point of measurement
marked with ⊕.
Turbulent flow consists of larger and smaller eddies. The dissipation occurs mainly in the smaller
eddies (the fine structure). Small eddies can be imagined as long and thin vortex threads (Tennekes,
1968). In the flow, these threads will gather so that certain regions get many small eddies, while less
in other regions. We can imagine that the threads are gathered in sheets around larger eddies and, in
particular, in the shear layer between them (Tennekes, 1968; Corrsin, 1962). The dissipation will be
distributed according to the distribution of small eddies.
Also for the dissipation, we can define an intermittency factor. This is the fraction of time for a point
in space, or the fraction of volume for a point in time, where the dissipation is large.
Chapter 8
Energy transfer in turbulent flow

8.1 Break-up of eddies in turbulence
This section will present a simplified picture of the eddy structure of turbulent
flow. Here it may be appropriate to remind about the usefulness of both
simple and complex models, see Sec. 2.7. A simplified model does not explain
everything, but that is not the point either.
Here we will look at turbulence as a series of large and small eddies. The large ones are broken
down to smaller, and the smallest eddies dissappear in diffusion. We hardly know who was the first
to think along such lines. However, the summarizing rhyme of Richardson (1922:66)1 is clearly the
most famous.
We can look at a simplification of the momentum equation:
∂u ∂u ∂ 2u
+u = ν 2. (8.1)
∂t ∂x ∂x
Here we introduce dimensionless quantities, ũ = u/u o , x̃ = x/L and t˜ = t · u o /L. The dimension-
less equation becomes
∂ ũ ∂ ũ ν ∂ 2 ũ 1 ∂ 2 ũ
+ ũ = = . (8.2)
∂ t˜ ∂ x̃ u o L ∂ x̃ 2 Re ∂ x̃ 2
The viscous terms go towards zero for high Reynolds numbers, Re. This means that the mechanism
that spreads and evens disturbances is not working anymore. The flow will be unstable for small
disturbances.
The mean flow has characteristic scales u and L. u is a mean velocity and L is given by the external
dimensions, for instance the diameter of a pipe.
1 «We realize thus that: big whirls have little whirls that feed on their velocity, and little whirls have lesser whirls and so
on to viscosity – in the molecular sense.» In his book it is written as prose but the rhyme is often reproduced in verse form.
122 8 Energy transfer in turbulent flow
✲
✲ ✻
uL
✲ L The Reynolds number is Re L = ≫ 1. The
ν
u ✲ flow is unstable and the pattern is broken down
✲ ❄ and eddies are formed.
We look at eddies with characteristic length ℓ′ and velocity u ′ . These are “large” eddies.
ℓ′ u′
u ′ ℓ′
Reynolds number: Reℓ′ = ≫ 1.
ν
The motion is unstable and the eddy is broken down to smaller eddies with characteristic length ℓ′′
and velocity u ′′ .
u ′′ ℓ′′
Reynolds number: Reℓ′′ = ≫ 1 . The mo-
ℓ′′ ν
u ′′ tion is unstable and the eddy is broken down to
smaller eddies.
The process goes on until the Reynolds number reaches an order of magnitude of one.
η v
vη
Reynolds number: Reη = = Recrit .
ν
Here the inertial forces and viscous forces (diffusion) balance each other. If the eddy becomes even
smaller, viscous stresses will break up the eddy. The molecules diffuse in all directions and do not
constitute a joint motion anymore. The kinetic energy of the eddy will be lost as heat (dissipation).
The viscous forces set a limit for the smaller scales.
This breakdown of the flow into larger and smaller eddies can be illustrated in different ways. There
are several simplified pictures – models – of the process. We just have to keep in mind that as soon
as we say anything specific about this, then we have simplified the situation.
The breakdown to smaller and smaller eddies occurs partly by large eddies accelerating adjacent fluid
to an eddy motion of a smaller scale. And it occurs partly by eddies being stretched to be thinner and
longer. In both cases, kinetic energy is transferred to smaller scales. This thought picture is sketched
in Fig. 8.1.
The energy transfer can be sketched as in Fig. 8.2. This is a simplified picture. Averaged over time
there is a continuous transition from the largest length scale L to the smallest η. We can talk about a
spectrum, and this we will consider more in the next section.
Nevertheless, the continuous process can be modeled as a series of such steps or levels according
8.1 Break-up of eddies in turbulence 123
Figure 8.1: Image of vortex stretching, energy transfer to smaller scales
Mean flow u, L
Energy transfer
u ′ , ℓ′ u ′′ , ℓ′′ u ′′′ , ℓ′′′ ··· v, η
Dissipation
Figure 8.2: Energy transfer in turbulence, schematic
to the pattern in Fig. 8.2. We then talk about a cascade model. (Cascade is French for a stepwise,
often artificial, waterfall; from Italian cascata fall, waterfall.) Such models are proposed by, among
others, Onsager (1949), Tennekes and Lumley (1972:260f) and Magnussen (1975b, 1981; see also
Chapter 11).
Different authors will put emphasis on different aspects of the process. It will pay off to read other
presentations in addition to this one, for instance Bradshaw (1971) and Tennekes and Lumley (1972).
The spectrum of scales in turbulence led Bradshaw to the definition of turbulence (page 19).
Transport and mixing
Transport with the turbulence, «turbulent trans-

port», is primarily conducted by the large ed-
dies. A large eddy moves a “package” of fluid,
for example with another temperature, over the
distance ℓ′ . A small eddy does the same, but
ℓ′ the transport has less contribution since the dis-
ℓn tance ℓn is small.
We say that turbulent transport is a large-scale
phenomenon. The length scale ℓ′ is a diffusion
Figure 8.3: Thought picture of transport length scale.
with the turbulence in large and small ed-
dies The small eddies have not much to say for
the transport but are important for combustion.
Combustion requires mixing of molecules of
fuel and oxygen – and of products to ignite. If fuel and air is not well mixed from before, there
will be separate eddies of fuel and air. The species are mixed at the interface of two large eddies.
This sheet is however, small compared to the volume.
In small eddies, the mixing is more efficient. The interfaces are larger. Viscous forces (diffusion)
break up the small eddies and spread, mix, the molecules.
In the regions where dissipation occurs, also the molecular mixing takes place. Most of the dissipa-
tion occurs in the small eddies – and most of the molecular mixing, and hence also the reactions.
8.2 Energy spectrum for turbulent flow

We define a two-point correlation (cf. page 114 and
u ′j Fig. 8.4):
✻ Ri j (Er , t) = u ′i (E
x , t)u ′j (E
x + rE, t). (8.3)
rE ✿
This can be transformed through a Fourier transforma-
✲ u′
✒ i tion,
F
Ri j (Er , t) −→ E i j (E
κ , t). (8.4)
xE
E i j is the energy-spectrum tensor. κE is the wave-
number vector and has the dimension (length)−1. An
inverse Fourier transform of the energy-spectrum ten-
Figure 8.4: Fluctuations in two sor E i j (E
κ , t) leads back to the correlation Ri j (Er , t).
points
The energy-spectrum tensor E i j (E
κ , t) depends on the
coordinate directions. Furthermore, we can define the three-dimensional energy spectrum,
E(κ, t) = 2πκ 2 E ii (E
κ , t), (8.5)
8.2 Energy spectrum for turbulent flow 125
where the wave number κ = |E κ | and E ii = E 11 + E 22 + E 33 . The energy spectrum E(κ, t) is

independent of direction. It is defined so that
Z ∞
E(κ, t)dκ = 21 u ′i u ′i = k(t). (8.6)
0
This is the turbulence energy. Often, we consider steady-state conditions and do not include the time
dependency. The sketch in Fig. 8.5 shows a typical curve for the energy spectrum E(κ).
E(κ)
D(κ)
κ
κe κd η−1
Figure 8.5: Sketches of the three-dimensional energy spectrum E(κ) and dissipation spec-
trum D(κ)
The vertical axis is linear. The horizontal axis of the figure is the wave number κ, the number of
waves or eddies per unit of length. Here, a logarithmic scale is used. The energy spectrum could also
have been expressed as a function of frequency, that is eddies per unit of time. The wave number
can be called a “space-frequency”. A point at the horizontal axis represents a certain length scale.
The ordinate value E expresses how much turbulence energy that relates to this lengthscale. Summed
(integrated) over all wave numbers or length scales we thus obtain the total turbulence energy.
We can say that the spectrum represents the average energy in one point (E x , t). It tells how the energy
on average is distributed over larger and smaller eddies that pass by this point.
The sketch shows a peak in the curve at low wave numbers (κ = κe ). We talk about the energy-rich
eddies, or the energy-rich scale range. We then mean the range about this peak. When we talk about
“large” eddies, we think on this range. The corresponding length scale is ℓe ∼ κe−1 .
Larger turbulence eddies exist. They can carry large energy but they are few and make little contri-
bution to the average. Small eddies (large wave number) are numerous, but each carry little energy
so that their sum is low.
The dissipation spectrum D(κ, t) is defined such that

Z ∞
D(κ, t)dκ = ε(t). (8.7)
0
Then is
D(κ, t) = 2νκ 2 E(κ, t). (8.8)
The sketch above shows a typical curve for a dissipation spectrum D(κ) together with an energy
spectrum E(κ). The curve shows how the dissipation on average is distributed on larger and smaller
eddies. The large eddies are not much affected by viscous forces and they represent a small part of
the dissipation. In the smaller eddies, the local gradients are larger and therefore, also the friction
work is larger. The contribution to the dissipation from this range of scales is larger. The most
dissipative range of scales is found about the peak κ = κd .
We have seen that the largest eddies can be as large as allowed by external boundaries, and that the
smallest can be as small as the viscous forces allow. We can define a turbulence Reynolds number,
for instance Reλ = u ′ λ/ν (cf. page 119). It will give a measure for how strong the turbulence is in
a flow. It will also give a measure for the difference between the “large” and the “small” eddies. If
the Reynolds number is high, there will be a large difference between large and small eddies. For
3/2
isotropic turbulence is ℓe /η ≈ 0.1Reλ (Hinze, 1975:225).
E(κ)
t
Figure 8.6: Sketch of the change in the energy spectrum E(κ) for decaying turbulence
Example: Energy spectrum for isotropic decaying turbulence
In Sec. 4.2 we saw how equations for k and ε could be simplified and solved for isotropic turbulence.
(See more details in the exercise compendium.) How does this show up in a spectrum?
Experiments (summarized by Ferziger, 1980) show k ∼ t −n , where n is well inside the interval
[1,2]. From this we can find that ε ∼ t −(n+1) , Ret ∼ t 1−n , ℓe ∼ ℓ′ ∼ t (1−n/2) , η ∼ t (n+1)/4 and
8.3 Kolmogorov’s hypotheses 127
ℓe /η ∼ t 3(1−n)/4 . When the turbulence decays, k, ε and the turbulence Reynolds number have to
decrease. The length scale increases however. We can think that when the whirling motions lose
strength, the particles drift more and more apart and the eddies become larger.
The area under the curve for the energy spectrum is the turbulence energy k; hence, this area will be
reduced. Furthermore, the length scale ℓe becomes larger so that the peak of the E-curve moves to
lower wave number (larger length scale). The Kolmogorov length scale, Eq. (8.20), also increases.
Thus, the ratio ℓe /η is a measure of the width of the spectrum, and this increases. Altogether, the
E-curve becomes lower and drifts to the left in the diagram when the turbulence decays, as it is
sketched in Fig. 8.6.
8.3 Kolmogorov’s hypotheses

Local isotropy
At high turbulence Reynolds numbers, the “large” eddies are much larger than the “small” eddies:
κe ≪ κd , or expressed by the corresponding length scales: ℓe ≫ ld .
Kolmogorov then states that
– small eddies are independent of large eddies and of the mean flow;
– all directions are equally probable for the small eddies.
This is called local isotropy: The directional orientation for a small eddy is independent of large
eddies. They are isotropically distributed.
Equilibrium range in the energy spectrum
The time scales for small eddies are short. They quickly respond to changes in the surroundings.
Therefore, small eddies are always in approximate equilibrium with the local properties of the mean
flow. This range of scales are called the equilibrium range or the universal equilibrium range. This
holds generally at high Reynolds number (thus “universal”).
Statistical quantities for the equilibrium range only depend on viscous forces and of the supply of
energy from somewhat larger eddies. At high Reynolds numbers, we assume that little or no kinetic
energy dissipates from larger eddies. The rate of supplied energy from larger eddies is thus the same
as the dissipation, ε.
The equilibrium range is independent of the mean flow, of the total turbulence energy and of the
formation of eddies. In short, they are independent of the left side of the spectrum. We can express
statistical quantities for the equilibrium range as functions of the dissipation ε (actually transferred
kinetic energy), the viscosity ν and the wave number κ, possibly also the time t. The wave number,
or length scale, is included because the eddies are characterized by their own size.
Inertial subrange
In a part of the equilibrium range, the eddies are so large that they are independent of the viscous
forces. The eddies in this range of scales only depend on the energy transfer from larger eddies and
of their own size. Statistical quantities then become functions of the dissipation ε (i.e. transferred
energy) and the wave number κ, possibly also the time t.
This range of scales is called the inertial range or inertial subrange. Somewhat simplified we can
say that the eddies whirl ahead with the energy they have, without adopting or giving off much net
energy. They are too small to be affected by the mean flow and external dimensions. And they are
too large to have much dissipation.
The energy spectrum for steady-state turbulence is a function of the dissipation ε and the wave-
number κ: E(κ) = f (ε, κ) ∼ εα κ β . The function can be found from dimensional analysis. The
dimension for ε is [L 2 /T 3 ], for κ [L −1 ] and for E [L 3 /T 2 ]:
2 α β
α β L 1 L 2α−β
[E(κ)] = ε κ = · = . (8.9)
T3 L T 3α
2
This gives α = 3 and β = − 53 or, with a constant of proportionality:
E(κ) = CK · ε2/3 κ −5/3. (8.10)
This expression for the three-dimensional energy spectrum in the inertial subrange is known as Kol-
mogorov’s 35 -law. It fits reasonably well with experiments. The constant CK is called Kolmogorov’s
constant and is about 1.5, see review and discussion of experimental data by Sreenivasan (1995).
We will look at a small interval of scales, 1κn , about the wave number κn in the inertial range. The
characteristic velocity scale at this level we write as u ′n . We can regard this as a measure for the
velocity of the eddies of size ℓn ∼ κn−1 . The energy that is related to this interval is
u ′n 2 ≈ E(κn )1κn . (8.11)
In the diagram for E(κ), see the sketch in Fig. 8.5, this becomes a vertical slot about κn . With a
logarithmic scale, we can write 1κn ∼ κn ,
−2/3 2/3
u ′n 2 ≈ E(κn )κn ∼ ε2/3 κn ∼ ε2/3 ℓn . (8.12)
We resolve the dissipation and obtain

u ′n3
ε∼ . (8.13)
ℓn
This was for the inertial range. Previously, we have seen that Prandtl modeled the dissipation as
ε ∼ u ′ 3 /L, where u ′ and L are characteristic scales for the large eddies. The length scale L has the
same order of magnitude as the scale of the energy-rich eddies, ℓe ∼ κe−1 . Later, we will see that in
Magnussen’s model, ε ∼ u ∗ 3 /L ∗ , where u ∗ and L ∗ are characteristic scales for the smallest eddies.
(See also Hinze, 1975:225.)
8.3 Kolmogorov’s hypotheses 129
In the inertial subrange, the statistical quantities are functions only of the energy transfer (which is
equal to the dissipation ε) and the wave number κ. The wave number can be expressed by a length
scale so that we can express statistical quantities as functions of the dissipation and a length scale.
Structure functions
We will look at the quantity [u ′ (E x )]2 , where u ′ is the velocity for a local motion relative
x + rE) − u ′ (E
to a larger flow, thus like a fluctuation. Such functions are called structure functions and were
introduced by Kolmogorov (1941b).
For high Reynolds numbers, we can write for the inertial subrange
[u ′ (E x )]2 = f (ε, r ) = constant · εα r β .

x + rE) − u ′ (E (8.14)
Dimensional analysis gives α = β = 23 , so that
[u ′ (E x )]2 = c2 · (εr )2/3 .

x + rE) − u ′ (E (8.15)
The left-hand side can be rewritten as

2 2
d(x + r ) d x dr
[u ′ (x + r ) − u ′ (x)]2 = − = (8.16)
dt dt dt
and 2
dr
= c2 · (εr )2/3 . (8.17)
dt
This can be rearranged and integrated from r = ro , t = 0:

3/2
2/3
r (t) = ro + 32 cε1/3 t . (8.18)
Example (exam problem from the “algae summer” 1988):

An algae swarm of diameter do = 100 m goes with the flow in an ocean current that has a width of
L = 10 km and a velocity of U = 1 m/s. The swarm has a velocity relative to the ocean current
of u ′ = 0.1 m/s. How long time does it take before the diameter is d = 5 km, and how long has it
been moved during that time? (Given Eqs. (8.16) - (8.17) with c2 = 1.5.) The viscosity for water is
ν = 10−6 m2 /s.
Thus, we will try to determine the time t and the distance (t · U ).
First, we estimate the dissipation ε ≈ u ′ 3 /L = 10−7 m2 /s3 .
Taylor-scale: λ2 = 15νu ′ 2 /ε gives λ ≈ 1 m and Reynolds number Reλ = u ′ λ/ν ≈ 105 . This
Reynolds number is “high”; there
√ is an inertial subrange and we can use the development above:
We use Eq. (8.18) with c = 1.5 and ro = 100 m to determine t at r (t) = 5 · 103 m. This gives
t = 7 · 104 s = 19 21 hour. The distance is (t · U ) = 70 km.
Dissipative range
In the last part of the spectrum – the smallest eddies – statistical quantities are functions of the
dissipation ε and the viscosity ν. The viscous forces restrict the smallest characteristic quantities.
With length scale η and velocity scale v, the Reynolds number Reη = ηv/ν = 1 (by definition)
when there is a balance between inertial forces and diffusion.
Here, v = v(ε, ν) and η = η(ε, ν). Dimensional analysis gives
v = (νε)1/4 , (8.19)
3 1/4
ν
η= (8.20)
ε
and time scale
η ν 1/2 η2
τ= = = . (8.21)
v ε ν
These scales are called Kolmogorov’s micro-scales or just simply Kolmogorov-scales.
Example: Consider a “propeller” (fan, mixer or similar) in a

container. We feed 1 kW into 1 kg of fluid.
The production of turbulence energy is equal to the added power
(1 kW). When the system is in steady state, the dissipation equals
the production: ε ≈ Pk = ẇ = 1 kW/1kg = 103 m2 /s3 .
If the fluid is water, the viscosity is ν = 1.0 · 10−6 m2 /s and the
density ρ = 103 kg/m3 .
For 1 kg of water, the volume will be V = 10−3 m3 .
A “large” length scale we then obtain from the volume: L = V 1/3 = (m/ρ)1/3 = 0.1 m.
The Kolmogorov-scales become:
η = 5.6 · 10−6 m ; v = 0.2 m/s and τ = 3 · 10−5 s.
The ratio between largest and smallest length scales is L/η = 2 · 104 and thus V ≈ 1013η3 .
For air, the same exercise gives:
ν = 1.6 · 10−5 m2 /s, ρ = 1kg/m3 , V = 1m3 , L = V 1/3 = 1m.
The Kolmogorov-scales: η = 4.5 · 10−5 m ; v = 0.36 m/s and τ = 1 · 10−4 s.
L/η = 2 · 104 and V ≈ 1013η3 .
For numerical simulation without turbulence model (direct simulation), the grid mesh has to be so
fine that it captures the small scales. Hence, the grid must have at least V /η3 cells to resolve the
smallest scales. In the example, the dissipation is calculated as an average for the entire volume.
Locally, the dissipation ε can be larger. This reduces the length scale η so that we need more cells.
This we will return to in Sec. 10.10.
8.4 Energy spectrum and combustion 131
8.4 Energy spectrum and combustion

The theory and models for the energy spectrum are developed for flows with unchanged composition
and nearly constant temperature and density. What then, when we turn to cases with combustion?
Temperature, density and viscosity have large and in some cases sharp gradients.
Measurements of the energy spectrum and the dissipation in combustion zones are hard to conduct,
and not many experimental studies are available. One of the few is by Furukawa, Okamoto and
Hirano (1996). They measured the spectrum for a flow in three ways:
– continuous record in combusting flow;
– discrete record for the intermittent local reaction zone, that is, measurements recorded only
when the probe was in the reaction zone;
– continuous record for the same flow without combustion.
For the continuous records in combusting flow, they found that the spectrum ranged over about the
same wave numbers as for the record without combustion. Correspondingly, they found that the
Kolmogorov length scales were about the same.
Also in the local reaction zone, the wave numbers were similar. In this case, however, the Kolmogorov-
scale was 3–4 times larger than that derived from the continuous record. Since the viscosity increases
with temperature, it is reasonable that also the Kolmogorov scale increases (see page 146) in the
combustion zone. On the other hand, the eddies do not get any bigger – rather on the contrary. It
seems that the spectrum itself is less affected than the Kolmogorov scale.
The study also indicates that the combustion produces more small-scale turbulence. This turbulence
is dissipated in the same scale range, or in even smaller scales. Thus the turbulence energy and the
dissipation increase at high wave numbers (small length scales), so that the spectrum is stretched to
the right, while the Kolmogorov scale moves to the left on the wave number axis.
When the reaction zones expand, we can imagine that they both retard the turbulence and that they
enhance the turbulence. The retarding is because eddies have more rotational resistance when they
get bigger. The same expansion gives a motion that can produce more turbulence.
As said, the subject is not much investigated and the deductions are uncertain. Altogether, this may
indicate that we, in lack of better knowledge, should discuss and calculate on small-scale phenomena
in turbulence as if it had the viscosity – and thus the Kolmogorov-scales – of a cold flow. For sure,
we can not assume that the spectrum becomes narrower, even though the Kolmogorov length scale
increases.
8.5 Molecular mixing; scalar-dissipation

In Sec. 5.5 (page 87) we developed an equation for the variance of a scalar, with the equation for
the temperature variance T ′ 2 as example. Here, we will look at the destruction term, or the scalar
dissipation, in such an equation. We will see this from the viewpoint of breakup and stretching of
eddies, as the picture is used in this chapter.
We look at an eddy (or element) with temperature much larger than that of the surroundings, Fig. 8.7:
The energy flux (flux of T ) due to molecular diffusion is λ/(ρC p ) · (∂ T /∂ x j ) according to Fourier’s
Tsurr
T ∗ ≫ Tsurr ℓ′
T ∗ ≫ Tsurr
L∗ Tsurr
Figure 8.7: Thought image of eddy stretching; smaller length scale gives larger gradients
law. The local gradient of T , |∂ T ′ /∂ x j | ≈ 1T ∗ /L ∗ , is very large when L ∗ is small. Hence, the
local heat flux λ/(ρC p ) · (∂ T ′ /∂ x j ) is large.
Correspondingly, for mass fraction of a species, the local gradient is large because the length scale
is small.
We see that eddy stretching in the turbulence leads to good mixing of species and energy. In the
same eddies, we also have large local viscous stresses τ ∼ ν(∂u ′ /∂ x) ∼ ν(u ∗ /L ∗ ). These give
dissipation of mechanical energy into heat.
Mixing of species, mixing of temperature (energy) and dissipation are different phenomena but are
closely linked together.
In isotropic turbulence, the equation for temperature variance has a destruction term only:
∂T ′ 2 λ ∂T ′ ∂T ′
= −2 = −εT ′ 2 . (8.22)
∂t ρC p ∂ x j ∂ x j
Compare this with the k-equation, Eq. (4.2):
∂k ∂u ′ ∂u ′i
= −ν i = −ε. (8.23)
∂t ∂x j ∂x j
We can make a timescale θk from Eq. (8.23). This is a kind of loss-timescale, that is, a measure for
how fast an amount of turbulence energy is lost:
∂k k k
− ∼ ∼ ε, thus, θk ∼ . (8.24)
∂t θk ε
More generally, turbulence energy is also produced or supplied. We can then regard this as an
turnover timescale, a measure for how fast the turbulence is transferred through the system.
Similarly, from Eq. (8.22) we can make a timescale that is a measure for how fast the variance T ′ 2
is broken down:
∂T ′ 2 T′ 2 T′ 2
− ∼ ∼ εT ′ 2 , thus, θT ′ 2 ∼ . (8.25)
∂t θT ′ 2 εT ′ 2
8.6 More on minor scales 133
Since the eddy stretching is the same, the two time scales have to be approximately equal. This gives
the model ε
εT ′ 2 = c · T ′ 2 , (8.26)
k
where c is a constant with unity order of magnitude. The variance T ′ 2 have to diminish in the same
way as k, as sketched in Fig. 8.8.
T′ 2
Figure 8.8: Decay of turbulence energy and of temperature variance (sketch)
8.6 More on minor scales

The Kolmogorov scales represent or characterize the smallest eddies, we say.
This can easily be mis-interpreted so that they are the size of the smallest
eddies. They are not.
The Kolmogorov scales came forth when we assumed balance between inertial forces and viscous
forces (page 130). There are smaller scales. The smaller eddies, the larger viscous forces – and the
faster they are retarded and evened out. For the flow as a whole, they have little or no effect – at least
when the turbulence Reynolds number has a certain value, so that “large” scales are independent of
“small”.
As long as nothing particular happens with or in these “even smaller” (than “small”) eddies, one can
imagine that they behave similarly from case to case. They are only affected by the viscosity and
the energy supply (they are a part of the dissipative range), see e.g. Zhou (1993). Said otherwise:
They depend only on the Kolmogorov-scale. However, for mixing of chemical species, for instance
in combustion, one can imagine that theses “even smaller” scales can have some effect.
The Kolmogorov scales are defined from the energy spectrum and viscous destruction of motional
energy. From a scalar correlation (page 119), one can put up a scalar spectrum in the same way as
the energy spectrum (cf. e.g. Hinze 1975:283ff). And one can find different length scales from this
spectrum. Such a scale is the Batchelor scale
1/4
νD2 η
ηB = =√ . (8.27)
ε Sc
It comes forth by assuming equilibrium between convection and molecular diffusion – and is a
counterpart to the Kolmogorov length scale. For ordinary gas mixtures, for instance CO2 in air,
the Schmidt number Sc = ν/D is of order unity. For some species, the Schmidt number can be of
order 103 . Then the Batchelor scale becomes considerably smaller than the Kolmogorov scale. The
associated timescale becomes equal to the Kolmogorov timescale,
ηB2
τB = = τ. (8.28)
D
Here, nothing is said about which scalar we talk about. It can be temperature (energy), and then with
Prandtl number in place of Schmidt number, or it can be a mass fraction (concentration). For each
of the many species involved in combustion, we can define Batchelor scales.
This topic will not be further dealt with here, although we will touch it in Sec. 10.10 about direct
simulation of combustion. Micro-scales for scalar mixing is further discussed by Gibson (1968a,b),
Komori et al. (1991), Kerstein (1991) and others.
Even more on minor scales
The Kolmogorov scales represent the dissipation. Then we have accounted the dissipation over the
entire volume, or over the entire averaging time if looking at one point. However, the dissipation is
concentrated in certain regions (Sec. 7.3 and 11.4). We can imagine that dissipation is concentrated
in a smaller partial region, and let the volume of this partial region be a fraction γ ∗ of the entire
volume V . The dissipation restricted to and distributed over the smaller volume is ε∗ . The total
dissipation has to be the same, thus V · ρ · ε = γ ∗ V · ρ · ε∗ , or ε = γ ∗ ε∗ . A model for this volume
−3/4 −3/2
fraction is found in Magnussen’s model (EDC), Sec. 11.4, where γ ∗ ∼ Reℓ ∼ Reλ . Other
models also exist, cf. Ertesvåg and Magnussen (1997), that gives comparable relations.
From this, we get the length scale η∗ = (ν 3 /ε∗ )1/4 = (ν 3 /ε)1/4 · (ε/ε∗ )1/4 = η · (γ ∗ )1/4 , or
η∗ −3/8
∼ (γ ∗ )1/4 ∼ Reλ . (8.29)
η
Chapter 9
Various topics in turbulence

In this chapter, I have collected some sections that do not quite fit into the sectioning above. Parts of
it are summaries of things from several chapters.
9.1 Length scales and timescales in turbulent flow

This summarizes the scales we have looked at previously.
Mixing lengths, mixing time
Prandtl used the name mixing length for two length scales that he defined. The first one (Prandtl,
1925) is in the mixing length model, Sec. 2.6, where the length scale ℓ is defined by
∂u 1 ∂u 1
−u ′1 u ′2 = ℓ2 . (9.1)
∂ x2 ∂ x2
Usually, it is this length scale that is meant when someone is talking about “the mixing length”. In
the region for the logarithmic wall law (Sec. 4.4), the mixing length is proportional to the distance
to the wall; ℓ = κ x 2 .
The other length scale is defined by the dissipation term in the k-equation (Prandtl, 1945), Sec. 3.3:
k 3/2
ε = CD . (9.2)
L
These two length scales are proportional and of the same order of magnitude. Both are exchange
lengths for turbulence elements (“turbulence balls”) or length scales for turbulence diffusion.
136 9 Various topics in turbulence
The timescale for these motions can be written (cf. page 48) as
k
θk = . (9.3)
ε
Characteristic “large” length scale and timescale for turbulence
This is a length scale that represents, or gives a measure for, the large scales of the turbulence and
hence, a measure for the turbulence transport. See Sec. 2.8 and 8.1.
In a discussion or outline, all the large length scales above (mixing lengths, integral scales, length
scale for the energy rich scale range) could be used as a characteristic scale, ℓ′ , for the turbulence.
An associated timescale is θ = ℓ′ /u ′ , for instance θk = k/ε as above.
Macroscale (integral scale)
The macroscale or integral scale of isotropic turbulence is defined, Eq. (7.28), as

Z ∞ u ′ℓ u ′ℓ (r )
Lf = f (r )dr, where f (r ) = . (9.4)
0 u′ 2
This is a measure for how far a fluctuation is correlated with itself – how far the disturbance is felt.
L f is the longitudinal integral scale. The lateral, L g , is defined in the same way from the correlation
function g(r ), Eq. (7.29). L f and L g are about as large as the mixing length ℓ (and L) above.
If we define the correlation function in time instead of space as here, we can integrate over time and
find an integral timescale.
Taylor microscale
The Taylor microscale of isotropic turbulence is defined, Eq. (7.31), as

2
λ2f = − . (9.5)
f ′′ (0)
This is the longitudinal microscale, the lateral λg can be defined in the same way from g(r ). Usually,
it is λg that is applied.
The microscale can also be defined, Eq. (7.42), as
k
λ2 = 10ν . (9.6)
ε
For isotropic turbulence this corresponds to λg , but this definition is also valid for anisotropic turbu-
lence.
9.1 Length scales and timescales in turbulent flow 137
The microscale is a measure for the size of the eddies that mainly cause the dissipation (Hinze,
1975:42), or that contain most of the fine structure. (In the literature, opinions are somewhat divided
on the interpretation of this length scale.)
With a correlation function in time, we find a Taylor micro-timescale in the same way.
The energy-rich scale range
The length scale for the energy rich scale range, ℓe , is set equal to κe−1 or 2π · κe−1 (there is some
variation in the literature on this). As previously, κe is the wave number for the peak of the curve
for the three-dimensional energy spectrum (page 125). This length scale is of the same order of
magnitude as the mixing length.
The most dissipative scale range
We can also define a length scale for the peak of the curve for the dissipation spectrum (page 126),
ℓd ∼ κd−1 . Hinze (1975:224) refers experiments giving κd η ≈ 0.1. This means that ℓd is of order ten
times the Kolmogorov scale η.
Kolmogorov scale, microscale for the energy spectrum
The Kolmogorov microscale η = (ν 3 /ε)1/4 is a characteristic length scale for the smallest eddies of
the turbulence, page 130. The velocity scale is v = (νε)1/4 , and the Reynolds number Reη = ηv/ν
for these scales should be unity, by definition. Here, inertial forces and viscous forces will balance
each other. The associated timescale is τ = (ν/ε)1/2 = η2 /ν.
Batchelor scale, microscale for the scalar spectrum
These are scales that appear by assuming balance between local convection and molecular diffusion
1/4 √
in the small eddies, Sec. 8.6. The Batchelor length scale is ηB = νD2 /ε = η/ Sc. The
Bachelor timescale is τB = ηB2 /D, and is thus equal to the Kolmogorov timescale τ .
Convection time for the microscale
With a (characteristic) flow velocity U for the main flow, we can define
η
τU = . (9.7)
U
This is the time it takes for the Kolmogorov length scale η to travel past a point in the flow (Goulard,
Mellor and Bilger, 1976). The scale becomes less than the Kolmogorov timescale. The temporal
resolution in experiments or simulations should be finer than τU (cf. Sec. C.6).
Similarly, for a Batchelor length scale ηB we have

ηB
τU,B = . (9.8)
U
9.2 Length scales from turbulence models

When we do a simulation, we often wish to calculate a length scale. Usually we calculate a “large”
length scale, such as a mixing length or diffusion length scale.
1)
By assuming i) local equilibrium for turbulence energy (page 73) and ii) setting the turbulence vis-
cosity νt from the k-ε model equal to νt from Prandtl’s mixing-length model, we obtain an expression
for Prandtl’s mixing length.
Equilibrium (Pk ≈ ε) gives |∂u 1 /∂ x 2 | = (ε/νt )1/2 . From Prandtl’s mixing length (page 41), we
have |∂u 1 /∂ x 2 | = νt /ℓ2 . Now we set the gradients equal and introduce the turbulence viscosity
from the k-ε model, νt = Cµ k 2 /ε. This we arrange to an expression for the mixing length:
k 3/2
ℓ = Cµ3/4 . (9.9)
ε
2)
By setting νt = u ′ · ℓ′ (page 145), and use u ′ 2 = 32 k and νt from the k-ε model, we get
q
k 3/2
ℓ′ = 32 Cµ . (9.10)
ε
q
1/4
The ratio between the length scale in 2) and the length scale in 1) is 32 Cµ = 0.67.
√
Alternatively, we can say that νt = k · ℓ′ , and we get ℓ′ = Cµ k 3/2 /ε.
3)
If we use Eq. (9.20) and u ′ 2 = 32 k, we get
u′ 3 3/2 k 3/2
ℓ′ = A = A 32 . (9.11)
ε ε
The ratio between the length scale in 3) and the length scale in 1) equals 3.3 · A. The ratio between
the length scale in 3) and the length scale in 2) equals 5 · A. Thus, with A = 0.2 they are equal.
The three different scales as presented here are, however, so closely related that they can be regarded
as one scale.
9.3 Reynolds number for turbulence

In 1883, Osborne Reynolds discovered a general parameter for the transition from laminar to turbu-
lent flow in a pipe. The dimensionless parameter U · D/ν is later named after him.
9.3 Reynolds number for turbulence 139
In general, a Reynolds number is the ratio between inertial forces and viscous forces.
The original Reynolds number is the ratio between convective transport and diffusive transport in a
pipe. Later, similar parameters are identified and are also called Reynolds numbers. Here, the matter
is turbulence.
1) With a Taylor microscale, see page 119,

u′λ
Reλ = . (9.12)
ν
In isotropic turbulence, λ = λg (page 117). In anisotropic turbulence, λ2 = 15νu ′ 2 /ε or

λ2 = 10νk/ε.
In the literature, we often meet turbulence Reynolds numbers related to experiments. It is then often
(but not always) meant the Reλ from Eq. (9.12).
2) With an integral length scale (page 117 and 136):

u′ L
Re L = . (9.13)
ν
The length scale L can be obtained from experiments or from a simulation. Or it can be taken as a
characteristic “large” length scale for the turbulence. We could then also write
u ′ ℓ′
Reℓ′ = , (9.14)
ν
which is a characteristic Reynolds number for the large-scale turbulence (Sec. 8.1).
3) With a turbulence viscosity: The Reynolds number equals the ratio of turbulence viscosity to
molecular viscosity,
νt
Ret = . (9.15)
ν
This is the ratio between convective transport with the turbulence (“turbulence diffusion”) and
molecular diffusion.
4) With k and ε :
In turbulence models with an ε-equation, a Reynolds number is often defined as
k2
ReT = . (9.16)
νε
This Reynolds number is used in empirical functions of models for weak turbulence (low Reynolds
number). The first was by Jones and Launder (1972). See also Patel, Rodi and Scheuerer (1985).
The relation between Reλ and ReT we can find by using λ2 = 10νk/ε and u ′ 2 = 32 k. This gives
2
3k · 10νk 20 k 2
Re2λ = = = 20
3 ReT . (9.17)
ν2ε 3 νε
In a k-ε model, the turbulence viscosity is
k2
νt = Cµ f µ . (9.18)
ε
In the “standard” model for high Reynolds numbers is f µ = 1. In models for low Reynolds numbers,
fµ is an empirical function of ReT . In the model of Jones and Launder (1972) (see Launder and
Spalding, 1974) is f µ = 0.98 for ReT = 6000 (Reλ = 200). In Launder and Sharma (1974), f µ is a
little changed, so that f µ = 0.99 for ReT = 870 (Reλ = 76). For higher Reynolds numbers, fµ ≈ 1.
We see that for a high Reynolds number, ReT is proportional to Ret , although they are not equal:
Ret = Cµ fµ ReT .
As an approximation we can regard the “large” length scales L, ℓ, ℓ′ , k 3/2 /ε and ℓe for proportional
and of the same order of magnitude, so that the Reynolds numbers are proportional: Re L ∼ Reℓ ∼
Ret ∼ ReT ∼ Re2λ .
5) With the Kolmogorov length scale (page 130):

u′η
ReK = . (9.19)
ν
This combination of one scale for the small eddies (η) and one for the large (u ′ ) we can meet in the
3/2
combustion literature (see Sec. 10.2). When we recall that ℓe /η ∼ Reλ (cf. page 126) and regard
1/2 1/4
ℓe ∼ ℓ ∼ ℓ′ , we can find that ReK ∼ Reλ ∼ Reℓ′ .
High and low Reynolds numbers
It is not always easy to tell whether a flow has a “high” or a “low” Reynolds number. This depends
on the context or circumstance.
The Reynolds number Reλ tells something about the difference between the Kolmogorov length
scale (η) and the length scale for the most energy-rich range of a energy spectrum (ℓe ). In isotropic
3/2
turbulence, ℓe /η ≈ 0.1Reλ (Hinze, 1975:225).
The largest Reynolds numbers we find in flows in the atmosphere and in the ocean. Here, external
dimensions are very large. In the atmosphere, the Reynolds number Reλ can be 104 (Wyngaard,
1989). This is in any case a high Reynolds number. (See also the example calculation at page 129.)
In flows in laboratories and technical devices, the situation is different. There, a Reynolds number
Reλ of 150–200 is often regarded as “high”. In particular cases, the Reλ can be more than 103 .
Notice also that Reλ increases with the pressure but is reduced with the temperature, Eq. (9.27).
9.4 Alternative methods and models for numerical simulation of turbulence 141
The other two turbulence Reynolds numbers can be calculated from Reλ .
Example: For Reλ = 100 is ReT = 1500 and Ret = 135 (with f µ = 1). Thus is νt = 135ν. In this
case, the turbulence viscosity is much larger than the molecular viscosity. Seen in this way, we can
say that this Reynolds number is “high”. It is, however, not so high that we can say straightforwardly
that the small-scale range is independent of the large-scale range of the spectrum. We do not have
an inertial subrange in the energy spectrum (page 128). This means that we can not, without further
discussion, use a large-eddy simulation (page 142) in such a case.
Notice: The term “low Reynolds number” is also used in a totally different context. Certain flows
have a Reynolds number (for the main flow) much below unity. Flows where the viscous forces are
strongly dominant are called creeping flows or creeping motions, see e.g. Schlichting (1979:112f).
Examples are layers of lubrication oil (or grease) between two engine parts, the motion in salt sheets
or domes that penetrates upwards in the rock strata (diapirism) and motions in the mantle under the
Earth’s crust.
9.4 Alternative methods and models for numerical simulation

of turbulence
The three-dimensional energy spectrum E represents the turbulence energy in one point of space and
time (see Sec. 8.2). It comes from a two-point average value. On the basis of the energy spectrum, we
can distinguish between three different principles of numerical simulation. By numerical simulation
we put a grid over the calculation domain. This grid has a mesh width, or characteristic length scale,
1m .
Mean-value simulation
The numerical grid is sufficiently fine to catch up variations in the mean flow. We can find mean
values (statistical moments) like u, u ′i u ′j , ε and so on. The grid can have a mesh 1m that is larger
than the large turbulence eddies, but 1m is independent of ℓe . Here, all effects of the turbulence
have to be modeled.
When we solve an equation for the turbulence energy, we determine the area under the E-curve.
If we solve an ε-equation, we get a length scale. We can then find the width of the spectrum, for
instance the ratio of ℓe to η.
Mean-value simulation is the most common computation method for turbulent flows. Such meth-
ods are also called moment methods or statistical methods since we resolve values for statistical
moments.
The mean value can vary with time, cf. page 34. Preferably, the timescale for this variation should
be much larger than the timescale for the turbulence. If these scales drift into each other, it becomes
more difficult to give a clear interpretation of the results. Someone will point towards large-eddy
simulation (see below) as the solution of the problem. However, also in this method there should
preferably be a distinction between resolved and modeled motions.
Direct simulation
Turbulent flows can be simulated numerically by solving the equations of motion directly, without
any turbulence model. One then uses a grid that is so fine that it catches up the smallest eddies. Thus,
1m < η (ideally, but at least 1m ≤ η). In a direct simulation, one uses the equations of motion
(fundamental equations) directly in the same way as for laminar flow. No model for turbulence is
required. The simulation gives instantaneous values, u i , T , etc. These can be averaged and one can
determine fluctuations and correlations.
Direct numerical simulations require huge computer resources and in the beginning (mid 1970s) this
was possible only for very simple flows such as isotropic decaying turbulence. Later, simulations
have been made of shear flows such as a plane channel (Kim, Moin and Moser, 1987) and a boundary
layer (Spalart 1988). Such simulations give a basis for testing of models, like experimental data,
and they provide quantities that are hard or impossible to measure. The results can be used in
addition to or instead of measurements to test models for mean-value simulations. Preliminarily,
direct simulation is restricted to flows with a low Reynolds number, but the limits are extended in
accordance with the development in computer technology.
Moin and Mahesh (1998) give an overview of the development of direct simulation (without chemi-
cal reactions) up to now. They discuss some numerical topics and show the relation to experiments
and modeling. Direct simulation, in particular of turbulent combustion, we will come back to in
Sec. 10.10.
Large-eddy simulation
The grid has a mesh width that is less than the large eddies but larger than the smallest eddies, thus
ℓe ≫ 1m ≫ η. This resolves values for the large but not for the small eddies. The idea is that
the small eddies are easier to model. At a high Reynolds number, there is a big difference between
“large” and “small” eddies. The small are independent of the mean flow and of the large eddies.
This calculation method is called large-eddy simulation. The small scales have to be modeled, and
the term sub-grid modeling is also used. Reynolds (1989) discusses benefits and drawbacks of such
methods. For an introduction to the subject, see for instance Härtel (1996) and Speziale (1998).
3/2
For isotropic turbulence, ℓe /η ≈ 0.1Reλ (Hinze, 1975:225). In the atmosphere and in ocean
currents (see example at page 129), we can find Reλ = O(104 ) (cf. page 140). This gives ℓe /η =
O(106 ), which means a big difference between large and small scales. In flows in laboratories and
technical devices, Reynolds numbers Reλ at 150–200 are regarded as “high”. This gives ℓe /η =
O(103 ) and hence, it is not sure that the small scales are independent of the large.
RNG or dynamic turbulence models
In this context, it is suitable to include a technique called dynamic models or RNG-techniques (RNG:
“renormalization group (theory)”). This is a method which, like the methods above, is based on the
equations of motion. These equations are treated with other mathematical operations, and other
presumptions are made. It will lead too far to discuss the matter here, so interested readers have to
9.5 Simple estimates for turbulence quantities 143
find the literature themselves.

RNG can be used for developing or formulating models both for mean-value simulation and for
large-eddy simulation. It seems that mainly k-ε models are tested. In spite of many differences in
the development, the final result looks very much like the usual k-ε-model. The model constants
have got somewhat different values, and these are based on a different reasoning. An objection is
that it – in spite of a firmer theoretical basis – still is a k-ε model with a turbulence viscosity. Just this
is the main problem for all two-equation models, not the basis for the constants. (See the discussion
on page 46 on k-ε models and of the article of Launder and Spalding (1974).)
The theory was first used for turbulence modeling by Yakhot and Orszag (1986). A critical discus-
sion is given by Smith and Reynolds (1992) and parts of the model is reformulated by Yakhot and
Smith (1992). For further use and development of this approach, it will pay off to search the newer
issues of the relevant journals. Right now it seems to be one of the popular subjects of turbulence
research; – time will show what it leads to.
Transport equation for the probability density
The overview here also has to include transport equations for multi-dimensional probability density
functions. We will have a closer look at this in Sec. 10.9.
9.5 Simple estimates for turbulence quantities

In many instances, it will be useful to do simple estimates for turbulence
quantities. The purpose may be to find preliminary approximations. It may also
be initial values for a numerical simulation.
These estimates must be used with care. They can be misleading or give
completely wrong figures. Eventually, you will gather your own experience
within your field of work.
In addition to those mentioned here, the thermodynamic relations and empirical correlations found
in books on thermodynamics, heat and mass transfer, fluid mechanics, combustion theory, etc., will
of course also be relevant.
Velocity scale for the turbulence – turbulence intensity, turbulence energy
We often use a characteristic velocity scale u ′ for the turbulence. When we know the turbulence
energy, we can define u ′ 2 = 23 k, or alternatively u ′ 2 = k. As an estimate, one is as good as the
other.
In experiments, one has often measured the mean value u 1 and the normal stress u ′12 along the flow.
In the terminology of statistics, this is the variance of the flow velocity. The other normal stresses
(and hence, the turbulence energy) are more rarely found. – They are more demanding to measure.
The measurement of the fluctuations is often given as a turbulence intensity, u ′ /U . Here, u ′ is the
square root of the normal stress u ′1 2 , thus the standard deviation. U is the flow velocity; often the
mean velocity u 1 of the same location, but it can also be at a centerline axis or the velocity in an
inlet or outlet. The given turbulence intensity is not always clearly defined.
When we do not know the turbulence intensity, we have to guess. A value u ′ /U ≈ 0.1 or 0.05 is
often a usable attempt as long as we do not have special reasons to believe that the turbulence is
particularly strong or weak. Such an estimate does of course require that we have a main flow with
a flow velocity. It can hardly be used in an approximately homogeneous mixing chamber, cf. the
example at page 130.
Reynolds stresses
1 ′2 ′2 ′2
The turbulence energy is defined as k = 2 (u 1 + u 2 + u 3 ). When nothing else is known, one
2
easily grabs to isotropy and says that u ′1 2 ≈ u ′2 2 ≈ u ′32 ≈ 3 k. However, a better estimate is to set
u ′1 2 ≈ k. The other two are roughly equal; possibly u ′32 is a little larger than u ′22 when x 2 is normal
to and x 3 is parallel to a wall. This follows from experimental data, cf. e.g. Ertesvåg (1991:24).
p
The shear stress can, as an estimate, be set to −u ′1 u ′2 ≈ 0.3k; where 0.3 = Cµ , cf. Eq. (4.20).
Dissipation
Above, we have in several instances met expressions like

u′ 3
ε≈A , (9.20)
ℓ′
see Eq. (3.8) page 52, Eq. (4.18) page 74 and Eq. (8.13) page 128. Later, we will also use a similar
expression in Magnussen’s combustion model, EDC, page 175.
In Eq. (9.20), u ′ is a characteristic turbulence velocity scale and ℓ′ is a characteristic length scale.
See above about the velocity scale. The length scale is some or other “large” length scale. Often,
we do not have anything else than the outer dimensions, for instance a pipe diameter or a distance
to a wall. Nevertheless, an expression like Eq. (9.20) gives only a rough approximation, an order of
magnitude.
The numerical factor A can have a value of 0.2 (see Sec. 9.2) or, to make it more simple, a value of
unity. (We can also say that we include this numerical factor into the length scale.)
In some cases, we know the power fed into the system, that is, the production of turbulence en-
ergy (work per units of time and mass). If the system is approximately in steady state, we set the
dissipation equal to the production,
power feed
ε ≈ Pk = . (9.21)
mass
This is used in the example on page 130. With Eq. (9.21) for ε, we can use Eq. (9.20) for u ′ or ℓ′ .
Turbulence Reynolds number
A turbulence Reynolds number can be defined from the Taylor microscale λ, Eq. (7.43), as
u′λ
Reλ = . (9.22)
ν
The length scale can be estimated from Eq. (7.42), λ2 = 15νu ′ 2 /ε or λ2 = 10νk/ε. We must then
first have estimates for u ′ and ε. We will return to turbulence Reynolds numbers in Sec. 9.3.
Turbulence viscosity
We regard the turbulence viscosity as a product of a velocity scale and a length scale,
νt = u ′ · ℓ′ . (9.23)
These scales are the same as in Eq. (9.20), and the approximation here is as rough as there. Using
Eq. (9.20) with A = 1, we can alternatively introduce a numerical factor 0.2 in Eq. (9.23), cf.
Sec. 9.2.
The ratio between turbulence viscosity and molecular viscosity is also a Reynolds number, see Sec.
9.3. Such an estimate can, among other things, be used to evaluate whether we have a “high”
Reynolds number. It may, for instance, indicate whether we can use a “standard” k-ε model or not.
Kolmogorov scales
When we have estimated the dissipation and know the viscosity, we can find the Kolmogorov scales
(see example at page 130). The length scale η = (ν 3 /ε)1/4 and the timescale τ = (ν/ε)1/2 tell
something about how fine the resolution of the experimental apparatus has to be in order to catch
up the finest turbulence structures. It also tells something about the required meshing for a direct
numerical simulation (pages 142 and 130).
Viscosity: variation with pressure and temperature
The “large” scales, that is, the velocity scale u ′ , the mixing lengths and integral scales L, ℓ, are not
directly influenced by pressure and temperature. They will be functions of the external dimensions
of the main flow (which certainly to some extent will be influenced by pressure and temperature).
Hence, we can also calculate the dissipation ε (∼ u ′ 3 /ℓ) and the timescale θ = k/ε independently
of pressure and temperature.
For an estimate, we can assume that the (molecular) kinematic viscosity varies with pressure and
temperature as
ν ∼ P −1 T 3/2 . (9.24)
From this follows that the Taylor scale, Reynolds numbers and Kolmogorov scale vary as
λ ∼ ν 1/2 ∼ P −1/2 T 3/4 , (9.25)
−1 −3/2
Ret ∼ ν ∼ PT , (9.26)
′
uλ
Reλ = ∼ ν −1/2 ∼ P 1/2 T −3/4 , (9.27)
ν
η ∼ ν 3/4 ∼ P −3/4 T 9/8 . (9.28)
The highest temperatures in the combustion zone rarely lie more than 6–7 times over the temperature
of “cold” gas. The pressure can have larger variation – or it can be constant.
The quantities from these latter expressions should be used with some care. In some cases, it might
be a better approximation to use e.g. a Reynolds number and a Kolmogorov scale for a cold flow or
for a temperature that represents a larger volume than the reaction zone, see Sec. 8.4.
Spreading of a substance
In the example on page 129, we calculated at the spreading of an algae swarm in an ocean current.
The approach can be used to do estimates of the spreading of passive scalars in ocean and air currents.
Convective heat transfer
The heat transfer from a wall to a flow can be written as

qw = h c (Tw − T g ), (9.29)
where h c is the convective heat transfer coefficient [W/(m2 K)] (see in a book on heat transfer), Tw is
the wall temperature and T g is the temperature some distance into the boundary layer, preferably in
the logarithmic layer (see Sec. 5.6).
The heat transfer coefficient depends on the flow in the boundary layer, and as an estimate we
can set the numerical value equal to the numerical value of the momentum in the log-region, thus
h c /(W/(m2 K)) ≈ (ρu 1 )/(kg/(m2s)). For gases, the density is close to 1 kg/m3 i many cases.
Spreading of flames
Here, we take the following chapters in advance. For simple flames the propagation velocity can be
approximated by
UB ≈ 2u ′ . (9.30)
Byggstøyl and Magnussen (1983) developed this relation by using the EDC and the fuel mass frac-
tion equation for a plane, stationary and one-dimensional flame. The approximation agrees reason-
ably well with experimental data. Bradley (1992) also gives an overview of similar approximations
based on different models. Most of them give numerical factors of 1–2 in Eq. (9.30).
Example: Pipe flow
Air flows in a pipe with radius R = 0.1 m and a volumetric flow of 125 liter/s, which gives an
average velocity over the cross-section (bulk velocity) of u b = 4 m/s. The center-line velocity is a
little larger (approximately 1.2 times) than the bulk velocity. Viscosity: ν = 1.6 · 10−5 m2 /s.
The Reynolds number, Re D = u b D/ν = 5 · 104 , is 10 times larger than the critical value and hence,
the flow is turbulent.
In the center of the pipe, the turbulence is the least intense, but there is still some turbulence. We
can estimate u ′ /u b ≈ 0.05, so that u ′ = 0.2 m/s and k = u ′ 2 ≈ 0.04 m2 /s2 .
The length scale, in lack of any other knowledge, we set equal to the radius, ℓ′ ≈ R = 0.1 m.
Dissipation: ε ≈ u ′ 3 /ℓ′ ≈ 0.08 m2 /s3 .
Kolmogorov length scale: η = (ν 3 /ε)1/4 ≈ 5 · 10−4 m.
Turbulence viscosity: νt ≈ 0.2u ′ · ℓ′ ≈ 0.004 m2 /s (= 250ν), or νt ≈ 0.1k 2 /ε ≈ 0.002 m2 /s
(= 125ν).
Taylor microscale: λ = (10νk/ε)1/2 ≈ 0.01 m.
Reynolds numbers: Ret = νt /ν ≈ 250 and Reλ = u ′ · λ/ν ≈ 125.
Near the pipe wall, with distance y = 0.004 m: We use relations developed for the log-region.
Afterwards, we have to check whether the location actually lies in this region.
Near the wall, the turbulence is more intense than in the center, and we estimate u ′ /u b ≈ 0.1, or
u ′ ≈ 0.4 m/s and k ≈ u ′ 2 ≈ 0.16 m2 /s2 .
The shear velocity at the wall: u τ ≈ (u ′1 u ′2 )1/2 ≈ (0.3k 1/2 ) ≈ 0.22 m/s (cf. Eq. (4.20) page 74).
As an alternative, we can use empirical data for the Darcy friction factor f = 8τw /(ρu 2b ). With Re D
as above, we can find f ≈ 0.02 from a book on fluid mechanics (e.g. White, 1994:309f). This should
agree well with experimental data for smooth pipes. We then get u τ = ( f /8)1/2 u b ≈ 0.2 m/s.
Now we can determine y + = yu τ /ν ≈ 55 (alternatively 50 with the alternative estimate for u τ ).
This is in the log-region.
Dissipation: ε ≈ u 3τ /(κy) ≈ 6.7 m2 /s3 .
Kolmogorov length scale: η = (ν 3 /ε)1/4 ≈ 1.6 · 10−4 m.
Turbulence viscosity: νt = κyu τ ≈ 3.5 · 10−4 m2 /s (= 22ν).
Taylor microscale: λ = (10νk/ε)1/2 ≈ 0.020 m.
Reynolds numbers: Ret = νt /ν ≈ 22 and Reλ = u ′ · λ/ν ≈ 500.
Calculations for a pipe with the standard k-ε model gave these results:
For the center line: k = 0.05 m2 /s2 , ε = 0.1 m2 /s3 , νt = 2 · 10−3 m2 /s (= 125ν), λ = 0.009 m.
√
For the location y = 0.0038 m: y + = 50.3, u = 3.0 m/s, k = 0.136 m2 /s2 , u ′ /u b = k/u b =
0.09, ε = 6.2 m2 /s3 , u τ = 0.205 m/s, νt = 2.7 · 10−4 m2 /s (= 17ν), u ′1 u ′2 = 0.0414 m2 /s2 .
√
For the location y = 0.0064 m: y + = 83.8, u = 3.25 m/s, k = 0.130 m2 /s2 , u ′ /u b = k/u b =
0.09, ε = 3.6 m2 /s3 , νt = 4.25 · 10−4 m2 /s (= 27ν), u ′1 u ′2 = 0.0394 m2 /s2 .
Averaged over the whole cross-section: k = 0.10 m 2 2 2 3
2 −4 2 ′
√ /s , ε = 1.9 m′ /s ′. 3 These quantities give
νt = 0.09k /ε = 5 · 10 m /s (= 31ν), u /u b = k/u b = 0.08, ℓ = u /ε = 0.017 m ≈ 0.2R.
Example: Mixing chamber
Equal amounts of two gases are to be mixed in a mixing chamber. Inside the chamber, fans are
stirring the gases. The two gases have the same density, ρ = 1 kg/m3. The outlet mass flow of
mixture is ṁ = 0.1 kg/s and the chamber has a volume V = 0.1 m3 . We will estimate how much
power that has to be fed by the fans in the chamber.
The residence time becomes θb = ρV /ṁ = 1 s. As a length scale we can use ℓ′ ≈ V 1/3 ≈ 0.4 m.
The turbulence timescale is θ = ℓ′ /u ′ (= k/ε). We can use θb = θ as a criterion. In this time, the
gases are mixed and we estimate the turbulence velocity u ′ ≈ ℓ′ /θ ≈ 0.4 m/s.
Dissipation: ε ≈ u ′ 3 /ℓ′ ≈ 0.2 m2 /s3 . Provided power (work rate): Ẇ = ερV ≈ 0.02 W.
Here, we have assumed that the kinetic energy is equal for the outflow and the inflow of the chamber.
Since this is a rough estimate, we may wish to be on the safe side: If we say that the criterion should
be θb = 5θ , does this affect the power input? We then find that u ′ becomes 5 times larger, the
turbulence energy 25 times larger, and the dissipation and power input 125 times larger.
Example: Burner
For a burner, the typical length (e.g. diameter) can be D = 0.1 m, the velocity U = 10 m/s, and the
viscosity ν = 10−5 m2 /s (cf. Sec. 8.4 and page 146).
We estimate u ′ ≈ 1 m/s, ℓ′ ≈ D, θ = ℓ′ /u ′ ≈ 0.1 s; Dissipation: ε ≈ u ′ 3 /ℓ′ ≈ 10 m2/s3 .
Kolmogorov scales: η = (ν 3 /ε)1/4 ≈ 10−4 m, τ = (ν/ε)1/2 ≈ 10−3 s.
Convection time for Kolmogorov scale (page 137): τU = η/U ≈ 10−5 s.
Ratio of timescales: θ/τ = 102 , and θ/τU = 104 .
Example: Fire
For a medium-size fire, a typical length might be ℓ = 3 m, velocity U = 3 m/s, and viscosity
ν = 10−5 m2 /s (cf. Sec. 8.4 and page 146).
We estimate u ′ ≈ 0.3 m/s, θ = ℓ′ /u ′ ≈ 10 s;
Dissipation: ε ≈ u ′ 3 /ℓ′ ≈ 10−2 m2 /s3 .
Kolmogorov scales: η = (ν 3 /ε)1/4 ≈ 6 · 10−4 m, τ = (ν/ε)1/2 ≈ 3 · 10−2 s.
Convection time for Kolmogorov scale (page 137): τU = η/U ≈ 2 · 10−4 s.
Ratio of timescales: θ/τ = 3 · 102, and θ/τU = 5 · 104 .
Chapter 10
Turbulent combustion
10.1 Premixed and non-premixed flames
Flames are often classified according to whether the reactants are mixed before or inside the com-
bustion chamber or flame zone.
In a premixed flame the reactants are perfectly mixed before the chemical reaction.
In a non-premixed flame the reactants are supplied separately and mixed in the combustion cham-
ber.
The term diffusion flame is often used as a synonym for non-premixed flame. But we can also restrict
the definition and say that: In a diffusion flame the reactants are induced separately and mixed by
diffusion in the reaction (flame) itself.
The classical example of a premixed flame is the Bunsen burner. This is a simple laminar laboratory
flame, which is intensively analyzed in the literature. Another, more technical example is the Otto
engine. Fire/explosion after a gas leakage is premixed combustion. Propagation of a flame in a
mixture of reactants is discussed in quite a few combustion books.
An example of a non-premixed flame is a fuel jet flowing from a pipe into air. Another example is
an ordinary candle. Also these cases are elaborately treated in the literature. To the non-premixed
flames we have to count the Diesel engine and most other combustors.
It is easy to see that the classification into diffusion flames and premixed flames is an idealization.
Most practical flames will be something in-between.
We can look at the simple non-premixed flame of a fuel jet flowing into air, Fig. 10.1. At some
conditions, the flame will be lifted. That is, the reaction (flame) takes place at some distance above
the outflow. The reactants are then partly premixed before the reaction, and we do not have a pure
diffusion flame. However, we can still talk about a non-premixed flame.
150 10 Turbulent combustion
(a) Re = 10000 (b) Re = 13600
Figure 10.1: Photo of two turbulent, non-premixed flames. The fuel, acetylene, flows out
of a straight pipe with internal diameter 0.84 mm. The Reynolds numbers are given for the
outlet of the pipe, which is in the lower edge of the picture. The flame in (b) is somewhat
lifted from the outlet. The photos are approximately in the actual scale. (Photos: Rune N.
Kleiveland)
A premixed laboratory flame with strong turbulence is sketched in Fig. 10.2. The flame is stabilized
(anchored) by a parallel flow of hot gas. In a simple straight channel, it is practically impossible to
set up a stable, turbulent flame – and hardly a laminar flame. The setup shown can give a flame with
a relatively intense turbulence, a Reynolds number Reλ of order 103 (Ret = O(104 )). See details
and more references to experiments in Borghi (1988) and simulation in Gran (1994:147f).
For some turbulent premixed flames, the inflow conditions may be particularly important, both in
experiments and in simulations. In many set-ups, the flame will be positioned differently with differ-
ent distributions of turbulence in the inlet. This means that we preferably use experimental data from
the inlet as inlet conditions for the computations. Otherwise, the case that is simulated may become
something quite different from the experimental case that it is compared to. For real-life problems,
such experimental data are often not available. Then, it is important to make the inflow as realistic as
possible. In any case, one has to investigate how much the inlet conditions affect the computational
results. Similarly, when developing and designing devices with premixed combustion, not only the
geometric shape, but also the inlet flow conditions, may influence the behavior of the combustion.
10.2 Characterization of turbulent flames 151
Air
Fuel Product
Hot gas
Figure 10.2: Premixed turbulent flame
10.2 Characterization of turbulent flames

Chemical reactions and turbulent flow are each by their own very complex phenomena. Hence, it is
obvious that turbulent flames will become quite an intricate matter. Here we will look at some major
issues in this picture and find some characteristic quantities for combustion (on such quantities, see
Sec. 2.8). As before in the book, this means to do rough simplifications. For its use, however, this
can be both justifiable and useful.
Turbulence scales
Previously, we have looked at characteristic scales for turbulent flow. To get an overview, we sum-
marize:
For the large turbulence structures, we have the velocity scale u ′ , the length scale ℓ′ and the timescale
θ = ℓ′ /u ′ . From these we get a Reynolds number Reℓ′ = u ′ ℓ′ /ν. These scales are characteristic
for the motions that are of main importance for transport with the turbulence. Furthermore, we have
1/2
the Taylor length scale λ that gives the Reynolds number Reλ = u ′ λ/ν ∼ Reℓ′ , see Sec. 9.3. In
non-premixed flames, θ is a characteristic time for mixing of reactants; and for mixing heat (i.e.
products) with the reactants in premixed flames.
For small turbulence structures, we have the Kolmogorov scales (page 130) for velocity v, length η
and time τ . According to the definition, we can write v = (ν/τ )1/2 and η = (ντ )1/2. Actually, we
need only one of these in addition to the viscosity (– since all are functions of the dissipation and the
viscosity). These scales are characteristic for the motions that give a large dissipation of turbulence
energy, that is, with large viscous forces. There, also the molecular mixing will be the greatest. With
the Kolmogorov length scale, we can define a Reynolds number ReK = u ′ η/ν (page 140).
We may keep in mind (from page 126) that the ratio between “large” and “small” turbulence scales
3/2 3/4 1/2
is ℓ′ /η ∼ Reλ ∼ Reℓ′ . For the timescales we can find θ/τ ∼ Reℓ′ . We also can express
1/2 1/4
ReK ∼ Reλ ∼ Reℓ′ .
Laminar-flame scales
Characteristic quantities for laminar premixed flames are propagation velocity or flame speed u L ,
flame thickness δL , and chemical timescale τc . As a simple image we can think of a flame front with
extension δL that is moving with a velocity u L through a mixture of reactants. The timescale is then
a residence time in the flame zone, or a reaction time.
These quantities depend on the reaction rate and diffusivity, and we can write u L ∼ (D/τc )1/2 and
δL ∼ (Dτc )1/2 . (Compare with the med Kolmogorov scales above.) The timescale τc can be defined
−1
from the reaction rate of the fuel. We get the relation τc ∼ Rfu , which will depend on the species
taking part, the concentrations and the temperature.
In non-premixed flames (diffusion flames), the flame speed or the flame thickness do not have a
physical meaning. The timescale however, has a meaning and, in turn, we can define the velocity
scale u L and the length scale δL from the τc as above.
In many instances, the viscosity and the diffusivity are approximately equal – or they vary similar to
each other. This gives u L ∼ (ν/τc )1/2 and δL ∼ (ντc )1/2 . If we like, we can list the Schmidt number
Sc = ν/D as an additional parameter.
We have seen that time- and length scales for the turbulence are spread over a wide spectrum. In
order to simplify and make an easier overview, we found some few scales that are characteristic for
the turbulent motions. Similarly, for flames: There are a lot of time- and length scales spread over
a wide range (cf. Sections 10.3 and 10.10). Here we want to find some few scales that still can say
something about the combustion.
Laminar flames are thoroughly discussed in the literature. Here, we will be content with the as-
sumption that a reaction has one characteristic timescale, one velocity scale and one length scale.
When we do so – as for instance in the Borghi diagram (see below) – we have said that there is a
fixed or given relation between the “large” flame scale and a “small” reaction timescale of the fastest
reactions.
Dimensionless groups
For turbulent flames, the ”laminar” scales are used as characteristic scales. This can be done since
they are defined on the basis of the reaction rate and not of the geometry of the flow.
With this collection of characteristic scales, we can put up some dimensionless groups. In addition
to the Reynolds numbers above, we have the ratios between length scales: η/δL , ℓ′ /δL ; ratio of
velocity scales, u ′ /u L ; and ratios between timescales, Da = θ/τc and DaK = τ/τc . The latter
two are called Damköhler numbers (cf. Sec. C.5); the last one is also called Karlovitz number, Ka
(= Da−1K ). For a fast reaction (small τc ), the Damköhler number is large; for a slow reaction the
Damköhler number is small.
With some operations, one can find relations between the groups:
u′ δL −1/4 δL
∼ Reℓ′ · ′ ∼ Reℓ′ · (10.1)
uL ℓ η
10.2 Characterization of turbulent flames 153

τ η 2 −1/2
DaK = ∼ ∼ Da · Reℓ′ (10.2)
τc δL
′ 2 2
θ ℓ′ u L ℓ −1 η 1/2
Da = ∼ ′
∼ Reℓ′ ∼ Reℓ′ (10.3)
τc δL u δL δL

ℓ′ 3/4 η
∼ Reℓ′ (10.4)
δL δL
A simple dimensional analysis shows that we need only two independent dimensionless groups;
these must include
– two “large” turbulence scales, u ′ and ℓ′ , – or one of them together with a Kolmogorov scale
(v, η or τ );
– one chemical scale (u L , δL or τc );
– viscosity ν.
Furthermore, we can include the Froude number Fr = U 2 /(g L) (alternatively the Richardson num-
ber, page 238) that characterizes the buoyancy forces. For large Froude numbers, we can neglect
efficiencies of buoyancy; that is for Fr > 102 (Goulard, Mellor and Bilger, 1976), possibly for lower
values as well.
Flame types
One can readily imagine that flames where δL ≪ η differ much from flames where δL ≫ η or where
δL ≈ ℓ′ . The details of this picture have to be found by experimenting with flames. Different authors
have chosen different parameters for characterizing the different flame regimes. Borghi (1988) has
drawn diagrams with u ′ /u L and ℓ′ /δL as the axes; Williams (see e.g. Libby and Williams, 1994:56)
has chosen Damköhler number Da and Reynolds number Reℓ′ as the axes; Bray (1980:130) has
u ′ /u L and Reλ along the axes; see also Goulard, Mellor and Bilger (1976). All these explain and
account for the flame structure of premixed flames. In addition, Borghi gives a discussion of the
structure in turbulent non-premixed flames. Bilger (1980) also gives a well readable introduction to
theory and models for turbulent non-premixed combustion.
We can imagine a laminar flame front as a thin, flat reaction zone. With fast reactions (small τc , δL ;
large u L ) and weak turbulence (large θ , τ , ℓ′ ; small u ′ ), the reactions occur in sheets as in laminar
combustion. The turbulence will bend the reaction zone so that the flame becomes wrinkled.
With larger fluctuations, the wrinkles become folds, and pockets or islands of reactants can be
formed. The flame sheets are torn up. One can regard this as many small flames, flamelets. The
reaction region – the flame – becomes thicker.
With intense turbulence (small θ , τ , ℓ′ ; large u ′ ) and slow reactions (large τc , small u L ), the reactions
are distributed over a wide reaction region. The flame fills the entire combustion space and we have
a well stirred reactor. An idealized limiting case is called a perfectly stirred reactor.
One-dimensional non-premixed flames (diffusion flames) will behave in a similar sense. The limits
between the different regimes are less clear than for premixed flames. Figures 10.3 and 10.4 show
Borghi’s (1988) depiction of the conditions for the different types of flames.
Throughout a combustor, furnace or other device, the flow, concentrations and temperature will vary
much. Thus, we may see several flame types together.
′
u
log
uL
Thick
Thickened flames
D
flames
a
es
am
d fl
kle
ri n
dw
t
ns
ene
co
ck
hi
=
T st
con
τc
θ/
=
τ /τ c
=
=
η
D
aK
δL
rD
Laminar ≈ ηo
δL Wrinkled flames with
flames pockets of reactants
0 u′ = uL
Wrinkled
′
flame ℓ
Re log
ℓ δL
′
=
ℓ′
co
Re
ns
t
Figure 10.3: Different regimes of turbulent premixed flames, after Borghi (1988).
The classification can be useful in different instances:

– Simplified relations for estimate calculations will be valid for a restricted set of conditions.
With such a classification, one can stay close to the presumptions made in the simplification.
– Advanced models aim at capturing many different cases. Then, they have to be tested against
theory and experimental data in the different regimes that are identified.
– Each of the regimes of the diagrams represents a flame type that is more ready to be analyzed
and understood than if one tries to grasp turbulent flames as a whole.
Flames and turbulence
For sake of order, it can be added that the reaction (in particular the heat release) will affect the tur-
bulence, not only vice versa. Hot gas will rise up, and the buoyant flow will make turbulence. On the
other hand, expansion may damp the turbulence. Moreover, hot gas will get higher viscosity, which
damp the turbulence. The local Reynolds number becomes less in a flame than in a corresponding
flow without reaction, since the density is reduced and the viscosity increases.
10.3 Chemical kinetics, extinction 155
′
u
log
uL
Thickened flames
D
a
es
fl am
t
all
ns
sm
co
rn
=
To st
con
τc
θ/
=
τ /τ c
=
=
η
D
aK
δL
rD
Laminar ≈ ηo
δL Wrinkled and stretched
flames small flames
0 u′ = uL
′
ℓ
log
Re
δL
ℓ
′
=
ℓ′
co
Re
ns
t
Figure 10.4: Different regimes of turbulent non-premixed flames (diffusion flames), after
Borghi (1988). The figure looks much like Fig. 10.3 for premixed flames. Important differ-
ences are that the flame scales δL and u L do not have the same physical meaning, and that
Da and DaK have other critical values. Furthermore, the line u ′ = u L do not constitute any
particular limit in the diagram for non-premixed flames.
10.3 Chemical kinetics, extinction

Chemical kinetics
This subject we have touched briefly above (Sec. 1.8). Each elementary reaction – and there are
thousands – has its own properties or its own kinetics. By itself this is a large and comprehensive
subject. There are well readable expositions other places, for instance Warnatz, Maas and Dibble
(1999). In his dissertation, Inge Gran (1994) (also Gran and Magnussen, 1996b) has explained how
chemical kinetics can be linked to combustion simulations with Magnussen’s combustion model
(EDC).
Here, we will pick up one important aspect: An overall reaction is the sum of thousands of elemen-
tary reactions. The chemical timescales for the different elementary reactions can range over a wide
spectrum, possibly several decades (orders of magnitude). The slowest reactions (among others in
NO-formation) can have timescales of many seconds, while the fastest have timescales at 10−10
seconds (Warnatz, Maas and Dibble, 1999:101f).
Above, we have seen that length scales and timescales for turbulence are distributed over some
orders of magnitude. Rarely, there are more than 4–5 decades, in certain cases more (cf. pages
140 and 142). Timescales for molecular transport can be shorter, so that we lengthen the spectrum
with a few more decades. The timescales that dominate the transport – that is, “large” turbulence
timescales, which we resolve from a turbulence model – can often be found within 2–3 decades.
In other words, the chemical timescales will be distributed over a much wider spectrum than the
turbulence transport.
Fast reactions may always take place provided that the reactants are mixed molecularly. Slow reac-
tions, e.g. nitrogen reactions, might go on long after the actual flame zone; or they are terminated
before reaching equilibrium.
Extinction
Extinction of flames is a matter of local mixing of chemical species, local heat transfer and chemical
kinetics.
The local transport processes are related to gradients of concentration and temperature. Therefore,
2 2
the scalar dissipation rates (Sec. 5.5 and 8.5) εY ′ 2 ∼ D ∂Yk′ /∂ x j and εT ′ 2 ∼ α ∂ T ′ /∂ x j are
k
quantities that can characterize these processes in a turbulent flow. Furthermore, concentration,
temperature and chemical timescales will be decisive for whether the species react.
Extinction of flames, non-premixed and premixed, is further discussed by Warnatz, Maas and Dibble
(1999:195f,208f,258f). Sloan and Sturgess (1996) give an overview and discuss different models for
extinction in turbulent flames. In Sec. 11.7, we will look closer at how this is done in Magnussen’s
combustion model (EDC).
10.4 Combustion models – classification

Combustions models are a manifold issue. Any attempt on classification will give “grey zones”
between the classes. And one will find models that not really fit into the pattern. Some authors
prefer the word methods instead of models.
A common distinction is between models developed for premixed flames and models developed for
non-premixed flames. This division is useful primarily when working with models that clearly fit
into it.
The following set-up is an attempt to make an overview of the most well-known models. Here, the
models are grouped according to how they are formulated.
• Thermodynamic relations. These are simplified expressions that we find in textbooks on

thermodynamics. We calculate the fuel conversion, released energy, temperature and pres-
sure, and so on. The processes are ideal and with no spatial gradients. This is the first step in
most analyses.
10.4 Combustion models – classification 157
• Models for combustion speed or flame propagation velocity. This is often simple empirical
expressions. See e.g. the review by Bradley (1992).
• Mean-value models (statistical models), where one determines the mean values for concentra-
tion, energy, etc., and which may have an expression for the mean reaction rate, R k . They can
be further divided into
– models with a series-expanded Arrhenius expression,
– eddy models, for instance the models by Spalding (Eddy Breakup model) and Magnussen
(Eddy Dissipation model),
– models with a prescribed probability density function (pdf), for instance flamelet models.
• Models with a transport equation for a multi-dimensional joint probability density function.
For numerical simulations, the eddy models and flamelet models are the most used models in tur-
bulent combustion. In the following sections, we will briefly go through some of these models.
Reviews of some models are found in articles by (among others) Jones and Whitelaw (1982), Cor-
rea and Shyy (1987) and Borghi (1988), and in the books of Libby and Williams (1980,1994), Kuo
(1986), Warnatz, Maas and Dibble (1999).
Example of a combustion simulation
Figure 10.5 shows a mean-value simulation of a flare on a platform with wind at different speeds. The
program Kameleon was used with Magnussen’s combustion model (EDC), the k-ε model, gradient
models for heat and mass transfer (Sec. 5.2), radiation model (Sec. 12.2), etc. The purpose was to
investigate the radiation heat onto the flare tower and the burner. Notice that the first case is governed
by buoyancy whereas the other to a large extent is governed or stabilized by the wake behind the
burner itself.
A simplified calculation without such a flow-field computation would either approximate the flame
as a cylinder, with height and width according to empirical formulas, or as a point in the center of
the flame. The radiation heat can be set to a certain fraction of the combustion heat and is assumed
to be distributed evenly from the cylinder/point (see e.g. Drysdale, 1985:143f). Such expressions
are functions of fuel mass, heating value, wind speed and a characteristic diameter for the burner.
Without detailed experiments with the actual burner, such simple methods will hardly be able to
take into account that the burner has more outlets, that the flame may be lifted, or that the flame is
stabilized in the wind wake of the burner. Moreover, they can not be used in the region close the
flame.
The burner in Fig. 10.5 is designed to burn a certain (large) mass flow – the design value. In this
investigation, we found – somewhat surprising for the customer – that the radiation heat load in-
creased with lower mass flow rate. This was because a large mass flow rate and outflow velocity
lifted the flame from the burner. With less mass flow, the lift height was reduced and the flame
came closer to the burner. When we inspected the region just above the burner outlet, we could see
large regions with no reaction – partly because the fuel was not mixed with air, partly due to intense
turbulence with extinction (Sec. 11.7). Such conditions can hardly be studied without doing detailed
investigations, either numerically as here, or experimentally with the relevant flare.
Temperature (K)
(projected)
Plane: Y= 100.4
Z (m)
Max
2288.3
2000.0
170
1800.0
1600.0
1500.0
1300.0
1200.0
1150.0
160
1100.0
900.0
750.0
600.0
290.0
293.0
150
Min
110 120 130 140 150

Time= 110.78 X (m)
(a) Wind speed 5 m/s
Temperature (K)
(projected)
Plane: Y= 100.4
Max
Z (m)
160
2367.7
2000.0
1800.0
155
1600.0
1500.0
1300.0
150
1200.0
1150.0
1100.0
145
900.0
750.0
600.0
110 120 130 140 150 290.0
Time= 100.01 X (m) Min 293.0
(b) Wind speed 20 m/s
Figure 10.5: Mean-value simulations (with Kameleon) of a flare on an oil platform; with
different wind speeds. The purpose is to investigate the radiation heat at the tower and at the
burner itself. Notice that the first is governed by buoyancy whereas the other to a large extent
is governed or stabilized by the wake behind the burner.
10.5 Series expansion of the Arrhenius expression 159
10.5 Series expansion of the Arrhenius expression

In the turbulence modeling, we have used the Reynolds decomposition and av-
eraging, and achieved new equations and models for the averaged turbulence
quantities. This can also be done in order to find a model for the averaged
reaction rate, R k , in Eq. (2.9). We will, however, see that the approach is not
very useful.
For a simple reaction between fuel and oxidizer,

kf
fu + ox → prod, (10.5)
the reaction rate for the fuel can be written:
Rfu = −kf ρ 2 Yfu Yox , (10.6)

where
Ta
kf = BT β exp(− ). (10.7)
T
The relation with Eqs. (1.12)–(1.13) is seen by setting Ta = E a /Ru , kf = k/Mox and B = A/Mox .
We can introduce the Reynolds decomposition T , Yfu and Yox , and the equation of state p = ρRT .
By a Taylor-series expansion of the temperature T about T , one obtains the expression
Bp2 (β−2) Ta
R fu = − T exp(− ) · Y fu Y ox · FC . (10.8)
R2 T
Correlations with the pressure are neglected. FC is a series that contains correlations as T ′ 2 , Yfu′ Yox
′ ,
′ , T ′ Y ′ , T ′ 3 , T ′ 2 Y ′ , T ′ Y ′ Y ′ , and so on. This is written out in Sec. B.9.
T ′ Yox fu fu fu ox
If Ta ≪ T and T ′ ≪ T the series can converge rapidly, but even then there are quite a few cor-
relations to model. Also when Ta ≥ T , the series can converge, but will then contain very many
correlations of high order.
The conditions T ′ ≪ T and Ta ≪ T are satisfied in nearly isothermal reactions with a low activation
energy (E a = Ta Ru ). Such reactions we can find in some chemical production processes. In most
combustion reactions, however, the T ′ is of the same order of magnitude as T , and often Ta ≫ T .
From all these circumstances, it appears that the model of Eq. (10.8) is unsuitable as a model for
combustion.
10.6 Eddy models

These models are based on the breakdown of eddies and the molecular mixing
that results from this. One imagines that an eddy contains mainly one of the
species. When many such eddies with different substances are broken down,
the molecules are mixed. The aim is to model the source term R k in the
balance equations for chemical species, Eq. (2.9).
For a combustion reaction to take place,

– reactants must be mixed molecularly,
– hot reaction product must come close and “ignite” the reaction.
If the reactants are premixed, the first condition is always satisfied. But also here, the product must
be mixed in.
If one assume “infinitely fast” chemistry, the reaction will be determined by the mixing process or
the eddy breakdown.
Two models that are based on such eddy breakdown (or -breakup):
Spalding: Eddy Breakup model
Spalding (1976) developed the model having both premixed and non-premixed combustion in mind.
But it is often classified under premixed flames, for instance by Bray (1980:138) and Kuo (1986:478).
With a simple reaction, fu + ox → prod, the reaction rate for the product is modeled
q
ε
R prod = −(1 + r )R fu = CEBU ρ Yfu′ 2 . (10.9)
k
CEBU is a model constant.

In the literature, we find a number of models called “modified Eddy Breakup model”, which use
mean values instead of variance in Eq. (10.9). In principle, these models correspond to Magnussen’s
model below.
Magnussen: Eddy Dissipation model
This model we will come back to in the next chapter. For the overview, I will just mention a few
aspects.
The first version of the model was presented by Magnussen and Hjertager (1976). The model has
Spalding’s model as a background but differs by using the mean mass fraction, not the variance.
The idea is that reactions are restricted and determined by the species that is the least available.
The model is developed and used for premixed and non-premixed combustion. In the literature, the
model is often called an “Eddy Breakup” model, together with Spalding’s model.
10.7 Prescribed probability density 161
In the first version, the reaction rate for fuel is modeled as

ε 1 1
R fu = A · ρ · min(Y fu , Y ox , B · Y pr ). (10.10)
k r 1+r
Here, A and B are model constants.

In the next chapter, we will look more closely at a newer version of the model. But it should be
mentioned that the first model still is widely used. For instance, it is this model that is implemented
in many commercial CFD programs.
10.7 Prescribed probability density

In the models above, one attempts on modeling transport and source terms
for averaged quantities as concentration and energy. Another path can be to
consider the definition of mean value and try to model the involved functions.
Mean values can, according to Eq. (7.7), be expressed by

Z ∞
G= G(ξ ) f (ξ )dξ, (10.11)
−∞
where f (ξ ) is the probability density. Each location (Ex , t) has its own probability density. Hence,
f (ξ ) also depends on place and time. In general, G and f will be functions of more than one
variable, see Eq. (7.17). For practical application, one will usually simplify as shown in Eq. (10.11).
Then, we must establish somehow that this variable ξ is the single important one that everything else
can be a function of it. Up to now, we have not specified what ξ actually should be.
In order to use Eq. (10.11), we first need to know a functional relation as G = G(ξ ), where each
quantity (stochastic variable) G is expressed as a function of a characteristic variable ξ (or more).
Second, we have to know the probability density f (ξ ) as a function of the same variable.
We will treat these two matters separately:
Prescribed probability density: β-function
In the expression of Eq. (10.11), we have to know the probability density f (ξ ) in each location
(E
x , t). A simple way is to use a prescribed function (also: presumed or assumed). The probability
density function is then set up from empirical data. The shape is given when the expectation value
and variance are known.
An example of such an prescribed function is
ξ a−1 (1 − ξ )b−1
f (ξ ) = . (10.12)
B(a, b)
Here must 0 ≤ ξ ≤ 1. In the combustion literature, it is often called the β-function. In Fig. 7.3
(page 110), the function is shown with different values of the parameters a and b while Table 7.1
shows some statistical quantities calculated for the same values. The function is normalized by what
the mathematicians call the β-function:
Z 1
Ŵ(a)Ŵ(b)
B(a, b) = x a−1 (1 − x)b−1 d x = . (10.13)
0 Ŵ(a + b)
By using Eq. (10.11), one can determine (Richardson, Howard and Smith, 1953) the parameters a
and b (a > 0, b > 0):
a ξ (1 − ξ )
ξ= and ξ ′ 2 = . (10.14)
a+b 1+a+b
If we solve the equations for ξ and ξ ′ 2 , we thus can find the parameters a and b, and thereby
an approximation for the probability density f (ξ ). We can then determine the mean values and
variances of all other quantities. That is, if the quantity can be expressed as a function of ξ .
When choosing a function f (ξ ) for the probability density, several aspects have to be considered:
– The function must be able to reproduce relevant distributions. This can range from delta
functions in one point or in both end points (ξ = 0 and ξ = 1), very skewed distributions, and
to a normal distribution, cf. Fig. 7.2.
– It will be beneficial if the function gives an explicit relation between the parameters of the
function and the mean value and the variance. If this request is not satisfied, one has to iterate
every time to determine the parameters. This, of course, takes more computer time.
– It can be a benefit if the function is not split – that it is not composed of several parts as the
variable (ξ ) varies.
– The function should not cost too much computer time for integration, cf. Eq. (10.11).
However, having said this: It appears that the choice of function does not have a strong influence on
the computational result.
The β-probability density of Eq. (10.12) complies with all these considerations, which means that it
is of much use. Another function that is also used, is a clipped Gaussian function. This is a normal
distribution (Gaussian distribution), where the probability density (see Eq. (7.6)) outside the interval
ξ = [0, 1] is cut away, and the clipped parts replaced by delta functions (impulse functions) at ξ = 0
and ξ = 1. R
0
 −∞ g(c)dc when ξ = 0

f (ξ ) = g(ξ ) when 0 < ξ < 1 (10.15)

R ∞ g(c)dc when ξ = 1
1
where 2 !
−1/2 c−ξ
g(c) = 2πξ ′ 2 exp − . (10.16)
2ξ ′ 2
This function can represent the different distributions – possibly better than the β-probability density.
But it is sectioned in more parts, it does not give an explicit expression for the two parameters of the
function, and it contains an exponential function, which is demanding to calculate.
10.7 Prescribed probability density 163
Functional relations
Equation (10.11) is in principle valid both for premixed and non-premixed flames. But in practice,
it is mainly used for non-premixed flames. The difficulty of premixed flames is to find a suitable
characteristic variable ξ and the corresponding functional relations G = G(ξ ).
In non-premixed flames, the mixture fraction is used as the characteristic variable. In Sec. 5.4 (page
84), we defined the mixture fraction ξ such that
ϕbl = ϕ(ξ ) = ξ ϕ1 + (1 − ξ )ϕ2 . (10.17)
Here is 0 ≤ ξ ≤ 1, where ϕ is a conserved scalar (see page 84), and ϕ1 and ϕ2 are the values if the
inflow states (1) and (2).
The equation holds for properties with no source term, i.e. conserved scalars. For a property π that
is not conserved, the function π = π(ξ ) is generally non-linear in ξ . The function π(ξ ) can be found
by assuming complete reaction (all fuel to CO2 , H2 O); or one can assume equilibrium at the product
side (CO2 , CO, H2 O, H2 , etc.). A third alternative is to use tabulated relations for the various π(ξ ),
see about flamelet models below.
One can also formulate expressions valid for more than two inlets (or more correct: two inflow
states). The situation becomes more complex as each variable has to be expressed as a function of
more variables, G(ξ1 , ξ2 , . . . , ξn ).
If there occurs an infinitely fast reaction, the instantaneous mass fraction, temperature, etc. will
be functions of the mixture fraction: Yk (ξ ), T (ξ ). When the inflow conditions are known, we can
determine the mean values according to Eq. (10.11), for instance
Z 1
Yk = Yk (ξ ) f (ξ )dξ, (10.18)
0
Z 1
T = T (ξ ) f (ξ )dξ, (10.19)
0
Z 1
Yk′ 2 = (Yk (ξ ) − Y k )2 f (ξ )dξ, (10.20)
0
Z 1
T′2 = (T (ξ ) − T )2 f (ξ )dξ, (10.21)
0
Z 1
Rk = Rk (ξ ) f (ξ )dξ. (10.22)
0
The integrals go from 0 to 1 because 0 ≤ ξ ≤ 1 when the mixture fraction is chosen as the charac-
teristic variable.
Equations to be solved, an overview
We solve the equations for mean momentum (velocity) and pressure (or density). For this we use a
turbulence model, for instance with equations for k and ε or equations for the Reynolds stresses.
Instead of the equations for mean mass fraction, we solve the equations for mean and variance of
the mixture fraction, ξ and ξ ′ 2 . (Notice: This is for models with a prescribed probability density,
not in general.) In order to model these equations, one can solve equations for the scalar dissipation
εξ ′ 2 (destruction of ξ ′ 2 ) and for the turbulence fluxes ξ ′ u ′j . These quantities, however, can also be
modeled by algebraic expressions as in Chapter 5.
With ξ and ξ ′ 2 as the parameters of a prescribed probability density, one can integrate expressions as
in Eq. (10.18) and (10.20). One then has to know the quantities as functions of the mixture fraction,
e.g. Yk (ξ ). If the enthalpy h can be regarded as a conserved scalar (no source term, e.g. radiation),
also the mean enthalpy h and the mean temperature T can be found in this way. Otherwise, the
h-equation has to be solved.
10.8 Flamelet models

This may be the best known and most widely used sub-group of models with a
prescribed probability density of the previous section.
Non-premixed flames can be classified for instance, according to their turbulence intensity and typi-
cal length scale. This is done in a so-called Borghi-diagram, see Sec. 10.2. If the intensity is low and
the flame front (reaction sheet) is thinner than the smallest turbulence length scale (the Kolmogorov
length scale), we are in the flamelet-region. (Flamelet, from Latin: “small flame”.) The flame front
is disrupted and looks like many small laminar flames.
Each of these small flames is regarded as an one-dimensional laminar flame. From calculations or
measurements for simple laminar flames, the different quantities can be found as functions of the
mixture fraction, as G(ξ ) in Eq. (10.11). Such functions can be stored in a so-called flamelet library
and be used in expressions like Eqs. (10.18)–(10.21).
The use of the term flamelet is not fully consistent in the literature. The term can be restricted to
models that use data from simple laminar flames. But the word can also be used about models that
determine the functional relation G(ξ ) in other ways.
Models with prescribed probability density are presented and/or discussed by among others
Peters (1984,1986), Bilger (1988), see also Bilger (1980), Correa and Shyy (1987:258f,269f),
Borghi (1988:264f), Warnatz, Maas and Dibble (1999:188f).
10.9 Transport equation for probability density 165
10.9 Transport equation for probability density

The models or methods we have studied this far, say something about the
averaged quantities. This can be the expectation value (mean value) and
variance of a quantity, or a co-variance of two quantities. They do not say much
about the statistical distribution about the mean value. With a prescribed prob-
ability density, the probability density is partly obtained from the computation,
although it is mainly decided prior to the computation.
In this section we will shortly look at a more general method, and include some
basic theory. What is discussed above, Sec. 10.7, is a simplified view.
The state of a system is represented by three velocity components u 1 , u 2 , u 3 , and n scalar vari-
ables θ1 , . . . , θn . These scalar variables are energy and mass fractions. This gives a (3 + n)-
dimensional random vector [u 1 , u 2 , u 3 , θ1 , . . . , θn ] or [uu , θ ]. The vector is realized with the values
[v 1 , v 2 , v 3 , ϕ1 , . . . , ϕn ] or [vv , ϕ ]. The distribution function is
Fu 1 ,u 2 ,u 3 ,θ1 ,...,θn (v 1 , v 2 , v 3 , ϕ1 , . . . , ϕn ) = Fu ,θθ (vv , ϕ )

= F(vv , ϕ ) = P(u 1 < v 1 , u 2 < v 2 , u 3 < v 3 , θ1 < ϕ1 , . . . , θn < ϕn ) (10.23)
and the probability density function
f u 1 ,u 2 ,u 3 ,θ1 ,...,θn (v 1 , v 2 , v 3 , ϕ1 , . . . , ϕn ) = f u ,θθ (vv , ϕ )

∂ 3+n F(v 1 , v 2 , v 3 , ϕ1 , . . . , ϕn )
= f (vv , ϕ ) = . (10.24)
∂v 1 ∂v 2 ∂v 3 ∂ϕ1 . . . ∂ϕn
A function of random variables becomes a random variable itself, for instance

G = G(u 1 , u 2 , u 3 , θ1 , . . . , θn ). The expectation value is
Z ∞ Z ∞
G= ... G(v 1 , v 2 , v 3 , ϕ1 , . . . , ϕn ) f (v 1 , v 2 , v 3 , ϕ1 , . . . , ϕn )dv 1 dv 2 dv 3 dϕ1 . . . dϕn .
−∞ −∞
(10.25)
Mathematically, we can regard (v 1 , v 2 , v 3 , ϕ1 , . . . , ϕn ) as a “space”. With some mathematics, we
can develop the probability density in this space. The starting point is the equations for the random
variables u i (i = 1, 2, 3) and θk (k = 1, . . . , n).
Pope (1985) obtains

∂f ∂f ∂ p ∂f ∂
ϕ)
ρ(ϕ + ρ(ϕ
ϕ )v j + ρ(ϕ ϕ )g j − + ϕ )Sk (ϕ
ρ(ϕ ϕ) f
∂t ∂x j ∂ x j ∂v j ∂ϕk
" ! #   
∂ ∂τi j ∂ p′ ∂  ∂ Jik
= − + v,ϕ f + v, ϕ f  . (10.26)
∂v j ∂ xi ∂x j ∂ϕk ∂ xi
Here, Sk is the source term and Jik is the molecular flux in the equation for θk , g j is the body-force
acceleration and | shows “on condition of”. ρ and Sk must be known functions in the ϕ -space.
The terms on the left-hand side of the equation represent change with time, transport in physical
space, transport in velocity space with body forces and averaged pressure gradient, and transport in
the ϕ -space with chemical reactions. All these terms are functions of quantities that are known when
the equation is solved. Hence, they are closed and require no modeling. The terms at the right-hand
side are transport in velocity space with viscous stresses and with the fluctuating pressure gradient,
and transport in the ϕ -space with molecular flux. These terms have to be modeled.
In Reynolds-averaged equations for chemical reactions, the reaction terms have to be modeled, and
this is one of the greatest challenges. In the equation above, the reaction terms are known, while
terms with molecular transport are unknowns. If one can find a suitable model for the unknown
terms, and also find a method for solving the general equation, we can thus determine the multi-
dimensional probability density. We can then resolve all the variables that we might desire.
However, the involved challenges are huge. Up to now, the models and methods developed are tested
for simple systems only.
One approach is first to integrate the equation over the velocity space. The complexity is then re-
duced while, however, information on the velocity field is lost. This has to be resolved by other
means, for instance with Reynolds-averaged equations and an ordinary turbulence model. The inte-
gration also introduces a new turbulence term in the equation. This has to be modeled.
For simple chemical systems, with few random variables (θ1, . . . , θn ), the equations can be solved
with a so-called Monte Carlo method. (The name plays on the town name Monte Carlo, known for
hazard games.)
The method is in principle much more general than models with Reynolds-averaged equations.
Many researchers find it attractive as the reaction terms need no modeling. However, modeling
is not avoided, and even for very simple (simplified) chemical systems, the method is very resource
demanding. The future will show what such methods can give.
For further reading, see for instance Gran (1994), Pope (1985) and Kollmann (1992).
10.10 Direct numerical simulation of turbulent combustion

This section has some relations to Sec. 8.3 on the energy spectrum, Sec. 9.1
on length- and timescales, and Sec. 10.2–10.3 on flame scales and chemical
kinetics, Sec. 11.4 on fine structure. We have touched direct simulation in
Sec. 9.4. Said in another way: This section should not be the first you read in
this book.
Turbulent flows can be simulated directly from the basic equations, without models, cf. Sec. 9.4.
Such methods will primarily be supplements to mean-value simulations, not alternatives. Direct
simulations can give knowledge and data that is difficult or impossible to obtain from experiments.
We can make use of this in the development of more practically oriented methods.
In this section, we will discuss the requirements for a direct numerical simulation of a turbulent flow
with reaction. – How fine resolution is required in time and space? – How much computation time
is required for the simulation? For turbulent flows without reaction these questions are discussed in
10.10 Direct numerical simulation of turbulent combustion 167
the literature. For turbulent combustion, however, it seems that the matter is not fully discussed.
We will see that direct simulations of turbulent flows with combustion will demand substantially
larger resources than a direct simulation of flow without reaction.
Direct numerical simulation without models is a science by itself. The numerical techniques used
are not the same as in engineering calculations with mean-value models (statistical models), but
developed especially for such purposes. What can be simulated is to a large extent restricted by the
computer resources available and by the numerical methods. The expressions below for resolution in
time and space and for computational time are estimates. The exact figures depend on the numerical
techniques and the computer architecture and how this is utilized. The expressions can be used to
make estimates for resolution and computer time, and for comparing different cases.
The total computational time can be put up as the product

number of number of time per point
· · = Np · Nt · Tpt . (10.27)
grid points time steps per time step
The number of grid nodes (points) Np is the spatial resolution. How fine is the grid mesh required
be in order to catch the smallest lengths? Or rather: – How small are really the smallest lengths that
are important for the flow?
The number of timesteps Nt is the resolution in time. – How short time steps are required to catch
the most shortlived motions or the fastest reactions?
For each timestep in each point, we have to do computations that take the time Tpt .
Computation time per timestep per point
For each point and timestep, we have to solve the continuity equation, the energy equation and an
equation for momentum in each of the n dimensions. Basically, n equals 3. Furthermore, each of the
NS chemical species will require one equation for concentration (mass fraction). Strictly, we could
be satisfied with (NS − 1) equations, since the sum of the species equations equals the continuity
equation. The results are, however, more accurate when all species equations are solved. In addition
to solving the equation for each species, there will be an additional cost for calculating the source
term, Eqs. (1.12)–(1.13). When each species takes part in (on average) n̄ r reactions, this cost can be
assumed proportional to NS n̄ r . The computational time for each point and timestep becomes
Tpt ∼ (2 + n + NS + ANS n̄ r ) · Top , (10.28)
where Top is the computational time (CPU time) per operation. A is a factor that varies from one case
to another, with the method of numerical solution, and with the machine architecture. Experience
from some simulations (by Inge R. Gran) indicates that A is of order 10. Radiation, which gives a
source term in the energy equation, is omitted in the present discussion (cf. Sec. 12.2). It would give
an addition in Eq. (10.28). It would most likely be a substantial addition, although an estimate is
hard to give. One problem is that the numerical treatment is not sufficiently developed for radiation
calculations.
Characteristic scales in space and time
In previous chapters, we have looked at characteristic scales. We have seen that scales for length and
time are spread over wide spectra both for the turbulence (Chapter 8) and for the chemical reactions
(Sec. 10.3). The main flow has an external lengthscale L and a velocity scale U . The characteristic
velocity scale and length scale (“large”) for the turbulence are u ′ and ℓ′ , with the timescale θ = ℓ′ /u ′ .
We can regard these proportional to the scale of the main flow, so that L/U ∼ ℓ′ /u ′ = θ .
We have assumed the micro-scales to be characteristic for the “small” motions of the turbulence.
From the energy spectrum, that is, the motion itself, we have the Kolmogorov scales (Sec. 8.3);
length η = (ν 3 /ε)1/4 and time τ = (ν/ε)1/2 . For the molecular mixing (Sec. 8.6), we have the
1/4 √
Batchelor scales for each scalar (species and temperature); length ηB = νD2 /ε = η/ Sc and
time τB = τ . For these length scales, we have convection time τU = η/U and τU,B = ηB /U ,
Eq. (9.7) and (9.8). This is the time it takes for the Kolmogorov and Batchelor scales to flow past a
location in the flow.
The question that remains is whether these “large” and “small” scales are sufficiently representative
of the flow.
There are eddies less than the microscales (see Sec. 8.6). But these may not have significant contri-
butions to the flow. For flows without reactions or mixing, this is the accepted theory. Investigations
by among others Zhou (1993) support this conclusion. Moin and Mahesh (1998) also give arguments
for allowing the resolution of a direct simulation to be somewhat coarser than the Kolmogorov scales
while still reproducing the important quantities of the flow. This does not necessarily imply that the
smaller scales are unimportant for the molecular mixing of different species in a chemical reaction.
One can argue that scales that are not affected by chemical reactions, will adjust to larger scales up
to the Kolmogorov or Batchelor scales. To some extent this argument is a parallel to mean-value
simulation where we resolve the “large” scales or transport scales, and imagine that the turbulence
can be represented by these. And if reactions affect smaller scales, it is reasonable to believe that
also the chemical scales are less than the microscale, and therefore themselves give the limit for
resolution. All these scales form an unknown and difficult landscape, and to find the full picture we
might have to do – just a direct numerical simulation.
For the chemical reaction (overall reaction, flame) we have a “large” timescale that is a function of
−1
the fuel conversion rate, τc ∼ Rfu (Sec. 10.2). The corresponding “large” length scale is δL ∼
1/2
(Dτc ) .
Such length scales can be defined for each of the hundreds or thousands of elementary reactions. For
definition and discussion of such scales, see Warnatz, Maas and Dibble (1999:97ff). We will obtain
a minimum chemical timescale or reaction timescale, τr,min , and a minimum reaction length scale,
δr,min . For the length scale, we have the convection time τU,r,min = δr,min /U , see about Kolmogorov
and Batchelor scales above. This is the time it takes for the reaction length scale to flow past a
location in the flow.
Spatial resolution
The calculation domain has to be sufficiently large to fill the relevant space. We assume that the
external length scale L can represent this length in each of the n directions, so that the volume
10.10 Direct numerical simulation of turbulent combustion 169
becomes L n . Basically, there are n = 3 dimensions and the volume L 3 . We can, however, also think
of simplifications to a two-dimensional space, n = 2.
This space has to be subdivided so fine that we can catch the minimum scales, thus
n n n
L L n/2 η
Np ∼ = · max 1, Scmax , . (10.29)
min{η, ηB,min , δr,min } η δr,min
3/2 3/4
From above (page 126) we have the relation L/η ∼ ℓ′ /η ∼ Reλ ∼ Reℓ .
The spatial resolution must at least be
n
L 3n/2
Np ∼ ∼ Reλ . (10.30)
η
In many cases, the chemical scale will be the minimum length scale, so that
n n n
L η 3n/2 η
Np ∼ · ∼ Reλ · . (10.31)
η δr,min δr,min
The final factor may imply a substantially stricter requirement to spatial resolution.
Resolution in time
The largest time scale for the flow will be L/U , which will be proportional to θ = ℓ′ /u ′ . There might
be chemical timescales that are larger; either for the overall reaction or for a slow reaction. The least
timescale is the time needed for the least length scale to flow past the location (cf. Sec. C.6). The
time resolution must be the ratio of the largest to the smallest timescale:
largest time max{θ, τr,max }
Nt ∼ ∼
smallest time min{τU , τU,B,min , τU,r,min , τr,min }

θ θ θ θ τr,max τr,max τr,max τr,max
∼ max , , , , , , , . (10.32)
τU τU,B,min τU,r,min τr,min τU τU,B,min τU,r,min τr,min
3/2
The two first of these ratios relate to the turbulence, and we have: θ/τU ∼ L/η ∼ Reλ and
√ √ 3/2
θ/τU,B,min ∼ Scmax L/η ∼ Scmax Reλ . The next five factors of Eq. (10.32) tell something
about the relation between turbulence and reaction, while the last one says something about the
relations between different partial reactions.
Direct simulation of turbulence without reaction
When there is no chemical reaction, the expressions above will be simplified by canceling the chem-
ical scales. One can also argue that without chemical reactions and mixing of species, we need not
consider scales less than the Kolmogorov scales. The whole thing is reduced to
n
L θ 3n/2 3/2
Np Nt ∼ · ∼ Reλ · Reλ . (10.33)
η τU
The expression for computational time, Eq. (10.28), without changes in compositions, will be re-
duced to
Tpt ∼ (2 + n)Top . (10.34)
From this, we can see that a direct simulation of a turbulent flow without reaction require much less
resources than one with reaction. This can be seen in relation to the remark at page 29, that laminar
flames are much more complex than laminar flow without reaction.
Estimate of a practical case with and without combustion
The different scales vary from one flow case to another, and with the fuel. For comparison of
simulations with and without combustion, we can select a case with Reλ = 200, θ = 1 s, L = 1 m,
NS = 100, n̄ r = 10, τr,min = 10−9 s, δr,min = 10−6 m.
Now, we get Np ∼ (L/δr,min )3 ∼ 1018 in Eq. (10.29), since this is larger than (L/η)3 ∼ 2 · 1010.
For the timescales, we get Nt ∼ θ/τr,min ∼ 109 in Eq. (10.32), as this factor is larger than θ/τU ∼
L/η ∼ 3 · 103 and θ/τU,r,min ∼ L/δr,min ∼ 106.
For the computational time, Eq. (10.28), we get Tpt ∼ (2 + 3 + 100 + 10 · 100) · Top ∼ 103 Top . We
have then, somewhat optimistic, assumed the factor A = 1.
In total, this gives Np · Nt · Tpt ∼ 1030 Top .
Without reaction, we get Np · Nt ∼ Re6λ ∼ 6 · 1013 and Tpt ∼ 5Top, so that Np · Nt · Tpt ∼ 1014 Top .
In other words: According to this estimate, a simulation of the flow with combustion takes 1016
times more computational time than the corresponding flow without reaction. Or, if the latter takes
one second, the former will take 300 million years.
Simplifications and practical adaptions
We realize from the expressions above that a full numerical simulation of a practical case with
combustion becomes a big and heavy issue – which lies far into the future. With certain care,
however, we can find reactions tractable. The purpose will not be to solve industrial problems but to
study turbulent combustion and provide data that can be utilized to improve models we are using.
If the flow and turbulence can be calculated in two dimensions, with n = 2 in Eq. (10.29), the cost is
reduced. We then have to model the effects in the third directions based on the computational results
for the other two directions. If we choose a flow where the turbulence is not very intense, i.e. that
has a relatively low Reynolds number Reλ , we see that the requirment on the resolution is reduced.
Furthermore, we can choose a fuel (reactants)
– such that δr,min ≈ η, cf. Eq. (10.29);
– such that τr,max ≈ θ and τr,min ≈ τU , cf. Eq. (10.32);
– such that the reaction mechanism is not too extensive, that is, such that NS and n̄ r not are large.
Eq. (10.28).
This will reduce the cost so much that a simulation can be conducted, see e.g. Gran, Echekki and
Chen (1996).
Chapter 11
Magnussen’s combustion model,

the Eddy Dissipation Concept
This chapter will present Magnussen’s Eddy Dissipation Concept (EDC) somewhat more in
detail and “student-friendly” than what is found in scientific articles.
11.1 Basis and overview

EDC is a model for chemical reactions in turbulent flow. The model is developed
and tested by Magnussen and co-workers at NTNU/Sintef through a period of
25 years. In this section, as an introduction, some key features are listed. The
details will come gradually, and the section will be likely to show new content
when you re-read it after the rest of the chapter.
In simulations of turbulent flows, we solve equations for mean values. These are quantities that are
related to the large turbulence scales.
When we compute combustion, we need an expression for the reaction rate. This is the source term
in the equation for a mean mass fraction or mean concentration.
The combustion occurs where the reactants are mixed molecularly. This is mainly in the smallest
eddies or the fine structures. There, most of the dissipation of turbulence energy to heat also takes
place.
For reactants to react, the temperature has to be sufficiently high and the residence time sufficiently
long. Hot reaction products must get in touch to “ignite” the reaction. Reactants and products must
be mixed in the right ratios long enough for the reaction to take place.
The fine structures are not evenly distributed in time and space. In one location, there will be periods
172 11 Magnussen’s combustion model, «Eddy Dissipation Concept»
with much and periods with little fine structure. The fine structures appear intermittently.
Magnussen’s combustion model (EDC) has these main elements:
– A model for the energy transfer from larger to smaller scales (cascade model).
– With the energy transfer model, characteristic quantities for the lowest level of scales (the
fine structures) can be expressed as functions of scales from the large-scale level. The model
provides a relation between small and large scales.
– The large-scale level can be related to the mean flow by a turbulence model, for instance a k-ε
model, – or resolved directly by large-eddy simulation.
– The model regards the fine structures as a steady-state homogeneous reactor. The chemical
reactions occur here. The mass balances give the reaction rates for the reactor and the mean
reaction rates. These are the source terms of the equations for the mean mass fractions. The
energy balance for the reactor provides the temperature in the fine structures.
– By modeling the residence time and compare to the reaction time, one can model extinction.
– Computations for detailed chemical kinetics can be coupled to the model.
For fast reactions, the reaction rate is limited by the rate of mixing – which is determined by the
turbulence. For reactions with a certain reaction time, the residence time in the reactor must be
larger than the reaction time. Otherwise, local extinction (blow off) will occur.
Magnussen has also proposed a model for formation and combustion of soot in turbulent flames.
EDC is presented in different versions. The latest is found in Magnussen (1989). Earlier version
are found in Magnussen (1981), Magnussen, Hjertager, Olsen and Bhaduri (1978) and Magnussen
and Hjertager (1976). Extinction is modeled by Byggstøyl and Magnussen (1983). The energy-
cascade model is suggested by Magnussen (1975b) and worked over by Ertesvåg (1991), Ertesvåg
and Magnussen (1997). Treatment of detailed chemical kinetics is shown by Gran (1994), Gran and
Magnussen (1996a,b).
11.2 Energy transfer – cascade model

Reactions occur in small eddies. Large eddies provide transport, and the
transport is represented by turbulence models. This section explains the
model that relate the small-scale phenomena to large-scale phenomena in
the turbulence. It may be beneficial to read the preceeding section first. Also
Chapter 8 can give a better understanding of the following, and vice versa.
Averaged over time, the length and timescales in the turbulence are continously distributed over a
wide spectrum. Mechanical energy is transferred from the mean flow to large eddies and then further
to smaller and smaller eddies. The large eddies are the most inertial and contain the largest energy.
The smaller eddies swivel faster but contain less energy since their total mass is small. The frequency
is the largest in the smallest eddies. Here, the local viscous stresses are the largest. Viscous friction
transfers mechanical energy to heat. This transfer, the dissipation, takes place at all scale levels,
however, most in the smallest eddies. The spectrum of large and smaller eddies is often described
as an energy or turbulence cascade. (“Cascade” is French for a stepwise, often artificial, waterfall;
from Italian “cascata”: fall, waterfall.)
11.2 Energy transfer – cascade model 173
Mean flow
w′
❄
u ′ , L ′ , ω′ ✲ q′
w′′
❄
u ′′ , L ′′ , ω′′ = 2ω′ ✲ q ′′
′′′
❄w
wn
❄
u n , L n , ωn = 2ωn−1 ✲ qn
❄wn+1
w∗
❄
u ∗ , L ∗ , ω∗ ✲ q∗
Figure 11.1: Energy cascade
Figure 11.1 illustrates the model for transfer of mechanical energy from the mean flow, through
turbulence energy to heat. w′ is the feed of mechanical energy from the mean flow to the turbulence,
or production of turbulence energy. The sum of q ′ + q ′′ + . . . + q ∗ is the dissipation of turbulence
energy.
The first level of the turbulence structure is the large and energy-rich turbulence eddies. It is charac-
terized by a velocity scale u ′ , a length scale L ′ , and a frequency or strain rate ω′ = u ′ /L ′ . L ′ is an
exchange length, that is, a diffusion or integral scale. This level represents the whole spectrum by
including the effects of smaller scales. When we solve averaged turbulence equations, for instance
the k and ε-equations, we are at the first level.
The next level represents the part of the spectrum where the characteristic frequency is ω′′ = 2ω′ ,
velocity u ′′ and length L ′′ . Similar to the first, also this level includes the effects of all lower levels. In
the same way, the n-th level is characterized by ωn = 2ωn−1 , u n and L n . At each level, ωn = u n /L n .
In the smallest eddies, ω∗ , u ∗ and L ∗ are of the same order of magnitude as the Kolmogorov scales
(page 130).
The transfer from the first to the next level, w′′ , equals the sum of dissipation at all subsequent levels,
so that the dissipation ε = q ′ + w′′ . The feed of mechanical energy from the mean flow to the first
level is the production, w′ = Pk , if we assume local equilibrium. This is a function of quantities
belonging to the first level (Reynolds stresses), and of quantities belonging to the mean flow (mean
velocity gradients or mean strain rates).
Correspondingly, the transfer from the first til the second level will be a function of quantities be-
longing to the two levels. ω′ = u ′ /L ′ is a typical strain rate for the first level, while u ′′ is a typical
velocity scale for the second level. The frictional work, or dissipation, at the first level have to be a
product of the viscous stress (νω′ ) and a strain rate (ω′ ) at this level.
We can write the model as
w′′ = 23 CD1 ω′ 2u ′′ 2 and q ′ = CD2 νω′ 2 . (11.1)
Here, CD1 and CD2 are numerical factors that are regarded as constants. Magnussen (1975b,1981)
presumes that the expressions for w and q are similar for all levels. When we introduce ω′′ = 2ω′ ,
or ωn = 2ωn−1 , the model for transferred and dissipated energy at the n-th level becomes
wn = 23 CD1 ωn u 2n and qn = CD2 νωn2 , (11.2)
and the balance is wn = qn + wn+1 .

Unless the Reynolds number is low, the dissipation from the upper levels is small. Thus, qn ≪ wn
and wn ≈ wn+1 for a small n. We can then set u ′′ 2 = 21 u ′ 2 , and we obtain
w′′ = CD1 ω′ 32 u ′ 2 = CD1 ω′ k and q ′ = CD2 νω′ 2 . (11.3)
Here, the energy released from the first level is expressed by quantities at the same level. This can
be utilized as a model for the dissipation term in the k-equation. We can regard the latter equation
for w′′ as an approximation and Eq. (11.1) as the actual model for w′′ .
Now we can assign a value to the model constant CD1 . We use the model νt = u ′ · L ′ , where we
have chosen a constant of proportionality equal to 1 (see Eq. (3.10)). Introducing the approximation
ε = w′′ = 23 CD1 u ′ 3 /L ′ , we get νt = 32 CD1 u ′ 4 /ε = 32 CD1 k 2 /ε. Then, 32 CD1 corresponds to the
constant Cµ (=0.09) in the k-ε model (see page 74), and we can set CD1 = 0.135.
We then turn to the last level of the cascade, to the fine structures (n = ∗), where w∗ = q ∗ , and
w∗ = 32 CD1 ω∗ u ∗ 2 and q ∗ = CD2 νω∗ 2 . (11.4)
The expressions written here are the same as Magnussen (1981) uses in the development of the
combustion model. He uses the constants CD1 = 4ξ 2 = 0.134 (ξ = 0.183 i a constant used by
Magnussen) and CD2 = 15ξ 2 = 0.50. The development in this chapter follows Ertesvåg (1991:38),
and is somewhat more general than in the preceeding papers.
The direct dissipation from a level of the energy cascade is, according to the model, one-fourth of
the dissipation from the level below. When there are many levels (high Reynolds number), the total
dissipation is ε = 34 q ∗ . The three last levels account for more than 98 % of the dissipation. From the
11.2 Energy transfer – cascade model 175
model, we also find that u 2∗−1 = 52 u 2∗ , and u 2∗−2 = 10

21 2
u ∗−1 , where u ∗−1 and u ∗−2 are characteristic
scales for the second last and third last level. The ratio beweeen u 2n−1 and u 2n is close to 2 for all the
other levels in the energy cascade.
In the combustion model, we utilize the energy cascade to establish some relations that determine
the combustion: For high Reynolds numbers, ε ≈ w′′ and ε = 43 q ∗ . This can be written as
u′ 3
ε = w′′ = 23 CD1 , (11.5)
L′
u∗ 2
ε = 34 q ∗ = 43 CD2 ν , (11.6)
L∗ 2
u∗ 3
ε = 43 q ∗ = 43 w∗ = 2CD1 . (11.7)
L∗
The latter two equations give the Reynolds number for the lowest level,
u∗ L ∗ 2CD2
Re∗ = = , (11.8)
ν 3CD1
and the characteristic scales !1/4

3 1/4
∗ 2 3CD2 ν3
L = 2
, (11.9)
3 CD1 ε
!1/4
CD2
u =∗
2
(νε)1/4 . (11.10)
3CD1
These scales are of the same order of magnitude as the Kolmogorov scales.
With the constants CD1 = 0.134 and CD2 = 0.50 in Eqs. (11.5)–(11.10) , we obtain
u′ 3 u∗ 2 u∗ 3
ε = 0.20 = 0.67ν = 0.27 , (11.11)
L′ L∗ 2 L∗
u∗ L∗
Re∗ = = 2.5, (11.12)
ν
3 1/4
ν
L ∗ = 1.43 , (11.13)
ε
u ∗ = 1.75(νε)1/4 . (11.14)
In addition we can use a Taylor microscale λ and put up the expression (see page 118)
u′ 2
ε = 15ν . (11.15)
λ2
The cascade model has provided the characteristic scales for the fine structures as functions of scales
for the upper level. These scales can be found from a turbulence model, for instance a k-ε model.
11.3 Cascade and energy spectrum

For this section, it may pay off to have an eye at Sec. 8.2.
The cascade model above is sectioned stepwise along a frequency, ω. The energy spectrum is a
distribution of the energy along a wave-number axis, κ. This spectrum can also be subdivided
stepwise so that it becomes a cascade. A wave number κn represents the interval (κn− , κn+ ), which
− − +
equals (κn− , κn+1 ). The next step has wave number κn+1 and represents (κn+1 , κn+1 ), and so on.
Each step in Magnussen’s cascade contains the energy of all the subsequent steps. We can imagine
that smaller eddies move inside bigger eddies. The energy for a cascade level becomes
Z ∞
3 2
u
2 n = E(κ)dκ. (11.16)
κn−
Using this for the next step as well, we see that

Z aκn+1
3 2 3 2 3 2 wn+1
2 u n − 2 u n+1 = 2 u n 1 − 2w = E(κ)dκ. (11.17)
n aκn
Here, we have used the transfer model, wn , Eq. (11.2), in the first equality. Except in the smallest
scales, wn+1 approximately equals wn , so that the parenthesis is equal to 1/2.
In order to compare the cascade with the energy spectrum, we have to relate the two models. We
express the length scale of the n-th level by L n = (b · κn )−1 , where b is a constant. Furthermore, we
−
let each step represent a wave-number interval between κn− = aκn and κn+ = κn+1 = aκn+1 . In this
way, the wave-number cascade is subdivided such that each step represents an interval, not the rest
of the spectrum as in Magnussen’s cascade.
Here, the parameters a (0 < a < 1) and b are not a part of the cascade model as such, but only aids
to relate the two models. The value of the parameter a has to be chosen and it determines the shape
of the cascade along the wave-number axis.
The dissipation primarily takes place in the fine structures. In the cascade model, q ∗ = 43 ε, cf. Eq.
(11.6). This is the dissipation from the last level, which can be written
Z ∞
q∗ = D(κ)dκ, (11.18)
aκ∗
where κ∗ = (bL ∗ )−1 . From Eq. (8.8), we recall that D(κ) = 2νκ 2 E(κ).
This relates Magnussen’s cascade model to the three-dimensional energy spectrum E(κ). Now, the
cascade can be compared to experimental data for the inertial subrange and for the dissipative range.
An extensive outline is given by Ertesvåg and Magnussen (1997).
11.4 The fine structures 177
11.4 The fine structures

Here is more on mass and mass exchange in the fine structures. We make
further use of the cascade model of the preceeding section.
The ratio between the mass in the fine structures and the total mass is modeled as
∗ 3 !3/4
u 3CD2 νε 3/4
∗ −3/4
γ = = 2
= 9.8 ReT . (11.19)
u′ 4CD1 k2
We see from Eqs. (11.5) and (11.7) that this is γ ∗ ≈ L ∗ /L ′ . Equation (11.19) can be regarded as a
postulate. But we can also imagine this such that the fine structures constitute a sheet of thickness
L ∗ that is rolled around a cylindrical eddy of diameter L ′ . Then, γ ∗ is the ratio between the volume
(mass) of the sheet and that of the cylinder.
Figure 11.2: Artist’s expression of the mixing and reaction between fuel and oxidizer (oxy-
gen). From Grimsmo (1991:88).
We can imagine the fine structures as thin and long eddies or vortex threads (Tennekes’ model, see
page 120). In the flow, these threads will gather between larger eddies (cf. Fig. 11.2), such that
certain regions get much fine structure. Outside these regions, there is little fine structure. The mass
in the regions is larger than the mass of the fine structures themselves. The ratio between the mass
of these regions and the total mass is expressed as
!1/4
u∗ 3CD2 νε 1/4
−1/4
γλ = ′ = 2
= 2.1 ReT . (11.20)
u 4CD1 k2
γ ∗ and γλ is a kind of intermittency factors. They give the probability of finding fine structures or
fine-structure regions in a location.
The mass exchange between the fine structures and the surroundings, divided by the mass of the fine
structures, is

u∗ 3 1/2 ε 1/2 ε 1/2
ṁ ∗ = 2 ∗ = = 2.5 . (11.21)
L CD2 ν ν
We can imagine this mass exchange by thinking of an eddy with diameter L ∗ . Its inflow has the
velocity u ∗ over half the circumference and outflow over the other half. The volumetric flow into (or
out of), divided by the volume of the eddy, becomes 2u ∗ /L ∗ . With equal densities ρ, this will also
be the ratio between mass flow and mass.
The mass exchange between fine structures and surroundings, divided by the total mass, must be
ṁ ∗ γ ∗ :
!1/4
3 12CD2 νε 1/4 ε νε 1/4 ε
∗ ∗
ṁ = ṁ γ = 2 2
= 24 2 . (11.22)
4CD1 CD1 k k k k
The residence time of the fine structures is

1
τ∗ = . (11.23)
ṁ ∗
In published work on the combustion model, the numerical factors show a little variation. This is
due to various practicing of truncation. The notation will show some variation as well.
11.5 Reactor model

Here, we develop an expression for the reaction rate of a species. We use the
model from the preceeding sections.
We imagine that the chemical reactions take place in the fine structures. We will regard the fine struc-
tures as a well mixed reactor. Figure 11.3 sketches the reactor model. The mass inside the reactor
is MFS . The mass flows into and out of it are Ṁin and Ṁout . The reactor is regarded homogeneous.
The superscripts ∗ and o refer to reactor and surroundings. The entering mass has the properties of
the surroundings (Yko , T o , ρ o ) and the out-flowing mass has the properties of the reactor (Yk∗ , T ∗ ,
ρ ∗ ).
11.5 Reactor model 179
Q̇
❄ rad
Surroundings
MFS
ρ o , Yko , T o
Ṁin ✲ ✲ Ṁ
out
Fine structures
ρ ∗ , Yk∗ , T ∗
Figure 11.3: Schematical fine-structure reactor
The mass balance for a species k in a steady-state reactor can be written

MFS
Ṁin Y o − Ṁout Yk∗ = −Rk∗ ∗ . (11.24)
| {z k} | {z } ρ
mass k inflow mass k outflow
| {z }
mass k reacted
Here is Rk∗ the production rate1 for species k, that is, (−Rk∗ ) is the consumed mass of species k per
unit of time and volume. The volume of the fine structures is MFS /ρ ∗ . For a steady-state reactor,
Ṁin = Ṁout = Ṁ. From above, we have that ṁ ∗ is the mass flow into the fine structures (mass
exchange) divided by the mass of the fine structures, hence ṁ ∗ = Ṁ/MFS . Using this in Eq.
(11.24), we get
−Rk∗ = ρ ∗ ṁ ∗ (Yko − Yk∗ ). (11.25)
This is the reacted (consumed) mass of species k per unit of time and volume of fine structures.
In some papers, Rk∗ is called “net rate of mass exchange between the fine structures and the sur-
roundings” (i.e. for species k). This we see from Eq. (11.24): Net input of mass equals the reacted
mass.
Now we can find an expression for the mean reaction rate R k . This is the reacted mass of species k
per unit of time per unit volume of the entire fluid (reactor plus surroundings). We let ρ be the mean
density and Mtot the total mass. The reacted mass of k, divided by the total volume, is

∗ MFS Mtot −1 ρ
R k = Rk ∗ = ∗ γ ∗ Rk∗ . (11.26)
ρ ρ ρ
This is valid if all the fine structures react. This is not always the case. We have to assume that only
1 The sign follows from Eq. (1.5), where R is a positive source term when species k is produced. Notice that in some
k
papers by Magnussen and others, the reaction rate is written with the opposite sign, thus as a sink term.
a fraction χ of the fine structures react (0 ≤ χ ≤ 1), hence

ρ
Rk = χγ ∗ Rk∗ . (11.27)
ρ∗
With Eq. (11.25) and ṁ = γ ∗ ṁ ∗ , we now get
−R k = ρ ṁχ(Yko − Yk∗ ). (11.28)
Mean values and fine structures
Equation (11.28) expresses the reaction rate by Yko and Yk∗ . In simulations, however, we operate with
mean values. We have to establish a relation between mean values, values for the surroundings and
values for the reactor (fine structures).
Let us look at a region with turbulence and chemical reactions. We have said that γ ∗ is the mass of
the fine structures divided by the total mass in the region. Furthermore, (χγ ∗ ) must be the mass of
the reacting fine structures, divided by the total mass. Thus, this is the active reactor. The region we
look at has a mass Mtot and a volume Vtot. We write the mean density as ρ = Mtot /Vtot. The region
is subdivided into (active) reactor and surroundings. The active reactor has mass MR = χγ ∗ Mtot ,
and the surroundings has the remaining mass, Msurr = (1 − χγ ∗ )Mtot . The volume of the reactor is
MR 1 1
VR = = ∗ γ ∗ χ Mtot = ∗ γ ∗ χρVtot. (11.29)
ρ∗ ρ ρ
The volume of the surroundings is

Msurr 1 1
Vsurr = = o (1 − γ ∗ χ)Mtot = o (1 − γ ∗ χ)ρVtot . (11.30)
ρo ρ ρ
The two last equations, together with Vtot = VR + Vsurr , give

1 γ ∗χ 1 − γ ∗χ
= ∗ + . (11.31)
ρ ρ ρo
A volumetric average of a variable ϕ appears from

VR ∗ Vsurr o γ ∗χ 1 − γ ∗χ o
ϕ= ϕ + ϕ = ∗ ρϕ ∗ + ρϕ , (11.32)
Vtot Vtot ρ ρo
or
ϕ ϕ∗ ϕo
= γ ∗ χ ∗ + (1 − γ ∗ χ) o . (11.33)
ρ ρ ρ
For homogeneous and steady-state turbulence, the volumetric average is equal to the time and en-
semble averages (Hinze, 1975:5). In other cases, it is a good or less good approximation.
11.5 Reactor model 181
The mass average of a variable ϕ appears from

MR ∗ Msurr o
ϕ̃ = ϕ + ϕ , (11.34)
Mtot Mtot
which gives
ϕ̃ = γ ∗ χϕ ∗ + (1 − γ ∗ χ)ϕ o . (11.35)
This is the same as the Favre mean value (see Appendix B). This we see by introducing ρψ for ϕ
into Eq. (11.33). With the definition ψ̃ = ρψ/ρ, we obtain Eq. (11.35) for ψ̃.
A mean mass fraction has to be such a mass-averaged or mass-weighted quantity, thus
Ỹk = γ ∗ χYk∗ + (1 − γ ∗ χ)Yko . (11.36)
From this, we can resolve Yko and introduce it into Eqs. (11.25) and (11.28). This gives the reaction
rate of the fine structures and the averaged reaction rate:
ρ ∗ ṁ ∗
−Rk∗ = (Ỹk − Yk∗ ), (11.37)
1 − γ ∗χ
ρ ṁχ
−R k = (Ỹk − Yk∗ ). (11.38)
1 − γ ∗χ
The reaction rate is the source term of the equation for Ỹk :
∂ ∂
(ρ Ỹk ) + (ρ Ỹk u j ) = Diffusion(Ỹk ) + R k . (11.39)
∂t ∂x j
In several papers by Magnussen et al., the reaction rate is expressed by mass concentrations ρk
(often denoted ck ) instead of mass fractions. The mass concentration is a volumetric mean and,
hence, follow Eq. (11.33) with ϕ = ρk as the variable. We know that ρk = ρYk and hence,
ρk ρYk
= = Ỹk . (11.40)
ρ ρ
Furthermore, ρk∗ = ρ ∗ Yk∗ and ρko = ρ o Yko . This can be introduced in the expressions above, if
desired.
We are still left with the quantities Yk∗ and χ, which are unknowns. For χ there are several model
expressions, see especially the works from 1978, 1981 and 1989. The oldest (and simplest) can be
regarded as special cases or simplifications of the newer. In any case, the expression for χ is related
to how much combustion product that is present. Hot product is what ignite the reaction. Model
expressions from 1989 are listed in Sec. 11.8.
The mass fraction of the reactor, Yk∗ , can be determined in two different ways:
1) By presuming that the reactions are infinitely fast. We then can calculate Yk∗ from a simple
reaction equation (stoichiometric mass balance). This we will return to below.
2) By calculating more or less detailed on the chemical kinetics. This is something we and others
are working on. But here, no further considerations are made (see Sec. 10.3).
11.6 Fast reactions

The expression for the reaction rate in the previous chapter is valid for both
slow and fast reactions. When we presume that the reactions are infinitely fast,
the expression can be simplified.
Many reactions take place very fast. In the literature, we often meet the term fast chemistry limit.
This refers to infinitely fast reaction as a limiting value.
We can write an irreversible stoichiometric reaction as
1 kg fuel + r kg oxidizer → (1 + r ) kg product . (11.41)
Here, r is the stoichiometric amount of oxidizer for the relevant fuel (see page 29).
We will show that when the reaction is infinitely fast, the reaction rate for the fuel can be written
as
ρ ṁχ
−R fu = Ỹmin , (11.42)
1 − γ ∗χ
where
Ỹmin = min Ỹfu , r1 Ỹox . (11.43)
We return to the expression

−Rk∗ = ρ ∗ ṁ ∗ (Yko − Yk∗ ), (11.44)
which can also be written
ρ ∗ ṁ ∗
−Rk∗ = (Ỹk − Yk∗ ). (11.45)
1 − γ ∗χ
For fuel and oxidizer:

∗ ρ ∗ ṁ ∗
−Rfu = (Ỹfu − Yfu∗ ), (11.46)
1 − γ ∗χ
∗ ρ ∗ ṁ ∗ ∗
−Rox = (Ỹox − Yox ). (11.47)
1 − γ ∗χ
1) First, we look at a fuel-rich mixture: Yfuo > r1 Yox

o . The reaction is then limited by the oxygen.
∗
All the oxygen is used, so that Yox = 0. Then,
∗ ρ ∗ ṁ ∗
−Rox = Ỹox . (11.48)
1 − γ ∗χ
We also have
1 γ ∗χ
Yko = Ỹk − Y ∗. (11.49)
1−γ χ
∗ 1 − γ ∗χ k
11.6 Fast reactions 183
Of the fuel inflow, Yfuo , the amount reacted is 1r Yox

o , so that the leftover is
1 o
Yfu∗ = Yfuo − Yox . (11.50)
r
In the same way, the product increases as

1+r o
Ypr∗ = Ypro + Yox , (11.51)
r
and with Eq. (11.49) we can show that

1
Yfu∗ = Ỹfu − Ỹox , (11.52)
r
1+r
Ypr∗ = Ỹpr + Ỹox . (11.53)
r
Equation (11.52) gives Ỹfu − Yfu∗ = 1r Ỹox and hence,
∗ ρ ∗ ṁ ∗ 1
−Rfu = Ỹox . (11.54)
1 − γ ∗χ r
Equations (11.48) and (11.54) we will use below.
o > r Y o , the fuel amount will be the limiting factor. All the
2) If the mixture is rich on oxygen, Yox fu
∗
fuel is consumed: Yfu = 0.
Then is
∗ ρ ∗ ṁ ∗
−Rfu = Ỹfu . (11.55)
1 − γ ∗χ
As above, we find that
∗
Ỹox − Yox = r Ỹfu , (11.56)
Ypr∗ = Ỹpr + (1 + r )Ỹox . (11.57)
With this, we get

∗ ρ ∗ ṁ ∗
−Rox = r Ỹfu . (11.58)
1 − γ ∗χ
o = r Y o , Y ∗ = Y ∗ = 0, and Ỹ
3) If the mixture is stoichiometric, Yox fu ox fu ox = r Ỹfu . This makes Eq.
(11.58) equal to Eq. (11.48), and Eq. (11.55) equal to Eq. (11.54).
If we now define Ỹmin as in Eq. (11.43), both Eqs. (11.54) and (11.55) can be written
∗ ρ ∗ ṁ ∗
−Rfu = Ỹmin . (11.59)
1 − γ ∗χ
Equations (11.58) and (11.48) can both be written
∗ ρ ∗ ṁ ∗
−Rox = r Ỹmin . (11.60)
1 − γ ∗χ
∗ = r R∗ .
We see that Rox fu
Furthermore, we use the expressions R k = (ρ/ρ ∗ )γ ∗ χ Rk∗ and ṁ = γ ∗ ṁ ∗ , and obtain

ρ ṁχ
−R fu = Ỹmin , (11.61)
1 − γ ∗χ
and similarly, R ox = r R fu .
With mass concentrations (cf. Eq. (11.40)), we can write
ρ Ỹmin = ρY min = ρ min . (11.62)
Not much is said here about the product. We can determine R pr from the same procedure. But when
we have a simple reaction equation like Eq. (11.41), Ỹpr is given from Ỹfu and Ỹox . We then only
need to solve the equations for these two.
This presumption of fast reactions is a usual way to do, both in Magnussen’s model and in other
models. As long as we are mainly interested in fuel, air and energy, this is sufficient for very many
fuels. (Most solid and liquid fuels react in gaseous phase, exceptions are coke and char residues.)
Temperature in the fine structures
If we neglect the radiation heat transfer Q̇ rad to the reactor in Fig. 11.3, the reactor is adiabatic. The
energy balance gives
Ḣin = Ḣout, (11.63)

Ṁ C p (T o − Tref ) + Yfuo HR = Ṁ C p (T ∗ − Tref ) + Yfu∗ HR . (11.64)
Here, HR is the heating value of the fuel, and the specific heat capacity C p is assumed constant. We
let 1Yfu = Yfuo − Yfu∗ be the change in the fuel mass fraction (1Yfu > 0), that is, the consumed fuel,
and can rearrange to
HR
T∗ − To = (1Yfu ) . (11.65)
Cp
This energy balance is valid without presuming fast reactions. When the reaction is fast, we have
the two cases discussed above:
1) For a fuel-rich mixture, all oxygen is consumed: Then 1Yfu = r1 Yox
o , Y ∗ = 0 and
ox
o 1
Yox = Ỹox . (11.66)
1 − γ ∗χ
11.7 Extinction and blow-off 185
2) For a oxygen-rich mixture, all fuel is consumed: Then 1Yfu = Yfuo , Yfu∗ = 0 and
1
Yfuo = Ỹfu . (11.67)
1 − γ ∗χ
For both cases:

1 1
1Yfu = min Yfuo , r1 Yox
o
= min Ỹ 1
fu r ox =
, Ỹ Ỹmin . (11.68)
1−γ χ
∗ 1 − γ ∗χ
Furthermore, the temperature as a mass-weighted mean value, cf. Eq. (11.35), is
T̃ = γ ∗ χ T ∗ + (1 − γ ∗ χ)T o , (11.69)
and with Eqs. (11.65) and (11.68), we get the temperature in the fine structures,
HR
T ∗ − T̃ = Ỹmin . (11.70)
Cp
The fine-structure temperature T ∗ is for instance used in the soot model.
11.7 Extinction and blow-off

All chemical reactions will in practice have a certain reaction time. We often talk about a chemical
timescale τc . Above, we have defined a residence time or a hydrodynamic timescale, τ ∗ = (ṁ ∗ )−1 .
If the mass flow becomes so large that τc is larger than τ ∗ , the reactants will not have sufficient time
to be heated to reaction. The reaction is extinguished or blows off.
For the fine-structure reactor (Fig. 11.3), we can define a degree of reaction,
Yfuo − Yfu∗
Y = . (11.71)
Yfuo
When Y = 0, there is no reaction. If Y = 1, the reaction in the reactor is complete.

From the mass balance, Eqs. (11.24)–(11.25), we can write this as
Yfuo − Yfu∗ −Rfu∗ τ∗
Y = o = o ∗ . (11.72)
Yfu Yfu ρ
From energy balance, Eqs. (11.63)–(11.65), we get

Yfuo − Yfu∗ C p (T ∗ − T o )
Y = o = . (11.73)
Yfu HR Yfuo
Figure 11.4: Extinction in the fine-structure reactor, τ1∗ > τext

∗ > τ ∗ (sketch). From Byg-
2
gstøyl and Magnussen (1983).
These two expressions for Y can be plotted in a Y -T ∗ diagram, Fig. 11.4. The system has solutions
where the two curves cross each other. The reaction rate Rfu∗ = R ∗ (ρ ∗ , Y ∗ , Y ∗ , T ∗ ) can be calcu-
fu fu ox
lated with an Arrhenius expression, cf. Sec. 1.8 page 28 (see also for instance Kuo, 1986:115), or
by another chemical-kinetics model.
We can then calculate Y from the mass balance for different values of the residence time τ ∗ . The
energy balance gives a straight line in the diagram, independent of τ ∗ .
Where the two curves meet in one point, we have the limiting value for extinction; τc = τ ∗ = τext
∗ .
∗
For smaller τ , the system does not have any other solution than Y = 0, thus no reaction. For
τ ∗ > τc = τext
∗ the system has two solutions with Y > 0. The upper intersection gives a stable
solution.
Example:
From Eq. (11.21) we see that the timescale τ ∗ = (ṁ ∗ )−1 ∼ (ν/ε)1/2 . Near a wall, for instance in a
pipe or a combustion chamber, the dissipation ε is large. The timescale is the least at the wall and
there is a risk of local extinction.
In a free flow, for instance a jet, ε ∼ u ′ 3 /L ′ . If we regard L ′ as a mixing length, experiments in a
round jet show that L ′ ∼ b ∼ x and u ′ ∼ Us ∼ x −1 . Here, b is the jet width, Us is the centerline
velocity and x is the distance from the nozzle in the direction of the flow. This gives τ ∗ ∼ x 2 . When
x is small, there is a possibility that the residence time τ ∗ becomes too small. Local extinction at the
outlet gives a lifted flame.
11.8 Model expressions listed 187
11.8 Model expressions listed

Here follows a collected listing of the model expressions that you need for
implementing Magnussen’s combustion model EDC into a computer program
for turbulent flames.
The ratio between the mass of fine structures and the total mass:
∗ 3 νε 3/4
u
γ∗ = = 9.8 . (11.74)
u′ k2
The ratio between the mass of fine-structure regions and the total mass:
u∗ νε 1/4
γλ = ′ = 2.1 2 . (11.75)
u k
The mass exchange between the fine structures and the surroundings, divided by the mass of the fine
structures: ε 1/2
u∗
ṁ ∗ = 2 ∗ = 2.5 . (11.76)
L ν
The residence time in the fine-structure reactor: τ ∗ = 1/ṁ ∗ .
The mass exchange between the fine structures and the surroundings, divided by the total mass:
ṁ = γ ∗ ṁ ∗ . (11.77)
The mean reaction rate for a species, source term in the equation for the mean mass fraction Ỹk :
ρ ṁχ
Rk = − (Ỹk − Yk∗ ). (11.78)
1 − γ ∗χ
Only a fraction χ of the fine structures reacts. The model version from Magnussen (1989) is formu-
lated by Gran (1994:43) as
χ = χ1 · χ2 · χ3 , (11.79)
2
Ỹmin + Ỹpr /(1 + r )
χ1 = , (11.80)
Ỹfu + Ỹpr /(1 + r ) Ỹox /r + Ỹpr /(1 + r )
( )
1 Ỹpr /(1 + r )
χ2 = min · ,1 , (11.81)
γλ Ỹpr /(1 + r ) + Ỹmin
 
1  γ λ Ỹpr /(1 + r ) + Ỹmin 
χ3 = min ,1 . (11.82)
γλ  Ỹmin 
The three contributions to χ can be interpreted like this: χ1 is the probability that the reactants are
found together; χ2 expresses the degree of heating by products; χ3 limits the reaction due to lack of
reactants.
Ỹmin = min Ỹfu , 1r Ỹox . (11.83)
For fast chemical reactions, the model can be simplified to:

ρ ṁχ
R fu = − Ỹmin . (11.84)
1 − γ ∗χ
Gran (1994:44) rewrites χ Ỹmin into one joint expression, with the reaction rate for fast reactions:
ρ ṁχ1 n o
R fu = − min Ỹmin , Ỹpr /(1 + r ), Ỹmin + Ỹpr /(1 + r ) γλ . (11.85)
(1 − γ ∗ χ) γλ
This can be compared with Eq. (10.10). If you have several (partial) reactions in your calculations,
you have to identify fuel, oxidizer and product for each reaction.
If you want to do your calculations with chemical kinetics (that is, not assume the reactions to be
infinitely fast), you must determine the fine-structure values Yk∗ from a chemical-kinetics model.
Chapter 12
Etterrakst
In this chapter I have collected some topics that are not discussed in this book.
They are important, although left out to restrict the extent of the book. Some
of the topics are well presented other places. The purpose here is to give you
some starting threads for further work on these topics.
12.1 Numerical simulation

Techniques and methods for numerical simulation of flow problems is a large and wide subject by
itself. It will go too far to dig into it here, even for a brief overview.
In order to conduct a simulation, the physical processes have to be made clear. And to evaluate the
results of the simulation, one has to understand the processes one tries to obtain a picture of. This
is the aim of this book. One has to find the balances, sources and transport terms for the different
quantities.
Many simulations are based on the methods described by Patankar (1980). We have used this book
as the basic book in the course Numerical mass and heat transfer. Recently, we have used the book
by Versteeg and Malalasekera (1995). Both can beneficially be read besides this book. Furthermore,
you should be able to find some explanations and further references in the documentation of the
simulation code you are using – at least if it is a commercial code.
A certain overview of different methods you will get from the books of Anderson, Tannehill and
Pletcher (1984), Peyret (1996) and Ferziger and Perić (1996).
A notice to the terminology: Some authors distinguish between finite volume and finite difference
methods, while others think they are so close that both are termed finite difference (e.g. Anderson et
al., 1984).
Quite a few people will say that one has to use schemes with high order to ensure a reliable com-
190 12 Etterrakst
putation; at least second order. Othervice, there will be too much false, numerical diffusion. This is
correct, for a mathematician. (You have to learn from literature the meaning of terms like scheme
and order.)
The higher order the scheme has, the finer grid you need to obtain a stable solution. For practical
problems, you often do not have sufficient computational power for such fine grids. One thing is
to is to compute a channel with recirculation zones behind an obstacle or two. To compute the air
flow through an oil platform is something totally different. This means that the engineer has to
rely on “inaccurate” methods and “too coarse” grids. However, when the alternatives are way more
uncertain, the “inaccurate” numerical solution is not, after all, that bad.
12.2 Radiation
For non-premixed flames, radiation may contribute a large fraction of the heat transfer. Radiation is
important, for instance, in fires and flares, and in stoves and furnaces. A large fraction of radiation
in flames comes from soot (see below). Flames with much soot give much yellow light.
At off-shore oil and gas installations and at refineries there is always a flare where gas is burned in
emergency situations. The radiation from the flare should be below certain limits [kW/m2 ] at loca-
tions where humans are staying. The limits differ according to the length of the stay: working places,
control rooms, escape routes, etc. There are also limits for radiation onto load-bearing structures.
Radiation is often modeled by assuming grey surfaces. When the gases are included (as participating
media), grey gases is the most usual approximation. You can read about grey surfaces and grey gases
in a book on heat transfer, e.g. Mills (1995), Bejan (1993) or Incropera and DeWitt (1990). There
are also models that include effects of the spectral distribution of the radiation, however mainly at a
research stage. A main obstacle is their computational demand and that they are not yet parallelized.
In her doctoral work, Ge Song (1996) treated the spectral distribution of radiation in flames, where
also further references can be found.
Lockwood and Shah (1981) have developed a method for treatment of radiation in flow problems.
The method can be used both with grey-gas models and for spectral models.
Radiation is a challenge in numerical codes for flow. Most other physical and chemical processes
have effects in the nearby region, for instance chemical reactions, molecular motions, and the effects
of body forces. Radiation has effects over a distance, such that one has to consider the path of
the rays. Surfaces and molecules in the flow affect a large region. This means that it is difficult
to subdivide the domain into sub-domains to parallelize the solution of the system of equations.
Mathematically, this is manifested as an integral term in the differential equation for energy. This
integral spans over the entire flow field and changes from location to locations. If you manage to
solve this problem, honor and attention can be expected.
12.3 Soot and unburned hydrocarbons 191
12.3 Soot and unburned hydrocarbons

Some decades ago, chimneys with clouds of soot and other smoke were the symbol of wealth and
industrial development. With time, it became more a nuisance – and eventually a health problem.
Soot and soot formation is a very complex problem where quite a few relations still are unclear. Up
to now, the models are relatively simple and not very numerous. One of them is Magnussen’s soot
model, which is closely related to the combustion model (EDC), see Magnussen (et al.) (1971, 1973,
1975a, 1976, 1978). Kennedy (1997) gives an overview of the state of soot modeling.
We can read more on soot for instance in the textbooks by Glassman (1996:398f), Warnatz, Maas
and Dibble (1996, 1999) and Chomiak (1990), and in the anthology edited by Bockhorn (1994).
12.4 Measurements
The purpose of measurements in flames can be to learn more about the physical and chemical pro-
cesses in the combustion. In more specific cases, it can be to monitor the operation of a plant or the
emissions from it. The measurements are usually indirect: For instance, when we measure tempera-
ture with an ordinary mercury thermometer, we actually measure a length.
We can classify measurements according to whether they take place inside or outside the combustion.
Measurements in the flue gas, e.g. temperature and composition, take place after the combustion.
Also the heat transfer is measured outside the combustion zone. In order to study details of the
combustion, one has to measure the quantities at many locations instantaneously and over time. To
characterize the combustion, one has to know the distribution of temperature, flow velocity, species
concentration and density. For laminar flames it can be sufficient to measure steady-state values of
these quantities. For turbulent flames, both average and fluctuating values in time and space have to
be obtained.
According to the method, one can distinguish between
– methods that use a probe (censor) inside the area of the measurement,
– optical methods that analyze natural light (or other radiation) from the flames or (white) light
sent through the flames,
– electric methods,
– laser-based methods that use reflected laser light, or radiation from molecules/atoms that are
excited by the laser light.
Details and references can, for instance, be found in Beér and Chigier (1972) – on simple experi-
mental methods, Eckbreth (1988) – on laser measurements, and Chomiak (1990) and Taylor (1993)
– on all the methods. The latter two, and Doebelin (1990), also give an introduction to aims and
procedures of experimental methods.
192 12 Etterrakst
12.5 Fuels
The term fuel covers a wide range of substances. Roughly 90% of all commercial primary energy is
from fossil fuels. Coal is 27%, oil is 40% and natural gas is 23%. (Not all agree that oil and natural
gas are fossil.) Furthermore, non-commercial energy, mainly fuelwood, is estimated to add another
5–10 %.
The reader knows these fuels and can read more other places. Statistics for reserves, production and
consumption of different (commercial) energy courses can be found for instance in the annual issues
of BP Statistical review of world energy.
Here, I will just point to the relation between mass and energy: We defined (page 29) the stoichio-
metric amount of air as the mass of air that is required to burn one kg of a fuel. For hydrogen and
hydrocarbons this number is approximately 15. This means that to transport the chemically bound
energy, one has to carry a small fraction of the substances that are involved in the chemical reaction.
The reaction products are released to the atmosphere. With an electric battery, in comparison, we
have to carry with us all the reacting substances, and we have to bring the products back home. An-
other side of the same matter: When a camel in the desert burns 1 kg fat from its hump, it produces
more than 1 kg water in the process.
12.6 Two and multiphase

This book is on flows in one phase, either gaseous or liquid. The picture then is really complex.
When the flow has two or more phases, a real challenge is faced. Gas bubbles in liquid flow and
particles or droplets in gaseous flow are very common. There can be two liquid phases, e.g. oil and
water; or there can be several phases, e.g. gas, oil, water, sand and foam in a separator on an oil
platform. When we inspect particles in a gas or a liquid, the form and size can have considerable
influence. There is a large difference between spheres and long threads (fibers). Also the volumetric
ratio between the phases is important; ranging from e.g., small air bubbles in a water flow, through
large bubbles, sluggish air/water flow, large droplets and small droplets to a water mist in an air flow.
Similarly, there is an smooth transition from “particles” to “solid bodies” in a flow.
Chomiak (1990) provides a long section on combustion of liquid and solid fuels, but says little about
the flow. The treatment of dust and clouds of droplets seems to be the areas where research has made
the largest advances. Such flows are found in e.g. powerplants and engines, where coal powder or
liquids are blown into a combustion chamber or furnace. Soot and other particles in the flue gas
is a similar problem. Some newer review articles can be a suitable entrance to this topic: Bachalo
(1994), Ayyaswamy (1995), Annamalai et al. (1992, 1993, 1994). Beér, Chomiak and Smoot (1984)
give a useful, although not updated, discussion of flows with particles (coal) and combustion. All
these articles contain extensive lists of references. See also in the section on soot above.
There are few or no models that are so general that all types of multiphase flow can be simulated.
Besnard and Harlow (1988) treat turbulence in multiphase flow and consider a somewhat wider
range of volume fractions. You have to search the literature to find someone who has treated what
is of your interest. A starting point may be to sit down with the latest volumes of the International
Journal of Multiphase Flow. Then there should be a hope to find a lead for further work.
12.7 Out into practice – What now? 193
12.7 Out into practice – What now?

Now you have read and learned what is written in this book. (Or: let us imagine that you have done
so.) First, you have seen even more things that you do not know much about. This is normal – and
intended: Your horizon is extended, you have experienced something called (scientific and personal)
development, hopefully learned something – and may be become a bit more humble towards all the
things you do not know.
Second, you may be left with the question: – What now? – How do I apply this? This, you will have
to find out yourself. This might be the beginning. Quite few have to read and learn much more on
the topic, or rather, special parts of it. Possibly you will turn to a path where this is “useful insight
into another field of work”, but not more. Or perhaps this forms an important part of the basis of
your professional life. Nevertheless, you can be sure that flows are turbulent for your lifetime.
With the basis given by this book, combined with previous courses, you should be well equipped to
explore more specialized literature. This holds both when you want to dig into theoretical studies,
work on experimental studies or practical devices with flow and combustion. Unless you are so
excessively sharp that you catch up this course with little efforts, you might have learned something
about yourself as well.
I any case, you will have to gather some practical experiences from the field you are going to work
on. Such experience you gain by practicing yourself and by learning from others. The lessons
learned from this book improve your ability to put experiences into a context and to understand the
background for why things are as they are.
In this book, I have barely touched numerical solution of the systems of equations – often called
simulation (“simulare”: to make like). There are some people who develop their own code from
scratch. Most workers use codes developed by others. There are a large number of such codes, and
nearly anyone can get results from a code. The codes can be graded from fairly simple programs that
you get (nearly) free, to advanced, commercial program systems. The former you are allowed to use
on your own responsibility, they have little or no do documentation and a high user threshold. The
latter can be equipped with a multitude of aids, user interface, model variants, well-tested procedures
and detailed documentation – and not least, a suction pipe into your budget. In any case you will
have to understand both the problem specification and the modeling in order to be able to interpret
and make use of the results you achieve.
Can this be used for something? In this book I have tried to communicate a certain amount of
formalistic knowledge – developments and such things, some understanding and insight – and some
scepticism.
Commercial programs have many models implemented, and documentation with equations and ref-
erences to articles where the models are presented. But you rarely get any clear advises on when
and where the different models are to be used. This is because the advice has to vary from case to
case, with the aims of the user, requirements on accuracy, available resources and so on. This has to
be evaluated by the user. And it is left to the user to aquire more knowledge and experience with the
194 12 Etterrakst
models.
Computations with numerical programs and turbulence models can be easy to perform, but hard to
do well. Such computations do not provide “the truth”. They have to be interpreted based on the
actual flow case and the circumstances. A simulation can usually not replace measurements of the
same case. On the other hand: Transferring or extrapolating experimental data from one case to
another can lead to false conclusions. Mathematical models give much better predictions than the
best qualified guess.
We usually lack detailed measurements for the specific case. Modeling and numerical simulation is
then the best alternative you have – and it is a good tool.
Appendix:
Equations that we love
A collection of various equations and expressions

for the benefit and pleasure of researchers and students
The intention here is not that you should learn anything or find anything new. The content is only the
things that all of us “know” from before. It is not meant as an introduction but rather a handbook.
It is an attempt to give an overview of the relevant equations and expressions. Other presentations
are found, for instance, in Bird, Stewart and Lightfoot (1960) (partially reproduced by Kuo, 1986)
and in Williams (1985). Many books show such equations, but the large majority are restricted to
equations for flows without chemical reactions.
One of the points of this appendix is that the equations and the information should be fully correct,
and that all relevant presumptions and simplifications made should be clearly stated. This ambition
is, however, within certain limits: I did not write this for nuclear scientists and people hovering at
very high altitudes. The target group is engineering scientists caring for earthly flows and reason-
ably continuous fluids. Moreover, I did not have very high pressure close to mind. An important
restriction is single-phase flows.
However, experience tells that there is always one more error. When you find errors or deficiencies,
have hints for improvements or wishes/suggestions for extensions, I would like to hear from you
(see page vi).
A study of a subject is in many ways also a study of language. Therefore, I have included a small
dictionary, Appendix D.
Appendix A
The fundamental equations – and

something more
A.1 Mass balance
A.1.1 Continuity equation
General form for single-phase flow:
∂ρ ∂
+ (ρu j ) = 0, (A.1)
∂t ∂x j
or
∂ρ ∂ρ ∂u j
+uj +ρ = 0. (A.2)
∂t ∂x j ∂x j
| {z }
Dρ
Dt
For a flow with constant density ρ Eq. (A.1) becomes

∂u j
= 0. (A.3)
∂x j
About the distinction between variable density and compressible flow, see page 26.
Reformulating the convective and transient terms
The convective and transient terms in the equation for a general scalar variable ϕ can be written as
∂ ∂
Cϕ = (ρϕ) + (ρϕu j ). (A.4)
∂t ∂x j
198 Appendix A The fundamental equations – and something more
By subtracting Eq. (A.1) times ϕ (the product is zero), we obtain

∂ϕ ∂ϕ Dϕ
Cϕ = ρ + ρu j =ρ . (A.5)
∂t ∂x j Dt
This rewriting is valid for flows both with constant and variable density.
A.1.2 On mass, amount of substance and volume

Note: The SI or ISO-standard is mol, not kilomol (kmol), for amount of substance, and kilogram
(kg), not gram (g), for mass (ISO 31-0:1996).
The mass fraction for a species k in a mixture is defined as
mk
Yk = . (A.6)
m
and the mole fraction is defined as

nk
Xk = . (A.7)
n
Here, m and n are, respectively, the total mass and amount of substance and m k and n k are the mass
and amount of species k. The relations between them are n = m/M, n k = m k /Mk . The molar mass
(the mass of one mole) for a species is Mk , and the molar mass for a mixture of NS species is
NS
X
M= X k Mk . (A.8)
k=1
The mass and mole fractions are related by

mk n k Mk X k Mk X k Mk
Yk = = PN = PN = , (A.9)
m S
n i Mi S
X i Mi M
i=1 i=1
nk m k /Mk Yk Yk /Mk
Xk = = = M = PN . (A.10)
n m/M Mk S
Yi /Mi i=1
I the last equality, the following expression was used

NS
X NS
X NS
X NS
Yi mi ni 1 X n 1
= = = ni = = .
Mi m Mi m m m M
i=1 i=1 i=1 i=1
The mass concentration for a species k is

mk
ρk = = Yk ρ. (A.11)
V
A.1 Mass balance 199
The amount-of-substance concentration (also known as the concentration, mole concentration or

molarity) is expressed as
nk nk Yk ρ
ck = = = . (A.12)
V m k /(Yk ρ) Mk
Actually, both these quantities are now and then referred to as the «concentration». Here, V is
volume and ρ is density (or mass density – when the context requires this clarification). When
summing over all species k of the mixture,
X
Yk = 1, (A.13)
k
X
X k = 1, (A.14)
k
X X
ρk = ρ Yk = ρ. (A.15)
k k
An amount of gas can be specified as a volume at a certain, defined state (pressure and temperature).
The most common are
Standard cubic meter, defined at 15 ◦C (288.15 K) and 101.325 kPa (1 standard atmosphere). This is
defined as the Standard reference conditions for natural gas (ISO 13443:1996) and for measurements
of gases and liquids from crude oil (ISO 5024:1976).
Normal cubic meter, defined at 0 ◦ C (273.15 K) and 101.325 kPa (1 atm).
Notice that when the heating value is given per unit of volume, the heating value can possibly be
calculated at one temperature (e.g. 25 ◦ C) and the volume at another temperature (e.g. 15 ◦ C).
There are also other definitions of standard or normal state; that is, with somewhat different values of
pressure or temperature. In British or American standards, the temperature is often 60 ◦ F (15.56 ◦C).
ISO 13443:1996 gives rules for conversion between different standards.
A.1.3 Equation for species mass fraction, concentration – Mass diffusion

When the fluid is a mixture of different species in the same phase, we can formulate a continuity
equation for each species,
∂ρk ∂ ∂
+ (ρk u j ) = (− jk, j ) + Rk , (A.16)
∂t ∂x j ∂x j
or, expressed in terms of the mass fraction Yk (ρk = ρYk ) as

∂ ∂ ∂
(ρYk ) + (ρYk u j ) = (− jk, j ) + Rk . (A.17)
∂t ∂x j ∂x j
If we prefer to keep the density apart from the reaction rate, we can introduce the notation Rk = ρωk .
The mass diffusion jk, j has three different contributions:

(i) Diffusion due to gradients of concentration, often expressed in terms of a diffusion velocity
(ρk Vk, j = ρYk Vk, j ) and modeled by Fick’s law (see below).
(ii) Mass diffusion due to a gradient of temperature, known as thermodiffusion or the Soret effect.
(iii) Mass diffusion due to a pressure gradient, pressure diffusion.
In certain reacting flows, (ii) and (iii) are of importance, but in combustion these contributions both
are small and can be neglected. See Warnatz, Maas and Dibble (1999:58,155,158f) and more refer-
ences there.
Fick’s law
When neglecting the Soret effect (thermodiffusion) and the pressure diffusion, the diffusive flux can
be modeled by Fick’s law:
∂Yk
− jk, j = −ρk Vk, j = −ρYk Vk, j = ρD . (A.18)
∂x j
Note: For constant ρ, but only then, this can be written as −ρk Vk, j = D∂ρk /∂ x j .
The diffusion coefficient or mass diffusivity Dk will vary from substance to substance and it will
be a function of the mass fractions of all other species in the mixture, see for instance Warnatz,
Maas and Dibble (1999:56f,159). This gives a system of coefficients that may become completely
unmanageable. Therefore, we use one common value D for all the species in the mixture (Kuo,
1986:206f; Libby and Williams, 1980:5).
With a common diffusion coefficient D, Eq. (A.16) can be written on the form
∂ ∂ ∂ ∂Yk
(ρYk ) + (ρYk u j ) = (ρD ) + Rk . (A.19)
Here, the Schmidt number Sc = µ/(ρD) can be introduced.
Summing over all species k in the mixture gives

X X
ρk Vk, j = ρ Yk Vk, j = 0 (A.20)
k k
and X
Rk = 0. (A.21)
k
Summing Eq.(A.16) over all species k in the mixture gives the continuity equation, Eq. (A.1).
A.2 Momentum 201
A.2 Momentum
A.2.1 Equation for momentum
On the terms momentum and moment, see page 25.
General form of the equation for single-phase flow:
∂ ∂ ∂
(ρu i ) + (ρu i u j ) = (σi j ) + ρ f i , (A.22)
∂t ∂x j ∂x j
σi j = − pδi j + τi j , (A.23)
or
∂ ∂ ∂p ∂
(ρu i ) + (ρu i u j ) = − + (τi j ) + ρ f i . (A.24)
A mixture
P of substances with different acceleration due to body forces f k,i will have an acceleration
fi = k Yk f k,i . This is relevant in systems with electromagnetic forces. On electric fields and
flames, see for instance Chomiak (1990:378) and Calcote and Gill (1994).
For a Newtonian fluid the stress tensor can be written

τi j = µ + +λ δi j , (A.25)
where µ and λ are the first and second coefficients of viscosity. This can be rewritten to

τi j = µ + + (µ B − 32 µ) δi j , (A.26)
where µ B = λ + 32 µ is the bulk viscosity. The Stokes hypothesis implies that λ = − 23 µ, or µ B = 0.

Truesdell (1952:288ff) summarized the background of and discussion on this expression; see also,
for instance, Tritton (1988:58,72) and Aris (1962:112).
The term stress is used both for τi j and for σi j . The former is the viscous stress tensor and the latter
is the total stress tensor.
In order to distinguish µ from the turbulence viscosity, the term (dynamic) molecular viscosity can
be used. The term laminar viscosity is misleading since µ also applies to turbulent flows.
Note: The literature shows some variation in use of terms and symbols for the quantities denoted as
λ and µ B here. The words can even be interchanged!
Now, the equation can be written as

∂ ∂ ∂ ∂u k ∂u i ∂u j
(ρu i ) + (ρu i u j ) = − pδi j + (µ B − 32 µ) δi j + µ + + ρ f i . (A.27)
∂t ∂x j ∂x j ∂ xk ∂x j ∂ xi
This is known as the Navier–Stokes equation (or equations, as there are three components), alterna-
tively with µ B = 0. The term Navier–Stokes equations is also used for the system of Eq. (A.27), the
continuity equation and the energy equation.
For flows with constant density ρ, the continuity equation gives ∂u

∂ x k = 0, and the momentum equa-
k
tion can be written as

∂u i ∂u i 1 ∂p ∂ ∂u i ∂u j
+uj =− + ν + + fi . (A.28)
∂t ∂x j ρ ∂ xi ∂x j ∂x j ∂ xi
Here, the equation is divided by the density. When the viscosity is constant as well, we get

∂u i ∂u i 1 ∂p ∂ ∂u i
+ uj =− +ν + fi . (A.29)
∂t ∂x j ρ ∂ xi ∂x j ∂x j
Body force acceleration can be expressed from a potential, 9, as

∂9
fi = − . (A.30)
∂ xi
The body force is often the gravitational force. With the coordinate z directed towards the gravity
acceleration (upwards), the potential becomes 9 = gz, and the acceleration
∂z
f i = −g . (A.31)
∂ xi
For a Cartesian coordinate system with x 3 = z, we can write fi = −gδi3 .
A.2.2 Euler equation, Bernoulli equation

If viscous forces are neglected, Eq. (A.24) becomes
∂ ∂ ∂p
(ρu i ) + (ρu i u j ) = − + ρ fi . (A.32)
∂t ∂x j ∂ xi
This is the Euler equation. As the term with the second derivative is omitted, we can divide by the
density:
∂u i ∂u i 1 ∂p
+uj =− + fi . (A.33)
∂t ∂x j ρ ∂ xi
The equation is still valid regardless of density varying or not, cf. Eq. (A.5).
The Bernoulli equation can be developed from the Euler equation. For a non-transient and one-
dimensional flow along a streamline in the x s direction:
∂u s 1 ∂p
us =− + fs . (A.34)
∂ xs ρ ∂ xs
The body force is the gravity that acts against the coordinate direction z. The acceleration in the x s
direction becomes
∂z
f s = −g cos α = −g . (A.35)
∂ xs
A.2 Momentum 203
Here, g is the gravity acceleration and α is the angle between the unit vectors for the coordinate
directions x s and z. When assuming the density ρ and the gravity acceleration g as constants, we
can write
∂ 1 2 p
u + + gz = 0, (A.36)
∂ xs 2 s ρ
or, integrated along the streamline,
1 2 p
2 u s + ρ + gz = constant. (A.37)
This is the Bernoulli equation. Notice that this is a special case of the momentum equation. The
equation can also be developed from the general energy equation, Eq. (A.42), or from Eq. (A.43) for
mechanical energy.
A.2.3 Pressure
The pressure enters quite a few equations and relations. However, the quantity called pressure can
differ between different contexts. For the «insiders», the meaning will be obvious, however not
always for everyone else. It may be convenient to list some definitions and clarify terminology.
The static pressure, also known as the thermodynamic pressure, is the quantity p found in the equa-
tion of state (for an ideal gas, p = ρ RT ); in the the definition of enthalpy, h = e + p/ρ; in the
total stress tensor, Eq. (A.23); and in the Bernoulli equation, Eq. (A.37). The gradient of the static
pressure is found in the momentum equation.
Notice that in numerical computations, it can be convenient to include more terms together with
the pressure. If we introduce Eqs. (A.26) and (3.26) for the viscous and turbulence stresses into
the mean
momentum equation, Eq. (B.34), we can identify a term with the gradient of the quantity
p − 32 µ − µ B + 32 µt + 23 ρ̄ k̃ ∂ ũ l /∂ xl . This might be the quantity that is calculated as the «pres-
sure» in the computer code. It can also be beneficial to subtract a chosen, constant reference pres-
sure. The numerical value for pressure is large, and the deviation from a reference pressure can be
more convenient to work with than the absolute pressure. When the absolute pressure is required,
for instance in an equation of state or a polytrophic process, care must be taken to insert the right
quantity.
The dynamic pressure is the quantity 21 ρu 2i , which is found, for instance, in the Bernoulli equation.
The hydrostatic pressure or gravitational pressure is used for the quantity ρgh, where h is the eleva-
tion difference between two locations and g is the gravity acceleration (the absolute value). The use
of the term can show some variation as it can mean ρref gh, where ρref is a (constant) reference value.
In liquids the difference is small as the density is approximately constant. The term hydrostatic
pressure can also be used for the quantity p + ρgh or p + ρgz, which is the pressure of a quiescent
fluid, cf. Eq. (A.37).
A hydrostatic pressure distribution is the gradient ∂ p/∂z = −gρref , where z is directed upwards.
The stagnation pressure is also known as the total pressure or the effective pressure. This is the sum
pt = p + 12 ρu 2i + ρgz. (A.38)
In many instances of flow, the latter term is neglected, either because it is small or because its
variation is small. In hydrostatics, there is no velocity and the second term is zero.
The pressure head is the pressure recalculated into the height of a liquid column, usually of wa-
ter or mercury. Examples are the static pressure head p/(ρv g) and the dynamic pressure head
1 2
2 ρu i /(ρv g), where ρv is the density of the liquid in the column.
The gauge pressure (AE: «gage») is the pressure that is measured, that is the overpressure, the
pressure above the atmospheric pressure (from «gauge»: graded scale for measurement, for instance
on a glass tube in a liquid barometer).
A.2.4 Boussinesq approximation for buoyancy

Note: The term Boussinesq approximation is used in at least three different meanings: 1) an ap-
proximation for the buoyancy term in the momentum equation, like here; 2) an approximation for
turbulence stresses, a model with turbulence viscosity (Boussinesq viscosity), see Sec. 2.5; and 3)
an approximation in wave mechanics.
In the Boussinesq approximation for buoyancy, all variations in fluid properties (viscosity, specific
heat capacity, thermal conductivity, diffusivity, etc.) are neglected, except the density. Moreover, all
variations in density are also neglected, except when these give buoyancy.
Put more simple: The density is assumed constant ρ = ρ◦ in all terms except the buoyancy term,
where it is ρ = ρ◦ + 1ρ. Furthermore, the buoyancy term is split into an elevation part and a
temperature part:
∂9 ∂
ρ f i = −(ρ◦ + 1ρ) =− (ρ◦ 9) + 1ρ fi . (A.39)
∂ xi ∂ xi
Here, 9 is the body force potential introduced in Eq. (A.30). The elevation term can be included into
the pressure term. Since the potential due to gravity is linear, the gradient is a constant - however
not the absolute value. A distinction between the two pressure quantities is important mainly when
the pressure is used as a boundary condition.
The change of density due to temperature can be expressed by
1ρ = ρ − ρ◦ = −αρ◦ 1T = −αρ◦ (T − T◦ ), (A.40)
where α is the coefficient of volume expansion (volume expansivity), see a book on thermodynamics.
For an ideal gas, α = T −1 .
Now we can write the momentum equation as

∂ ∂ ∂ ∂ ∂u i ∂u j
(ρ◦ u i ) + (ρ◦ u i u j ) = − ( p + ρ◦ 9) + ρ◦ ν + + (−ρ◦ α)(T − T◦ ) f i .
∂t ∂x j ∂ xi ∂x j ∂x j ∂ xi
(A.41)
We often direct the coordinate x 3 upwards, against the gravity, so that we can write fi = −gδi3 , cf.
Eq. (A.31).
A.3 Energy balance 205
A discussion of the conditions for the Boussinesq approximation is found, for instance, in Tritton
(1988:164f,188f). There are several conditions, but one is 1ρ/ρ◦ ≪ 1 . Opinions on how much
«much less than» is, can be diverse. This depends on how large deficiencies we can accept before
we regard the approximation as not good enough. A possible and not unusual practice is to set a
limit such that 1ρ/ρ◦ should be less than 0.15–0.25. Many flows in buildings (ventilation), in the
atmosphere and in the ocean will be within this limit. Another condition is that the density varies
with temperature only. In fires and other combustion cases, the density can vary with a factor of
10–30, both due to temperature and due to composition.
In the formulations above, the variation in density, and hence in buoyancy, are caused by temperature
variations. Similar expressions can be developed for small variations in concentrations, for instance,
salinity in sea water.
A.3 Energy balance

A.3.1 General equation of energy
Total energy
A general energy equation, the first law of thermodynamics on differential form for single-phase
flow:
∂ ∂ ∂q j ∂ ∂ X
(ρet ) + (ρet u j ) = − + Q̇ − ( pu j ) + τi j u i + ρ Yk f k,i (u i + Vk,i ). (A.42)
k
et = e + 21 u i u i is the internal energy and kinetic energy. The equation is developed by, among
others, Kuo (1986:200). The development of a differential form of the equation requires that spatial
derivatives exist for q j , p, τi j and u i ; then the equation is generally valid for single-phase flow.
The term Q̇ is an «inner» source, where among other, the net transferred radiation heat is included.
Notice that the reaction heat does not enter here, as it is a part of the internal energy (see below).
Formally, radiation heat can also be formulated like the diffusive flux: Q̇ rad = −∂q j,rad/∂ x j .
Somewhat inexact, et is also called total energy. But we might have included the potential energy in
et as well. The last term on the right-hand side is work conducted by body forces. The first part of
this term can be reformulated to a changeP of potential energy and it can be moved to the left-hand
side: Using Eq. (A.30), we can write ρ( k Yk f k,i )u i = ρ f i u i = −ρu i ∂9/∂ x i . This term could be
rewritten so that we got an equation for (et + 9).
Mechanical energy
An equation for mechanical energy 12 u i u i is obtained by multiplying the momentum equation by u i :

∂ 1 ∂ ∂p ∂τi j
(ρ u i u i ) + (ρ 1 u i u i u j ) = −u i + ui + ρ fi u i . (A.43)
∂t 2 ∂x j 2 ∂ xi ∂x j
Internal energy
Subtracting Eq. (A.43) from Eq. (A.42) provides an equation for internal energy:
∂ ∂ ∂q j ∂u j ∂u i X
(ρe) + (ρeu j ) = − + Q̇ − p + τi j +ρ Yk fk,i Vk,i . (A.44)
| {z } k
8
The lastPterm cancels when P the body-force acceleration fk,i is equal for all species, f k,i = f i ;
thus, ρ k Yk fk,i Vk,i = ρ fi k Yk Vk,i = 0. This is the case when gravity is the only body force.
When, for instance, electromagnetic forces are in action, different species can have different f k,i , cf.
Eq. (A.24). 8 is the dissipation function, which is the heat produced from viscous friction.
Total enthalpy
An equation for the total enthalpy h t = et + p/ρ = e + p/ρ + 21 u i u i is obtained by adding

∂p ∂
∂t + ∂ x j ( pu j ) to both sides of Eq. (A.42):
∂ ∂ ∂p ∂ ∂q j X
(ρh t ) + (ρh t u j ) = + τi j u i − + Q̇ + ρ Yk f k,i (u i + Vk,i ). (A.45)
∂t ∂x j ∂t ∂x j ∂x j
k
Static enthalpy
An equation for static enthalpy, h, is obtained by subtracting the equation for mechanical energy,
Eq. (A.43) from Eq. (A.45), thus h = h t − 12 u i u i . Alternatively, ∂p ∂
∂t + ∂ x j ( pu j ) can be added to
both sides of Eq. (A.44), thus h = e + p/ρ:
∂ ∂ ∂p ∂p ∂u i ∂q j X
(ρh) + (ρhu j ) = + ui + τi j − + Q̇ + ρ Yk fk,i Vk,i . (A.46)
∂t ∂x j ∂t ∂ xi ∂x j ∂x j
| {z } | {z } | {z } k
Dp 8
ρ Dh
Dt Dt
An equation for static enthalpy can also be obtained as a simplification of Eq. (A.45). If we can
assume that the kinetic energy ( 12 u i u i ) is much less than the total enthalpy h t , then h t will be ap-
proximately equal to the static enthalpy, h t ≈ h. This equation for static enthalpy will have source
terms somewhat different from those of Eq. (A.46).
Thermal enthalpy
P
For a mixture, the enthalpy is h = Yk h k . This can be split into chemical and thermal enthalpies:
k
X X
h = h ◦ + 1h = Yk h ◦f,k + Yk 1h k , (A.47)
k k
where h ◦f,k is the enthalpy of formation and 1h k is the thermal enthalpy for species k, see Sec. A.3.2.
An equation for thermal enthalpy is obtained from the species mass fraction equation , Eq. (A.17),
multiplying by h ◦f,k , and summarize over all species k:
X X X X
∂ ◦ ∂ ◦ ∂ ◦
ρ h f,k Yk + ρ h f,k Yk u j = −ρ h f,k Yk Vk, j + h ◦f,k Rk . (A.48)
∂t ∂x j ∂x j
k k k k
| {z } | {z }
h◦ h◦
The enthalpy of formation h ◦f,k is a constant and can therefore be included in the derivative. Sub-
tracting Eq. (A.48) from Eq. (A.46) provides
∂ ∂ Dp ∂q j X
(ρ1h) + (ρ1h u j ) = +8− + Q̇ + ρ Yk f k,i Vk,i
∂t ∂x j Dt ∂x j
k
X
∂ ∂h ◦
− ρD − h ◦f,k Rk . (A.49)
∂x j ∂x j
k
This is the equation for thermal enthalpy. Fick’s law, Eq. (A.18), was introduced in the second last
term of Eq. (A.48). The last term can be written as HR Rbr , see Eq. (A.69). Regarding the f k,i term,
see Eq. (A.44).
In general, 1h is a function of pressure, temperature and species mass fractions and hence, the term
«thermal» might be a little misleading. For an ideal gas, or at constant pressure, the enthalpy of a
species is a function of the temperature only; h k = h k (T ), see below.
Thermal internal energy

P
The internal energy for a mixture is e = k Yk ek . Like enthalpy, internal energy can be split into a
chemical and a thermal component:
X X
e = e◦ + 1e = Yk ek◦ + Yk 1ek . (A.50)
k k
P
Starting from Eq. (A.44) for internal energy e, an equation for e◦ = k Yk ek◦ , cf. Eq. (A.48), is
subtracted and the result can be formulated as
∂ ∂ ∂q j ∂u j X
(ρ1e) + (ρ1e u j ) = − + Q̇ − p +8+ρ Yk f k,i Vk,i
k
X
∂ ∂e◦
− ρD − ek◦ Rk . (A.51)
∂x j ∂x j
k
This is thePequation for thermal internal energy. Here, Fick’s law is introduced in the second last
term: −ρ k ek◦ Yk Vk, j = ρD∂e◦ /∂ x j . The last term can be written as UR Rbr , see Eq. (A.73).
Regarding the f k,i term, see Eq. (A.44).
As long as nothing is said about τi j , q j , Vk, j , Q̇ or 1h k , Eqs. (A.42)–(A.51) are general for single-
phase flow. This includes, for instance, high Mach number, real gases and non-Newtonian fluids.
The total stress tensor can be split into pressure and viscous stresses: σi j = − pδi j + τi j .
A.3.2 Enthalpy and internal energy in combustion – specific heat capacity

Thermal enthalpy; specific heat capacity
P
For a mixture the enthalpy is h = k Yk h k . The enthalpy of a species is h k = h ◦f,k + 1h k , where
h ◦f,k is the enthalpy of formation (chemical enthalpy) for species k at the reference state (T ◦, p◦ ).
This is a constant reference value for each species. The thermal enthalpy 1h k is the deviation, that
is 1h k = h k (T, p)−h k (T ◦, p◦ ) = 1h k,(T ◦, p◦ )→(T , p) . The reference state is usually T ◦ = 298.15 K
(25 ◦ C), p◦ = 0.1 MPa. Some databases use, however, the absolute zero as reference temperature.
In general, the enthalpy of a species is a function of both pressure and temperature. For an ideal gas,
or at constant pressure, the enthalpy is a function of the temperature only; h k = h k (T ). Then,
Z T
◦
h k = h f,k + C p,k d T. (A.52)
T◦
For a real gas, h k = h ◦f,k + 1h k,(T ◦, p◦ )→(T , p◦ ) + 1h k,(T , p◦ )→(T , p). The two first terms are similar
to those of Eq. (A.52). For the pressure term, see a thermodynamics book, for instance Sonntag and
VanWylen (1982:399, 1991:373), Moran and Shapiro (1993:462,1998:503).
For a pure substance the specific heat capacity (often called specific heat) at constant pressure is
defined as
∂h ∂h k
Cp = or C p,k = . (A.53)
∂T p ∂T p
P
Introducing h = k Yk h k into the definition, Eq. (A.53a):
X
∂h ∂h k X ∂Yk
Cp = = Yk + hk (A.54)
∂T p ∂T ∂T p
k k
or X
X ∂Yk
Cp = Yk C p,k + hk . (A.55)
∂T p
k k
For some reactions, the last term might have some influence. However, if the equilibrium is shifted
strongly towards the product side of the reaction balance, the last term can be neglected. This is
usually the case in combustion (for the overall/global reaction) and hence, we use the approximation
X
Cp = Yk C p,k . (A.56)
k
Thermal internal energy; specific heat capacity gas constant

P
The internal energy for a mixture is e = k Yk ek . Like enthalpy, internal energy can be split into
a chemical and a thermal component: ek = ek◦ + 1ek , where ek◦ = ek (T ◦, p◦ ). For an ideal gas,
ek = ek (T ), and Z T
ek = ek◦ + Cv,k d T. (A.57)
T◦
For a pure substance the specific heat capacity (often called specific heat) at constant volume is
defined as
∂e ∂ek
Cv = or Cv,k = . (A.58)
∂T v ∂T v
For a mixture, the approximation X

Cv = Yk Cv,k , (A.59)
k
can be used. See, however, the discussion of C p above.
The relation between internal energy and enthalpy is h = e + p/ρ = e + RT .
For a species in a mixture, the relations are
pk p Xk
h k = ek + Rk T = ek + = ek + , (A.60)
ρk ρ Yk
h ◦f,k = ek◦ + Rk T ◦ (A.61)
and
1h k = 1ek + Rk (T − T ◦ ). (A.62)
The gas constant for a species is R = Ru /M, or Rk = Ru /Mk , so that for a mixture mixture,
X X Yk
R= Yk Rk = Ru . (A.63)
Mk
k k
Notice that this is not an approximation, contrary to the expressions for C p and Cv above. The
universal gas constant Ru is independent of substances and temperature.
Heating value and enthalpy of formation
More comprehensive descriptions can be found in textbooks on thermodynamics, for instance Son-
ntag and VanWylen (1982:469, 1991:440), Moran and Shapiro (1993:587,1998:639).
The reaction enthalpy at the reference temperature is often denoted by h ◦RP (or h RP0 ) in books on
thermodynamics. It is given as enthalpy per kg of fuel or per kmol of fuel. The heating value HR for
a fuel is given as enthalpy per kg fuel, or often for gases, per normal or standard m3 (page 199).
The relation between the heating value and the enthalpy of formation, per kg of fuel, can be expressed
as

◦ 1 X ◦
X
◦
HR = −h RP = n k h̄ f,k − n k h̄ f,k
m fu
R P

1 X ◦
X
◦
= m k h f,k − m k h f,k . (A.64)
m fu
R P
Here, the sums are taken over all reactants (R) and reaction products (P) that occur in the stoichio-
metric reaction. n k is the amount of substance and m k is the mass of reactants and reaction products.
m fu is the mass of fuel.
For a fuel mixture, the heating value can be calculated by summing the contributions from the
different species. The composition might be known in terms of mass fractions, Yk , or mole fractions,
X k (= volume fractions for ideal gases). If the heating value of each species is known, the heating
value for the mixture can be found from
X 1 X 1 X
HR,mix = −h ◦RP,mix = Yk HR,k = X k Mk HR,k = X k (−h̄ ◦RP )k (A.65)
Mmix Mmix
k k k
Here, Mmix is the molar mass of the fuel mixture, cf. Eq. (A.8).
These expressions are valid as long as the species composition of the fuel is known. When the
composition is unknown, the heating value has to be determined from experiments. If the elemental
composition, empirical formulas can be used, giving approximate values for the heating value. Dif-
ferent molecular bindings have different binding energies and hence, such formulae can not be very
accurate.
Combustion can often be written as a simple, one-step reaction:

1 kg fuel + r kg oxidizer → (1 + r ) kg product . (A.66)
The heating value is

HR = h ◦f,fu + r h ◦f,ox − (1 + r )h ◦f,pr . (A.67)
The oxidant is usually air, which is a mixture of N2 , O2 and small amounts of noble gases, H2 O
and CO2 . For O2 , N2 and noble gases (but not for CO2 and H2 O), the enthalpy of formation is by
definition zero. This means that h ◦f,ox ≈ 0.
The reaction rates in the reaction of Eq. (A.66) are related by
1 1
Rfu = Rox = − Rpr , (A.68)
r 1+r
so that
X
h ◦f,k Rk = h ◦f,fu Rfu + h ◦f,ox Rox + h ◦f,pr Rpr
k

= h ◦f,fu + r h ◦f,ox − (1 + r )h ◦f,pr Rfu = HR Rfu . (A.69)
This can be used in the last term of Eq. (A.49) for thermal enthalpy. Notice that Rfu < 0.
Internal energy and reaction energy
The reaction energy can be expressed as

1 X X
UR = −u ◦RP = n k ēk◦ − n k ēk◦
m fu
R P
X X
1
= m k ek◦ − m k ek◦ . (A.70)
m fu
R P
For the relation between internal energy and enthalpy, see Eqs. (A.60)–(A.61).
The difference between heating value and reaction energy can be expressed as
X X X X
m br (HR − UR ) = n k (h̄ ◦f,k − ēk◦ ) − n k (h̄ ◦f,k − ēk◦ ) = Ru T ◦ nk − nk (A.71)
R P R P
For a simple, one-step reaction, Eq. (A.66), the reaction energy becomes
◦ ◦ ◦
UR = ebr + r eoks − (1 + r )epr . (A.72)
Notice that the internal energy ek◦ is not zero for gases like O2 , N2 and noble gases (cf. enthalpy of
formation).
With Eq. (A.68),
X
◦
ek◦ Rk = efu ◦
+ r eox ◦
− (1 + r )epr Rfu = UR Rfu , (A.73)
k
cf. Eq. (A.69). This is the last term of Eq. (A.51) for thermal internal energy. Notice that Rfu < 0.
A.3.3 Enthalpy gradient, energy gradient

P
The enthalpy for a mixture is h = k Yk h k . In general, the enthalpy for a substance is a function of
both pressure and temperature. For an ideal gas, or at constant pressure, the enthalpy is a function
of the temperature only; h k = h k (T ). Then,
Z T
h k = h ◦f,k + C p,k d T. (A.74)
T◦
The gradient becomes

∂h X ∂h k X ∂Yk
= Yk + hk , (A.75)
k k
where Eq. (A.74) gives Z

∂h k ∂ ◦ ∂ T
= h + C p,k d T . (A.76)
∂x j ∂ x j f,k ∂ x j T◦
| {z }
=0
The first term is zero since the enthalpy of formation h ◦f,k is a constant reference value.
From mathematics we can find a general relation
Z v(x j )
∂ ∂v ∂u
f (t)dt = f (v(x j )) · − f (u(x j )) · , (A.77)
∂ x j u(x j ) ∂x j ∂x j
or, in this case Z T

∂h k ∂ ∂T
= C p,k d T = C p,k · . (A.78)
∂x j ∂x j T ◦ ∂ xj
In the present case, the last term of Eq. (A.77) becomes zero since T ◦ is a constant reference value.
As before, C p,k = C p,k (T ). Now, Eqs. (A.78) and (A.76) are introduced into Eq. (A.75):
∂h X ∂T X ∂Yk
= Yk C p,k + hk . (A.79)
k k
In
P the first right-hand-side term, the temperature gradient can be taken outside the sum. Usually,
k Yk C p,k = C p can be used; see Eqs. (A.55)–(A.56), page 208. This gives
∂h ∂T X ∂Yk
= Cp + hk . (A.80)
k
From Eq. (A.74) to Eq. (A.79) no approximations or models are introduced. However, as said,
Eq. (A.74) assume ideal gases or a constant pressure.
With the same procedure and conditions, it can be shown that the gradient of internal energy e =
P
k Yk ek , can be expressed as
∂e ∂T X ∂Yk
= Cv + ek . (A.81)
k
A.3.4 Model for heat flux

The diffusive heat flux entering the energy equation has three contributions:
(i) Heat conduction (molecular thermal diffusion), modeled by Fourier’s law.
(ii) Flux of energy due to mass diffusion. This is an amount of energy carried by the flux of
mass ρYk Vk, j of a species k in direction x j . Here, thermodiffusion (Soret effect) and pres-
sure diffusion is neglected (cf. page 200). The model is Fick’s law, as in the continuity
equation for species k (the Yk equation), see Eqs. (A.18)–(A.19), page 200: ρYk Vk, j =
−ρD∂Yk /∂ x j , with the same mass diffusion coefficient D for all species (Kuo, 1986:206f;
Libby and Williams, 1980:5).
(iii) The Dufour effect: heat flux due to concentration gradients. This effect is not not much
investigated, but seems to of little importance. It is (almost) always neglected (Kuo, 1986:198;
Libby and Williams, 1980:8).
∂T X
q j = −λ + h k ρYk Vk, j (A.82)
∂x j
| {z } | k {z }
(i) (ii)
∂T X ∂Yk
= −λ − ρD hk . (A.83)
∂x j ∂x j
k
λ is the thermal conductivity [W/mK] for the mixture. It can be expressed from the conductivities
for the individual gases, λk , see Warnatz, Maas and Dibble (1999:54).
The heat flux is a function of the temperature gradient. This is convenient when we reformulate the
energy equation to a temperature equation. However, often we prefer to use an enthalpy equation,
and in many instances, we would rather express the heat flux as a function of the enthalpy gradient.
It can be useful to define a thermal diffusivity, α = λ/(C p ρ). This has the same dimension as the
mass diffusivity D and the kinematic viscosity ν (can be regarded as a «momentum diffusivity»).
A.3.5 Enthalpy equation with enthalpy gradient

For some numerical simulations, it is convenient to have a gradient diffusion term. For the enthalpy
equation, the heat flux then have to be reformulated to a function of the enthalpy gradient.
Above we had the model
∂T X ∂Yk
−q j = λ + ρD hk . (A.84)
∂x j ∂x j
k
Equation (A.80) for the enthalpy gradient in an ideal gas, or at constant pressure, can be rearranged
to
∂T 1 ∂h 1 X ∂Yk
= − hk . (A.85)
∂x j Cp ∂x j Cp ∂x j
k
This relation introduced into Eq. (A.84) gives

λ ∂h λ X ∂Yk X ∂Yk
−q j = − hk + ρD hk , (A.86)
Cp ∂x j Cp ∂x j ∂x j
k k
or X
λ ∂h λ ∂Yk
−q j = − − ρD hk . (A.87)
Cp ∂x j Cp ∂x j
k
Here, the thermal diffusivity α (see above) can be introduced. Furthermore, the Prandtl number, Pr,
Schmidt number, Sc, and the Lewis number, Le (see Sec. C.4), can also be introduced:
µ ν µ ν λ/C p α Sc
Pr = = Sc = = Le = = = , (A.88)
λ/C p α ρD D ρD D Pr

µ ∂h 1 1 X ∂Yk
−q j = −µ − hk . (A.89)
Pr ∂ x j Pr Sc ∂x j
k
When Sc = Pr, that is Le = 1, the last term is zero. This form of the relation is frequently used.
Notice, however, that across solid surfaces there are (usually) no mass diffusion. This means that
the Lewis number is not unity but infinite. Then we have to turn back to Eq. (A.82) and put term (ii)
equal to zero, that is Vk, j = 0 normal to the wall.
With Eqs. (A.87) or (A.89), the heat-flux term of the enthalpy equation can be reformulated to a
function of the enthalpy gradient. Equation (A.46) for static enthalpy becomes
!
Dh Dp ∂ µ ∂h ∂ µ 1 X ∂Yk X
ρ = +8+ − 1− hk + Q̇ + ρ Yk fk,i Vk,i .
Dt Dt ∂ x j Pr ∂ x j ∂ x j Pr Le ∂x j
k k
(A.90)
This equation is, accordingly, valid for ideal gases or at constant pressure, cf. Eq. (A.85) and
Eqs. (A.79)–(A.80).
Similarly, Eq. (A.49) for thermal enthalpy
X X
D1h Dp ∂ µ ∂1h
ρ = +8+ + Q̇ + ρ Yk fk,i Vk,i − h ◦f,k Rk
Dt Dt ∂ x j Pr ∂ x j
k k
◦ !
∂ µ 1 ∂h ∂ µ 1 X ∂Yk
− 1− − 1− 1h k . (A.91)
∂ x j Pr Le ∂ x j ∂ x j Pr Le ∂x j
k
This equation also becomes much simpler with unity Lewis number. The third last term (reaction
term) can be written as HR Rfu (recall that Rfu < 0), see Eq. (A.69). On the f k,i term, see Eq. (A.44).
A.3.6 Heat flux with gradient of internal energy

Also when solving an equation for internal energy, a gradient diffusion term might be convenient.
Writing the heat flux in terms of the energy gradient is not as straightforward as for the enthalpy.
The energy flux with mass diffusion is still a function of the enthalpy, and thermal diffusion is still a
function of C p , not of Cv .
The starting point is again the model for heat flux in Eq. (A.83). The temperature gradient can be
expressed in two different ways:
1) From Eq. (A.80), an enthalpy gradient is obtained that can be reformulated to an energy gradient:

∂T 1 ∂h 1 X ∂Yk 1 ∂e 1 ∂ p 1 X ∂Yk
= − hk = + − hk . (A.92)
∂x j Cp ∂x j Cp ∂x j Cp ∂x j Cp ∂x j ρ Cp ∂x j
k k
This gives the heat flux as

X
λ ∂e λ ∂ p λ ∂Yk
−q j = + − − ρD hk . (A.93)
Cp ∂x j Cp ∂x j ρ Cp ∂x j
k
The last term is zero if Sc = Pr, that is Le = 1. The second last term can not in general be neglected.
If more suitable, p/ρ = RT can be used.
2) The energy gradient is found in Eq. (A.81). There, the energy ek occurs whereas the heat flux
model contains h k . These have to be coordinated, and we get
∂T 1 ∂e 1 X ∂Yk 1 ∂e 1 X ∂Yk 1 X ∂Yk
= − ek = − hk + Rk T . (A.94)
∂x j Cv ∂ x j Cv ∂x j Cv ∂ x j Cv ∂x j Cv ∂x j
k k k
Now, the heat flux becomes

X
λ ∂e λ ∂Yk λ X ∂Yk
−q j = − − ρD hk + Rk T , (A.95)
Cv ∂ x j Cv ∂x j Cv ∂x j
k k
or X
λ ∂e λ ∂Yk X ∂Yk
−q j = − − ρD ek + ρD Rk T . (A.96)
Cv ∂ x j Cv ∂x j ∂x j
k k
One of these expressions can be introduced into Eq. (A.44) for internal energy, or in Eq. (A.51) for
thermal internal energy, cf. Eqs. (A.90)–(A.91). However, in this instance, it is quite hard to «get rid
of» the additional gradient terms.
Note the difference between Eq. (A.93) and Eq. (A.96) regarding the use of C p and Cv .
A.3.7 Equation for temperature

P
The enthalpy of a mixture is h = k Yk h k . In general, the enthalpy for a substance is a function of
both pressure and temperature. For an ideal gas, or at constant pressure, the enthalpy is a function
of the temperature only; h k = h k (T ). Then,
Z T
h k = h ◦f,k + C p,k d T, (A.97)
T◦
and the gradient, see Eqs. (A.74)–(A.80), is

∂h k ∂T ∂h ∂T X ∂Yk
= C p,k and = Cp + hk . (A.98)
∂x j ∂x j ∂x j ∂x j ∂x j
k
Similarly, the total derivative becomes

Dh DT X DYk
= Cp + hk . (A.99)
Dt Dt Dt
k
Using Eq. (A.46) for static enthalpy:

Dh Dp ∂u i ∂q j X
ρ = + τi j − + Q̇ + ρ Yk f k,i Vk,i , (A.100)
Dt Dt ∂x j ∂x j
k
and the continuity equation for a species k, multiplied by h k and summed over all species k:
X DYk X ∂ X
hk ρ = hk −ρYk Vk, j + h k Rk . (A.101)
Dt ∂x j
k k k
When subtracting Eq. (A.101) from Eq. (A.100), the result is an equation for the temperature.
On the right-hand side, the intermediate expressions can be used
∂q j X ∂
− − hk −ρYk Vk, j
∂x j ∂x j
k
!
∂ ∂T ∂ X X ∂
= λ − h k ρYk Vk, j − hk −ρYk Vk, j
∂x j ∂x j ∂x j ∂x j
k k
X
∂ ∂T ∂T
= λ + −ρYk Vk, j C p,k . (A.102)
k
Here, Eq. (A.82) is introduced for the heat flux, and Eq. (A.78) is used in the last term. The equation
becomes
X
DT Dp ∂ ∂T
ρC p = + λ + 8 + Q̇ + ρ Yk fk,i Vk,i
Dt Dt ∂x j ∂x j
k
X X ∂T
− h k Rk + ρ −Yk Vk, j C p,k . (A.103)
∂x j
k k
The assumptions made to achieve this equation was ideal gases (or constant pressure) and that the
Dufour heat flux was neglected. Fourier’s law is introduced and Fick’s law can readily be introduced
in the last term. However, nothing is said about the stress tensor τi j , the heat source Q̇, the body
force f k,i or the specific heat capacity C p,k .
P
Using Fick’s law, the last term of Eq. (A.103) is modeled as ρD k ( ∂Y ∂T
∂ x j C p,k ) ∂ x j ; with the same
k
diffusion coefficient D for all species k. Alternatively, if the specific heat capacity is the same for
all species,
P C p,k = C p = C p (T ), the last term is zero, cf. Eq. (A.20). The second last term is then
equal to k h ◦f,k Rk , cf. Eqs. (A.74) and (A.21). This in turn, can be reformulated to HR Rbr , see
Eq. (A.69).
A.3.8 Equation for thermal enthalpy and heating value

The static enthalpy
P is the enthalpy
P of formation and the thermal enthalpy summed over all species of
a mixture; h = k Yk h k = k Yk (h ◦f,k + 1h k ). Equation (A.46) for static enthalpy has no source
term due to chemical reactions. This is contrary to Eq. (A.49) for thermal enthalpy and Eq. (A.103)
for temperature. A drawback of the static enthalpy is that the numerical value does not give an
obvious meaning. For many mixtures h is negative, since many species have a negative enthalpy of
formation.
A more comprehendible quantity is the sum of heating value and thermal enthalpy, Yfu HR + 1h.
This is the enthalpy that is, or can become, thermal enthalpy, and that can be extracted as heat, for
instance from a stove.
An equation for this quantity is obtained by multiplying the equation for Yfu , Eq. (A.17), with HR :

∂ ∂ ∂ ∂
(ρYfu HR ) + (ρYfu HR u j ) = ρD (Yfu HR) + Rfu HR . (A.104)
Here, Fick’s law is introduced and HR (a constant value) is taken inside the derivatives. Then, this
is added to Eq. (A.49) for thermal enthalpy, giving
∂ ∂
(ρ(Yfu HR + 1h)) + ρ(Yfu HR + 1h)u j
∂t ∂x j
X
Dp ∂ λ ∂
= + 8 + Q̇ + ρ Yk fk,i Vk,i + (Yfu HR + 1h)
Dt ∂x j Cp ∂x j
k
!!
∂ λ ∂ ∂h ◦ X ∂Yk
− − ρD (Yfu HR + 1h) + − 1h k . (A.105)
∂x j Cp ∂x j ∂x j ∂x j
k
In the development of the right-hand side, Eq. (A.87) for the heat flux was used, and h = h ◦ + 1h
was split, Eq. (A.47). The reaction terms of Eq. (A.49) and Eq. (A.104) cancel against each other,
cf. Eq. (A.69). The development is made for a simple, one-step reactionP as in Eq. (A.66). If multiple
fuels are considered, the contributions can be summed. Then, a term br Yk HR,k will replace the
term Ybr HR in the equation above. The last term of the equation cancels if the Prandtl and Schmidt
numbers are equal (unity Lewis number).
Remark: The quantity (Ybr HR + 1h) is suitable for presenting a result. Provided that the reactions
are simple and complete overall reactions like Eq. (A.66), the equation is conveniently used. The
substances can readily be classified as fuel, oxidizer or product. Difficulties can appear when the
same species is a product in some reactions and a reactant in some others, cf. Eq. (1.10), page 27.
Equation (A.105) is hard to make general, whereas Eq. (A.90) for h readily can be extended to
include a large number of reactions. Moreover, if an equation for static enthalpy h is solved, the
quantity (Ybr HR + 1h) can be calculated from the final result.
A.4 A short note on tensor notation and coordinates

In order to save writing, and to have a better overview of the equations, we
can use tensor notation. Section D.5 includes a short glossary of coordinate
systems.
Cartesian tensor notation
The equations of this book, and in many others, are written in Cartesian tensor notation. The reason
is that most equations get a simpler outlook in this way. We start out with a three-dimensional
Cartesian coordinate system:
x3
x2
x1
A vector has three components, for instance a space vector and a velocity vector:
Vector form: xE , component form: [x 1 , x 2 , x 3 ] or [x, y, z].
Vector form: uE, component form: [u 1 , u 2 , u 3 ] or [u, v, w] or [u x , u y , u z ].
The stress tensor has nine components and can be written as

   
τ11 τ12 τ13 τx x τx y τx z
τ =  τ21 τ22 τ23  or τ =  τ yx τ yy τ yz  . (A.106)
τ31 τ32 τ33 τzx τzy τzz
In Cartesian tensor form, a vector (1st order tensor) is denoted with one index, for instance
x i with i = 1, 2 or 3; x i = [x 1 , x 2 , x 3 ],
u i with i = 1, 2 or 3; u i = [u 1 , u 2 , u 3 ].
The stress tensor is a 2nd order tensor and get two indices:
τi j with i = 1, 2 or 3 and j = 1, 2 or 3.
A second-order tensor, like τi j , has 9 (32 ) components in a three-dimensional room. A third-order

tensor, ai j k , has 27 (33 ) components, a fourth-order tensor, bi j km , has 81 (34), and so on. We rarely
encounter tensors with order three or higher.
A.4 A short note on tensor notation and coordinates 219
The indices are chosen arbitrarily, so that u i is the same as u j or u n – that is, provided that the
indices i , j and n are not used elsewhere in the term in question. Similarly, τi j is the same as τik ,
τ j k or τ pn . Each of the indices has the value 1, 2 or 3.
Example: In general, τ13 6= τ23 . These are two (of nine) individual components of the tensor.
Nevertheless, we have τi j = τik = τ pn . In the latter instance, each of the three symbols represents
the entire tensor (all the 9 components), and it is one and the same tensor. If we in addition say that
(i, j ) = (1, 3) and ( p, n) = (2, 3), it is then obvious that τi j 6= τ pn . In that case, we are back to
talking about individual components of the tensor.
When two indices in a tensor or a term are the same, we should sum over these indices from 1 to 3,
for instance
3
X
u i u i should be read as u i u i = u1u 1 + u 2u 2 + u 3u 3,
i=1
3
X
∂u i ∂u i ∂u 1 ∂u 2 ∂u 3
should be read as = + + ,
∂ xi ∂ xi ∂ x1 ∂ x2 ∂ x3
i=1
3
X
∂u i ∂u i ∂u i ∂u i ∂u i
uj should be read as uj = u1 + u2 + u3 .
∂x j ∂x j ∂ x1 ∂ x2 ∂ x3
j =1
This is known as Einstein’s summation rule.

The last example has two j -s but only one i . The single index should not be summed. The sum on
the right-hand-side represents 3 different sums, one or each value of i . We recognize the convection
term of the momentum equation. This can be regarded as three equations, one for each of the
components. In the momentum equation, the index for the component (here u i ) is the same in all
terms of the equation. However, indices to be summed over can vary from term to term. When the
convection term is u j (∂u i /∂ x j ), the stress term can be written as ∂τi j /∂ x j or ∂τik /∂ x k , but not
as ∂τki /∂ x i or ∂τik /∂ x i . It is more orderly when we use the same index for summing in different
terms.
More examples:
ui ui = u j u j , but generally u i u i 6= u i u j ,
∂u i ∂u n ∂u i ∂u k
= , but generally 6= .
∂ xi ∂ xn ∂ xi ∂x j
The expressions to the right are unequal since one term has one index that is repeated, and hence
should be summed, whereas the other have two different indices and should not be summed. The for-
mer represent one single quantity each, which is the sum of three terms. Each of the latter represent
9 different combinations of the two indices, that is, 9 different quantities.
In a term of an equation it does not matter whether it reads ∂u i /∂ x i or ∂u n /∂ x n , that is, provided the
indices i and n are not used otherwise in the term.
Some equations contain terms with two or more pairs of repeated indices. Then you go on summing.
The product u i u i has 3 terms that should be summed, τi j (∂u i /∂ x j ) has 32 =9 terms, ai j b j k cik has
33 =27 terms to be summed, and so on.
Two special tensors are the Kronecker delta δi j and the permutation tensor ǫi j k :
 
1 0 0
1 when i = j
δi j = or δi j =  0 1 0  , (A.107)
0 when i 6= j
0 0 1

 1 when (i, j, k) is (1,2,3), (2,3,1) or (3,1,2) ,
ǫi j k = −1 when (i, j, k) is (3,2,1), (2,1,3) or (1,3,2) , (A.108)
 0 when i = j and/or i = k and/or j = k.
For two-dimensional cases, the indices get the values 1 or 2, rather than 1, 2 or 3. For one-
dimensional cases all indices get the value 1.
This section is on Cartesian tensors. It is common to present the equations on Cartesian form, also
when using non-Cartesian coordinate systems.
Tensor analysis, both Cartesian and general, is further described in, for instance, Tyldesley (1975),
Aris (1962), Irgens (1982) and Sokolnikoff (1964). The former two are the easier to digest, and are
also more related to fluid mechanics, whereas the latter two have a more general aim.
Cylinder coordinates, spherical coordinates
Certain forms of the fundamental equations are presented in cylinder coordinates (polar coordinates)
and in spherical coordinates by Bird, Stewart and Lightfoot (1960:83,317) (also reproduced by Kuo,
1986:175–201).
Appendix B
Equations for turbulent flow

B.1 Decomposition into turbulent and mean fields: Reynolds and Favre
Reynolds (1895) decompose the field into mean values (average values) and fluctuations (deviations
from the mean values):
u i = u i + u ′i , p = p + p′ , ϕ = ϕ + ϕ ′ , . . . , etc. (B.1)
A mean value can be a time mean, an ensemble mean or a volume mean, see Sec. 2.2 page 34.
Common for all these is that u ′ i ≡ 0, p′ ≡ 0, ϕ ′ ≡ 0, and so on.
Introducing this into the fundamental equations provides the turbulence equations or Reynolds equa-
tions, Sec. 2.3.
Favre (1965) include the density ρ into the mean value and defines the decomposition as
u i = ũ i + u ′′i , ϕ = ϕ̃ + ϕ ′′ , . . . , etc., (B.2)
where
ρu i ρϕ
ũ i ≡ , ϕ̃ ≡ , . . . , etc. (B.3)
ρ̄ ρ̄
Here, ρu ′′ = 0, ρϕ ′′ = 0 and ϕe′′ = ρϕ ′′ /ρ̄ = 0, but in general, u ′′ , ϕ ′′ , etc. are non-zero. The
overline ( ) denotes the usual average as above. The equations obtained by introducing this into the
fundamental equations can be called the Favre equations, or mass-averaged or mass-weighted mean
equations. For the equations, the averaging procedure is the same as for the Reynolds equations.
If ρ ′ = 0, the Favre decomposition is equal to the Reynolds decomposition: ũ = u, ϕ̃ = ϕ and
u ′′ = u ′ = 0, ϕ ′′ = ϕ ′ = 0, and so on.
222 Appendix B Equations for turbulent flow
Rules for ordinary averages (Reynolds averages)
A variable is decomposed into a mean value and a fluctuation: φ = φ + φ ′ , ψ = ψ + ψ ′ .

Rules can be developed from the definition of mean value, Eqs. (2.2), (2.3) or (2.4):
φ = φ, (B.4)
φ + ψ = φ + ψ. (B.5)
This rule means that an equation can be averaged by averaging each of the terms. Furthermore,
φ · ψ = φ · ψ, (B.6)
ψ φ ′ = ψ · φ ′ = 0, (B.7)
φ · ψ = φ · ψ, (B.8)
φ ψ = φ · ψ + φ′ ψ ′, (B.9)
dφ dφ
= , (B.10)
ds ds
Z Z
φds = φds, (B.11)
!
φ 1 φ
=φ· = . (B.12)
ψ ψ ψ
Example on developing a rule: Equation (B.7) by using the definition in Eq. (2.4).
Z ∞ Z ∞
ψ φ′ = ψ (φ(c) − φ) f (c)dc = ψ · φ ′ (c) f (c)dc = ψ · φ ′ = 0.
−∞ −∞
Since ψ is already integrated, it can be taken outside the integration. The mean value of a
fluctuation is zero.
B.1 Decomposition into turbulent and mean fields: Reynolds and Favre 223
Rules for Favre averages or mass-weighted Reynolds averages
A variable is split as φ = φ̃ + φ ′′ , with the definition
ρφ
φ̃ ≡ . (B.13)
ρ̄
Here, the overline ( ) denotes the same average as shown above.

By using Eqs. (B.13) and (B.8), the following can be expressed
ρφ ρφ
ρ φ̃ = ρ · = ρ̄ · = ρ̄ φ̃. (B.14)
ρ̄ ρ̄
Equation (B.13) can be formulated as ρ̄ φ̃ = ρφ, so that
ρ φ̃ = ρφ = ρ̄ φ̃. (B.15)
Since ρφ = ρ φ̃ + ρφ ′′ is also valid, then ρφ ′′ = 0, but φ ′′ 6= 0 if ρ ′ 6= 0. It can be shown that

ρ̄φ ′′ = ρ̄ φ − φ̃ = ρ̄φ − ρφ = −ρ ′ φ ′ . (B.16)
Other results:
ρ φ̃ ψ̃ = ρ̄ φ̃ ψ̃, (B.17)
ρφ
ρ φ̃ψ ′′ = · ρψ ′′ = φ̃ρψ ′′ = 0, (B.18)
ρ̄
φ̃ψ ′′ = φ̃ · ψ ′′ (6= 0 if ρ ′ 6= 0), (B.19)
ρφψ ′′ = φ · ρψ ′′ = 0, (B.20)
φψ ′′ = φ · ψ ′′ (6= 0 if ρ ′ 6= 0), (B.21)
Example of developing a rule: Equation (B.17) by using the rules above.

! ! ! !
ρϕ ρψ ρφ ρψ ρφ ρψ
ρ φ̃ ψ̃ = ρ · · = ρ̄ · · = ρ̄ · · = ρ̄ φ̃ ψ̃.
ρ̄ ρ̄ ρ̄ ρ̄ ρ̄ ρ̄
Here, Eqs. (B.6), (B.8) and (B.12) are used together with the definition (B.13).
ρφ ′′ ψ ′′
ρφ ′′ ψ ′′ = ρ̄ = ρ̄ ^
φ ′′ ψ ′′ , (B.22)
ρ̄
ρφ ρψ
φ̃ ψ̃ = · = φ̃ ψ̃, (B.23)
ρ̄ ρ̄
g = φ ψ̃.
φ ψ̃ = φ ψ̃ = φψ (B.24)
Notice that
g
dφ d φ̃
6= . (B.25)
ds ds
B.2 Continuity equation

General form, see Sec. A.1.1 page 197.
Reynolds averaged equation
Introduce u j = u j + u ′j . Assume that ρ does not fluctuate but that it can vary and hence, can have
gradients. Average the the equation,
∂ρ ∂
+ (ρu j ) = 0. (B.26)
∂t ∂x j
Here we could also say that we let ρ fluctuate, ρ = ρ̄ + ρ ′ , however, neglecting the correlations of
ρ ′ , that is ρ ′ u ′j . We then would have got ρ̄ in place of ρ in Eq. (B.26). The distinction is not very
interesting. We either assume the density as non-fluctuating, or we use the Favre averaged equations
(see below).
Constant density: On the distinction between variable density and compressible flow, see page 26.
∂u j
For ρ = constant we have Eq. (A.3): ∂x j = 0. From Eq. (A.3), or from Eq. (B.26), we get
∂u j
= 0. (B.27)
∂x j
When subtracting Eq. (B.27) from Eq. (A.3), a continuity equations for the velocity fluctuations is
obtained:
∂u ′j
= 0. (B.28)
∂x j
B.3 Turbulence equations for momentum 225
Favre averaged equation, mass-weighted Reynolds equation
Introduce the decomposition u j = ũ j + u ′′j and average the equation (the usual average):
∂ ρ̄ ∂
+ (ρ̄ ũ j ) = 0. (B.29)
∂t ∂x j
I the last term, ρ̄ ũ j = ρu j , cf. the definition, Eq. (B.3).
Reformulating the convective and transient terms
In Eqs. (A.4)–(A.5) the convective and transient terms of the equation for a general variable ϕ was
reformulated. Using Eq. (B.26) the transient and convective terms of an equation for a Reynolds
averaged variable can be ϕ reformulated in the same manner,
∂ ∂ ∂ϕ ∂ϕ
Cϕ = (ρϕ) + (ρϕ u j ) = ρ + ρu j . (B.30)
∂t ∂x j ∂t ∂x j
Also this reformulation is valid both at constant and variable density. Notice that u j ϕ 6= u j ϕ. This
is an error too frequently seen in the convective term.
By using Eq (B.29) the transient and convective terms of an equation for a Favre averaged variable
ϕ̃ can be written as
∂ ∂ ∂ ϕ̃ ∂ ϕ̃
Cϕ̃ = (ρ̄ ϕ̃) + (ρ̄ ϕ̃ ũ j ) = ρ̄ + ρ̄ ũ j . (B.31)
∂t ∂x j ∂t ∂x j
B.3 Turbulence equations for momentum

The fundamental equation: see Sec. 1.7, page 25.
Reynolds averaged equation
Introduce u i = u i +u ′i and assume that ρ does not fluctuate but can vary, so that it can have gradients.
Average the equation:
∂ ∂ ∂p ∂
(ρu i ) + (ρu i u j ) = − + τ i j − ρu ′i u ′j + ρ f i . (B.32)
u ′i u ′j are known as the Reynolds stresses or the Reynolds stress tensor. On terminology, see page 36.
The Boussinesq approximation for the buoyancy term of the momentum equation is discussed in
Sec. A.2.4. In the Reynolds averaged equation, the thermal buoyancy term becomes
(−ρ◦ α)(T − T◦ ) fi , (B.33)
where f i = −gδi3 when x 3 is directed upwards against the gravity. Furthermore, the density is
assumed constant, ρ = ρ◦ , in all other terms. Notice that a hydrostatic component of the gravity
term is included in the pressure term. This is not altered by the Reynolds average.
Favre equation, mass-weighted Reynolds equation
Introduce the decomposition u i = ũ i + u ′′i , see the definition in Eq. (B.3). Fluctuations of ρ are,
according to definition, included in ũ j . The equation is averaged in the usual manner (see Rules for
Favre averages page 223):
∂ ∂ ∂p ∂
(ρ̄ ũ i ) + (ρ̄ ũ i ũ j ) = − + τ i j − ρu ′′i u ′′j + ρ f i . (B.34)
Here can also be written ρu ′′i u ′′j = ρ̄ u]

′′ u ′′ , and ρ f = ρ̄ f˜ in the equation. The new quantities
i j i i
′′ ′′ ] ′′ ′′
ρu i u j or u i u j are the mass-weighted Reynolds stresses or Favre stresses. These are also simply
called Reynolds stresses when it is clear from the context that they are mass weighted. Using a
turbulence viscosity, the model turns out as in Eq. (3.26).
Notice that also on this form, p and τi j appear with the ordinary mean (overline).
B.4 Equation for species mass fraction

Fundamental equation: see Sec. A.1.3 page 199.
The source term is often written as Rk = ρωk . It is positive when species k is produced.
Reynolds equation
Introduce the decomposition u j = u j + u ′j and Yk = Y k + Yk′ . Assume that ρ and D do not fluctuate
but can vary so that they can have gradients. Average the equation, arrange the terms and obtain:
!
∂ ∂ ∂ ∂Y k ′ ′
(ρY k ) + (ρY k u j ) = ρD − ρu j Yk + Rk . (B.35)
Introduce the decomposition u j = ũ j + u ′′j and Yk = Ỹk + Yk′′ . For density, ρ = ρ̄ + ρ ′ . We assume
that the diffusion coefficient D does not fluctuate but can vary so that it can have gradients. Average
the equation in the usual manner and arrange the terms to
! !
∂ ∂ ∂ ∂ Ỹk ∂ ′′ ′′ ∂ ∂Yk′′
(ρ̄ Ỹk ) + (ρ̄ Ỹk ũ j ) = ρ̄D − ρu j Yk + Rk + D·ρ . (B.36)
∂t ∂x j ∂x j ∂x j ∂x j ∂x j ∂x j
B.5 Energy equation 227
The last term is in general non-zero but is often neglected.

The Favre fluxes (mass-weighted Reynolds fluxes, turbulence fluxes) can be expressed as ρu ′′j Yk′′ =
ρ̄ u]
′′ Y ′′ , and the source term as R = ρω = ρ̄ ω̃ .
j k k k k
B.5 Energy equation

To outline all variants will be a too lengthy exercise. The procedure is similar to that of the other
equations, see for instance the equations for species mass fraction above. A Reynolds-averaged
equation for enthalpy is shown at page 36, Eq. (2.10). The averaged source term hides correlations
that appear from terms of pressure, dissipation, diffusion, and others. These have to be discussed
and modeled or, alternatively, neglected.
The same can be said for the Favre-averaged energy equations. This far, I have not found any
author discussing the various correlations that appear. They are, as far as I have seen, neglected by
everyone – usually in silence. Restricting to the enthalpy equation, as Eq. (2.10), we can «on certain
conditions» write
!
∂ ∂ ∂ ∂ h̃ ′′
(ρ̄ h̃) + (ρ̄ h̃ ũ j ) = ρ̄α ′′
− ρu j h + Sh . (B.37)
Then, it is left to you to evaluate how accurate you will elaborate or investigate what «on certain
conditions» means in your special case.
B.6 Turbulence energy

Reynolds equation
The equation for Reynolds-averaged turbulence energy at constant density is developed in Sec. 3.2,
see Eq. (3.4).
The Boussinesq approximation for buoyancy in the momentum equation was discussed in Sec. A.2.4,
see also Eq. (B.33). The momentum equation got a term
(−ρ◦ α)(T − T◦ ) fi , (B.38)
where f i = −gδi3 when x 3 was directed upwards against gravity. Furthermore, we assumed the
density constant, ρ = ρ◦ , in all other terms. When doing the usual operations to achieve a Reynolds
averaged equation for turbulence energy, a source term,
ρ◦ G k = ρ◦ α(−u ′i T ′ ) f i (B.39)
appears from the term of Eq. (B.38). The pressure term of the turbulence energy equation is not
altered.
Favre equation, mass-weighted turbulence energy
The Favre or mass-weighted turbulence energy can be written as
′′ u ′′ = 1 1 ρu ′′ u ′′ .
k = 12 u] (B.40)
i i
2 ρ̄ i i
This quantity is often denoted k̃ in order to distinguish it from the Reynolds-averaged turbulence
energy, which in such contexts can be denoted with an overbar, k̄.
The development of the equation for the mass-weighted turbulence energy follows nearly the same
procedure as the Reynolds-averaged turbulence energy:
(1) The starting point is the instantaneous momentum equation.
(2) Multiply the equation by u ′′i :

u ′′i ∂t∂ (ρu i ) + ∂ ∂x j (ρu i u j ) = − ∂∂pxi + ∂∂x j (τi j ) + ρ f i . (B.41)
(3) Introduce the decomposition u i = ũ i + u ′′i , p = p + p′ .

(4) Average the equations in the usual way, and rearrange the terms.
After some mathematical operations, the equation appears:
∂ ∂ ∂ ũ i ∂
(ρ̄k) + ρ̄k ũ j = − ρu ′′i u ′′j + τi j u ′′i − 21 ρu ′′i u ′′i u ′′j − p′ u ′′j
∂p ∂u ′′ ∂u ′′
− u ′′i + p ′ i − τi j i . (B.42)
∂ xi ∂ xi ∂x j
The first right-hand-side term is the production term, the second is a diffusion term and the last one is
a dissipation term. In addition there are two terms that are zero when the density is constant. Notice
that the body-force acceleration does not appear in any term here, since ρu ′′i f i = 0. Body forces
(buoyancy etc.) acts, however, indirectly through other terms of the turbulence energy equation.
The terms can be arranged somewhat different. The model for τi j , Eq. (A.25), can be introduced.
However, contrary to in Eq. (3.4), there is no simple arrangement that can provide a viscous gradient
term («diffusion») in this equation.
The diffusion and dissipation is customarily modeled as in the Reynolds averaged turbulence energy
equation. Models for the two new terms are also proposed, but often they are simply neglected.
B.7 Reynolds stresses

Reynolds equation
The equation for the Reynolds stresses is shown and explained in Sec. 6.2-6.3, see Eq. (6.2). The
development is shown in the solutions to exercises.
B.8 Turbulence equations in cylinder coordinates 229
When using the Boussinesq approximation for buoyancy in the momentum equation (Sec. A.2.4, cf.
Eq. (B.33)), the Reynolds stress equation gets a source term like the equation for turbulence energy,
see Eq. (B.39):
ρ◦ G i j = ρ◦ α(−u ′j T ′ ) fi + ρ◦ α(−u ′i T ′ ) f j . (B.43)
Here, f i = −gδi3 when x 3 is directed upwards against gravity.
Development of the equation for mass-weighted Reynolds stresses (Favre stresses, Favre-averaged
Reynolds stresses):
(1) The starting point is the equation for instantaneous momentum, (ρu i ), which is multiplied by u ′′j :

u ′′j ∂t∂ (ρu i ) + ∂∂xk (ρu i u k ) = − ∂∂pxi + ∂∂xk (τik ) + ρ f i . (B.44)
(2) Multiply the equation for instantaneous momentum (ρu j ) by u ′′i :

u ′′i ∂t∂ (ρu j ) + ∂∂xk (ρu j u k ) = − ∂∂p
xj + ∂
∂ xk (τ j k ) + ρ f j . (B.45)
(3) Introduce the decomposition u i = ũ i + u ′′i , p = p + p ′ .

(4) Add the two equations.
(5) Average the equation in the usual manner. Rearrange the terms:
∂ ] ∂ ] ∂ ũ j ∂ ũ i
ρ̄ u ′′i u ′′j + ρ̄ u ′′i u ′′j ũ k = − ρu ′′i u ′′k − ρu ′′j u ′′k
∂t ∂ xk ∂ xk ∂ xk
∂
+ τik u ′′j + τ j k u ′′i − ρu ′′i u ′′j u ′′k − p′ u ′′i δ j k − p′ u ′′j δik
∂ xk
∂p ∂p ∂u ′′ ∂u ′′j
− u ′′i − u ′′j + p′ i + p′
∂x j ∂ xi ∂x j ∂ xi
∂u ′′j ∂u ′′i
− τik − τ jk .
∂ xk ∂ xk
(B.46)
Here, as in the turbulence energy equation, a gradient term for viscous diffusion can not readily be
arranged. The body-force acceleration does not appear in any term here, since ρu ′′i f j = 0, cf. the
equation for turbulence energy. The pressure terms can be rearranged from what is shown here.
B.8 Turbulence equations in cylinder coordinates

The equations of motion in cylinder coordinates can be found in Bird, Stewart and Lightfoot (1960:83ff),
also reproduced by Kuo (1986:175ff). The same mathematical operations as shown above are used.
Here, the equations are given on the mass-weighted (Favre) form.

Equations with Reynolds averages can be developed from the Favre equations: Let ũ i = u i , u ′′i = u ′i
and u ′′i = 0, cf. the discussion at page 26 and definitions in Sec. B.1. In the equations, ρ̄ (or simply
ρ) can be set outside the average signs. Furthermore, a gravity term as in Eq. (B.43) have to be
included.
The coordinate system is (r ,θ ,z), where r is the radial coordinate, θ the tangential coordinate and z is
along the cylinder axis. The velocity components are (vr ,v θ ,v z ). The equations can straightforwardly
be reformulated to a (x,r ,θ ) system by replacing z and v z by x and v x (or u).
B.8.1 Momentum
∂ 1 ∂ 1 ∂ ∂ ∂p 1 ∂
(ρ̄ ṽr ) + (r ρ̄ ṽr ṽr ) + (ρ̄ ṽ θ ṽr ) + (ρ̄ ṽ z ṽr ) = − + r τ rr − r ρ̄ vg
r
′′ 2
∂t r ∂r r ∂θ ∂z ∂r r ∂r
1 ∂ ∂ 1
g

′′ 2 − ρ̄ ṽ 2 + ρ f
+ τ rθ − ρ̄ vg ′′ v ′′ +
r θ τ r z − ρ̄ vg′′ v ′′ −
r z τ θθ − ρ̄ v θ θ r (B.47)
r ∂θ ∂z r
∂ 1 ∂ 1 ∂ ∂ 1∂p 1 ∂
(ρ̄ ṽ θ ) + (r ρ̄ ṽr ṽ θ ) + (ρ̄ ṽ θ ṽ θ ) + (ρ̄ ṽ z ṽ θ ) = − + r τ rθ − r ρ̄ vg
′′ v ′′
r θ
∂t r ∂r r ∂θ ∂z r ∂θ r ∂r
1 ∂ 1
τ θθ − ρ̄ vg ′′ 2 + ∂ τ θ z − ρ̄ vg ′′ ′′ τ rθ − ρ̄ vg
′′ ′′
+ θ θ vz + r v θ − ρ̄ ṽ r ṽ θ + ρ f θ (B.48)
r ∂θ ∂z r
∂ 1 ∂ 1 ∂ ∂ ∂p 1 ∂
(ρ̄ ṽ z ) + (r ρ̄ ṽr ṽ z ) + (ρ̄ ṽ θ ṽ z ) + (ρ̄ ṽ z ṽ z ) = − + r τ r z − r ρ̄ vg
′′ ′′
r vz
∂t r ∂r r ∂θ ∂z ∂z r ∂r
1 ∂ ∂
g

+ τ θ z − ρ̄ vg θ
′′ v ′′ +
z τ zz − ρ̄ v ′′ 2 + ρ f
z z (B.49)
r ∂θ ∂z
B.8.2 Equations for Reynolds stresses
∂ g 1 ∂ 1 ∂ ∂
ρ̄ vr′′ 2 + r ρ̄ ṽr vg ′′ 2 +
r ρ̄ ṽ θ vg′′ 2 +
r ρ̄ ṽ z vg′′ 2 =
r
∂t r ∂r r ∂θ ∂z
∂ ṽ r 1 ∂ ṽ r ∂ ṽr ṽ θ ′′ ′′
− 2ρ̄ vg ′′ 2
r − 2ρ̄ vg′′ ′′
r vθ − 2ρ̄ vg′′ ′′
r vz + 4ρ̄ vg v
∂r r ∂θ ∂z r r θ
1 ∂ ′′ 1 ∂
+ r 2vr τrr − r ρvr′′ 3 − r 2vr′′ p′ + 2vr′′ τrθ − ρvr′′ 2 v θ′′
r ∂r r ∂θ
∂ ′′ 1
+ 2vr τr z − ρvr′′ 2 v z′′ − 2 vr′′ τθθ − ρvr′′ v θ′′ 2
∂z r
∂ p̄ 1 ∂ ∂v ′′ 1 ∂v ′′ ∂v ′′
− 2vr′′ + 2 p′ r vr′′ − 2τrr r − 2 τrθ r − 2τr z r (B.50)
∂r r ∂r ∂r r ∂θ ∂z
B.8 Turbulence equations in cylinder coordinates 231
∂ g 1 ∂
′′ 2 + 1 ∂

′′ 2 + ∂

ρ̄ v θ′′ 2 + r ρ̄ ṽr vgθ ρ̄ ṽ θ vg
θ ρ̄ ṽ z vg
′′ 2 =
θ
∂ ṽ 1 ∂ ṽ ′′ ′′ ṽ θ − 2ρ̄ vg
∂ ṽ ′′ 2 ṽ r
− 2ρ̄ vg g
θ ′′ 2 θ ′′ ′′ θ − 2ρ̄ v
− 2ρ̄ vg
′′ ′′
r vθ θ − 2ρ̄ vg θ vz r vθ θ
∂r r ∂θ ∂z r r
1 ∂ ′′ ′′ 2
1 ∂
′′ ′′ 3 ′′

+ r 2v θ τrθ − r ρvr v θ +
′′ 2v θ τθθ − ρv θ − 2v θ p ′
r ∂r r ∂θ
∂ ′′ 1
+ 2v θ τθ z − ρv θ′′ 2 v z′′ + 2v θ′′ τrθ − 2ρvr′′ v θ′′ 2
∂z r
′′ ∂v ′′ ∂v ′′ ∂v ′′
1 ∂ p̄ 1 ∂v θ 1
− 2v θ′′ + 2 p′ − 2τrθ θ − 2 τθθ θ − 2τθ z θ (B.51)
r ∂θ r ∂θ ∂r r ∂θ ∂z
∂ g 1 ∂ 1 ∂ ∂
ρ̄ v z′′ 2 + r ρ̄ ṽr vg
′′ 2 +
z ρ̄ ṽ θ
g
v ′′ 2 +
z ρ̄ ṽ z
g
v ′′ 2 =
z
− 2ρ̄ vg ′′ ′′
∂ ṽ z
− 2ρ̄ vg ′′ ′′ 1 ∂ ṽ z − 2ρ̄ v g ′′ 2
∂ ṽ z
r vz θ vz z
∂r r ∂θ ∂z
1 ∂ ′′ 1 ∂ ∂
+ r 2v z τr z − r ρvr′′ v z′′ 2 + 2v z′′ τθ z − ρv θ′′ v z′′ 2 + 2v z′′ τzz − ρv z′′ 3 − 2v z′′ p′
r ∂r r ∂θ ∂z
∂ p̄ ∂v ′′ ∂v ′′ 1 ∂v ′′ ∂v ′′
− 2v z′′ + 2 p′ z − 2τr z z − 2 τθ z z − 2τzz z (B.52)
∂z ∂z ∂r r ∂θ ∂z
∂ g 1 ∂ 1 ∂ ∂
ρ vr′′ v θ′′ + r ρ̄ ṽr vg′′ v ′′ +
r θ ρ̄ ṽ vg′′ v ′′ +
θ r θ ρ̄ ṽ vg′′ v ′′ =
z r θ
− ρ̄ vg ′′ ′′
∂ ṽr
− ρ̄ vg′′ 2
∂ ṽ θ
− ρ̄ vg′′ 2 1 ∂ ṽ r − ρ̄ vg ′′ ′′
1 ∂ ṽ θ
− ρ̄ vg ′′ ′′ ∂ ṽ r − ρ̄ vg
′′ ′′
∂ ṽ θ
r vθ r θ r vθ θ vz r vz
∂r ∂r r ∂θ r ∂θ ∂z ∂z
ṽ ṽ
+ ρ̄ 2vg g θ ′′ ′′ r
′′ 2
θ − vr
′′ 2 − ρ̄ vg r vθ
r r
1 ∂ ′′ 1 ∂
+ r v θ τrr + r vr′′ τrθ − r ρvr′′ 2 v θ′′ − r v θ′′ p′ + v θ′′ τrθ + vr′′ τθθ − ρvr′′ v θ′′ 2 − vr′′ p′
r ∂r r ∂θ
∂ ′′ 1 1
+ v θ τr z + vr′′ τθ z − ρvr′′ v θ′′ v z′′ + vr′′ τrθ − ρvr′′ 2 v θ′′ − v θ′′ τθθ − ρv θ′′ 3
∂z r r
∂ p̄ 1 ∂ p̄ 1 ′ ∂ 1 ∂v ′′
− v θ′′ − vr′′ + p r v θ′′ + p′ r
∂r r ∂θ r ∂r r ∂θ
′′ ′′ ∂v ′′
∂v ∂v ′′ 1 ∂v 1 ∂v ′′ ∂v ′′
− τrr θ − τrθ r − τrθ θ − τθθ r − τr z θ − τθ z r (B.53)
∂r ∂r r ∂θ r ∂θ ∂z ∂z
∂ 1 ∂ 1 ∂ ∂
ρ̄ vg
′′ ′′
r vz + r ρ̄ ṽr vg ′′ ′′
r vz + ρ̄ ṽ θ vg′′ ′′
r vz + ρ̄ ṽ z vg ′′ ′′
r vz =
∂ ṽ r ∂ ṽ z 1 ∂ ṽ r 1 ∂ ṽ z ∂ ṽr ∂ ṽ z ′′ ′′ ṽ θ
− ρ̄ vg
′′ ′′
r vz − ρ̄ vgr
′′ 2 − ρ̄ vg ′′ ′′
θ vz − ρ̄ vg ′′ ′′
r vθ − ρ̄ vg z
′′ 2 − ρ̄ vg′′ ′′
r vz + 2ρ̄ vg θ vz
∂r ∂r r ∂θ r ∂θ ∂z ∂z r
1 ∂ ′′ ′′ ′′ 2 ′′ ′′ ′
1 ∂
′′ ′′ ′′ ′′ ′′

+ r v z τrr + r vr τr z − r ρvr v z − r v z p + v τrθ + vr τθ z − ρvr v θ v z
r ∂r r ∂θ z
∂ ′′ 1
+ v z τr z + vr′′ τzz − ρvr′′ v z′′ 2 − vr′′ p′ − v z′′ τθθ − ρv θ′′ 2 v z′′
∂z r
∂ p̄ ∂ p̄ 1 ′ ∂ ∂v ′′
− v z′′ − vr′′ + p r v z′′ + p′ r
∂r ∂z r ∂r ∂z
∂v ′′ ∂v ′′ 1 ∂v ′′ 1 ∂v ′′ ∂v ′′ ∂v ′′
− τrr z − τr z r − τrθ z − τθ z r − τr z z − τzz r (B.54)
∂r ∂r r ∂θ r ∂θ ∂z ∂z
∂ g 1 ∂
′′ v ′′ + 1 ∂

′′ v ′′ + ∂ ρ̄ ṽ vg

ρ̄ v θ′′ v z′′ + r ρ̄ ṽr vgθ z ρ̄ ṽ θ vg
θ z z
′′ v ′′ =
θ z
− ρ̄ vg ′′ ′′
∂ ṽ θ
− ρ̄ vg ′′ ′′
∂ ṽ z
− ρ̄ vg ′′ ′′ 1 ∂ ṽ θ − ρ̄ v g ′′ 2 1 ∂ ṽ z − ρ̄ v g ′′ 2
∂ ṽ θ
− ρ̄ vg′′ ′′ ∂ ṽ z
r vz r vθ θ vz θ z θ vz
∂r ∂r r ∂θ r ∂θ ∂z ∂z
ṽ θ ṽ r
− ρ̄ vg ′′ ′′
r vz − ρ̄ vg ′′ ′′
θ vz
r r
1 ∂ ′′ ′′ ′′
1
+ r v z τrθ + r v θ τr z − r ρvr v θ v z + ′′ ′′ v z′′ τrθ − ρvr′′ v θ′′ v z′′
r ∂r r
1 ∂ ′′ 2
∂ ′′
+ ′′ ′′ ′′
v z τθθ + v θ τθ z − ρv θ v z − v z p + ′′ ′ v z τθ z + v θ′′ τzz − ρv θ′′ v z′′ 2 − v θ′′ p′
r ∂θ ∂z
1 ∂ p̄ ∂ p̄ 1 ′ ∂v z′′ ∂v ′′
− v z′′ − v θ′′ + p + p′ θ
r ∂θ ∂z r ∂θ ∂z
′′
∂v ′′ ∂v 1 ∂v ′′ 1 ∂v ′′ ∂v ′′ ∂v ′′
− τrθ z − τr z θ − τθθ z − τθ z θ − τθ z z − τzz θ (B.55)
∂r ∂r r ∂θ r ∂θ ∂z ∂z
B.8.3 Simplifications, axial symmetry

Axial symmetry with tangential velocity
No mean values varies tangentially to the axis; that is, normal to a plane along the axis. Hence, all
terms ∂( )/∂θ = 0. The equation system is expressed in two dimensions. There can still be a
tangential velocity component ṽ θ 6= 0, and this can vary with r and z, but not with θ . This equation
system can be used for swirl flows.
B.9 Series development and average of the reaction term (Arrhenius) 233
Axial symmetry without tangential velocity
As above, all terms ∂( )/∂θ = 0. In addition, the tangential velocity component is zero, ṽ θ = 0.
Then, the ṽ θ equation is not needed and not the equations for vg
′′ ′′ g′′ ′′
r v θ and v θ v z , since these quantities
only appear in the ṽ θ equation.
B.9 Series development and average of the reaction term (Arrhenius)

This section refers mainly to Sec. 10.5
An expression for the average reaction rate can be found by first, to develop the exponential function
of temperature in a series and then, to introduce the Reynolds decomposition. This procedure is
known from literature (Borghi, 1974,1988), but seems to be little and incompletely discussed. The
reason might be that the conclusion is obvious, provided that the one takes the time to do the devel-
opment and consider the outcome. On the other hand, we now and then see proposals of using this
approximation as a combustion model.
For a simple, global reaction fu + ox → prod the reaction rate for the fuel can be expressed
as Rfu = −kf ρ 2 Yfu Yox with the reaction coefficient according to the Arrhenius expression (cf.
Eqs. (1.13) and (10.7)), kf = kf (T ) = BT β exp(−Ta /T ) .
These expressions give
Ta
Rfu = −Bρ 2 T 2 · T β−2 Yfu Yox exp(− ) (B.56)
T
We assume ideal gases, ρT = p/R, where R is the gas constant for the mixture (and here, we
skip the problem of calculating this quantity). Furthermore, we neglect correlations of the pressure,
which means that we keep the pressure outside the averaging.
The Reynolds decomposition is introduced for mass fractions, giving
!
′ + Y′ Y ′ ′
Y fu Yox fu ox + Yfu Yox
Yfu Yox = Y fu Y ox 1 + (B.57)
Y fu Y ox
In order to simplify writing, we define the functions

Ta
f (T ) = exp(− ) , g(T ) = T β−2 and h(T ) = g(T ) · f (T ) (B.58)
T
The function h(T ) can developed as a Taylor series about T = T . The deviation from T is the
fluctuation T ′ = T − T , and
1 1
h(T ) = h(T ) + h ′ (T ) T ′ + h ′′ (T ) T ′ 2 + h ′′′ (T ) T ′ 3 + . . . (B.59)
2 6
When collecting this, averaging the equation and setting mean quantities outside the average, the
relation appears as
p2
R fu = −B Y fu Y ox h(T ) FC
R2
p2 β−2 Ta
= −B 2 Y fu Y ox T exp(− ) FC
R T
= Rfu (Y fu , Y ox , T ) FC (B.60)
where
! !
′ + Y′ Y ′ ′ h ′ (T ) h ′′ (T ) h ′′′ (T )
Y fu Yox fu ox + Yfu Yox
FC = 1 + 1+ T′ + T′ 2 + T′ 3 + ...
Y fu Y ox h(T ) 2h(T ) 6h(T )
(B.61)
or
!
Yfu′ Yox
′ Yfu′ T ′ ′ T′
Yox Yfu′ Yox
′ T′ T h ′ (T )
FC =1 + + + +
Y fu Y ox Y fu T Y ox T Y fu Y ox T h(T )
! 2
T′ 2 Yfu′ T ′ 2 ′ T′ 2
Yox ′ T′ 2
Yfu′ Yox T h ′′ (T )
+ 2
+ 2
+ 2
+ 2
T Y fu T Y ox T Y fu Y ox T 2 h(T )
! 3
T′ 3 Yfu′ T ′ 3 ′ T′ 3
Yox ′ T′ 3
Yfu′ Yox T h ′′′ (T )
+ 3
+ 3
+ 3
+ 3
T Y fu T Y ox T Y fu Y ox T 6 h(T )
+ ...... (B.62)
Before considering the content of h ′ (T ), h ′′ (T ), h ′′′ (T ) and so on, we get an idea of considerable
challenges in modeling a large amount of correlations of increasing order; Yfu′ Yox ′ , Y ′ Y ′ T ′,
fu ox
′ 2 3
Yfu Yox T , T , and so on.
′ ′ ′
h(T ) = g(T ) · f (T ) (B.63)

Ta g(T ) · f (T )
h ′ (T ) = g ′ (T ) · f (T ) + g(T ) · f ′ (T ) = (β − 2) + (B.64)
T T
′′ ′′ ′ ′ ′′
h (T ) = g (T ) · f (T ) + 2g (T ) · f (T ) + g(T ) · f (T )

Ta Ta T 2 g(T ) · f (T )
= (β − 2)(β − 3) + 2(β − 2) − 2 + 2 a2 (B.65)
T T T T2
h ′′′ (T ) = g ′′′ (T ) · f (T ) + 3g ′′ (T ) · f ′ (T ) + 3g ′ (T ) · f ′′ (T ) + g(T ) · f ′′′ (T )

Ta
= (β − 2)(β − 3)(β − 4) + 3(β − 2)(β − 3)
T
2
Ta T Ta T2 T 3 g(T ) · f (T )
−6(β − 2) + 3(β − 2) a2 + 6 − 6 a2 + a3 (B.66)
T T T T T T3
B.9 Series development and average of the reaction term (Arrhenius) 235
n
T · h n (T )
Here, it is seen that in the series for FC , Eq. (B.62), the (n + 2) term containing , will
n! · h(T )
1 Ta T′ n
increase as compared to the (n + 1) term.
n T T ′ (n−1) T
When Ta ≪ T and T ′ ≪ T , the series FC might converge fast. However, if Ta > T (alternatively
Ta ≫ T ), a large number of terms (large n) is required for convergence. This gives a virtually
impossible modeling problem of correlations of the (n + 2)-th order.
The method can be adequate for reactions that are approximately isothermal, such that T ′ ≪ T ,
where, in addition, Ta ≪ T .
Appendix C
Dimensionless parameters
C.1 Origin and names
In thermal and fluid science, quite a few dimensionless parameters are named after a person, usually
one that was the first to use this specific parameter. Examples are Mach, Stanton, Prandtl, Knudsen,
Reynolds, see in books on fluid mechanics or heat and mass transfer.
G.G. Stokes found in i 1850 that what we today call Reynolds number was a criterion for dynamic
similarity, and in 1873, H. von Helmholtz used the groups now known as Froude and Mach numbers.
The first one to assign a personal name to such groups is supposed to be M.G. Weber. In 1919 he
named the Froude, Reynolds and Cauchy numbers. Since then other groups have been associated to
both Stokes, Helmholtz and Weber. The naming has to a large extent been unsystematic or incidental.
Consequently, some groups have had several names, while the same name have been assigned to
several groups (Catchpole and Fulford, 1987.)
C.2 Reynolds number

Osborne Reynolds, 1842–1912, see page 244.
A Reynolds number is the ratio of an inertial force to a viscous force. The original Reynolds number
is the ratio of convective transport to diffusive transport of a pipe flow (Reynolds, 1883):
UD
Re = . (C.1)
µ/ρ
Reynolds found that this is a general parameter for the transition from laminar to turbulent flow in
a pipe. U is the average velocity in the pipe (volume flow divided by area) and D is the diameter.
When the momentum equation is made dimensionless, such a parameter appears, see Eq. (8.2).
238 Appendix C Dimensionless parameters
Later, similar parameters are identified and are also called Reynolds numbers. On turbulence Reynolds
numbers, see Sec. 9.3.
C.3 Richardson number, Froude number

Lewis Fry Richardson, 1881–1953, see page 244.
William Froude, 1810–1879, see page 242.
In general, a Richardson number Ri is the ratio of a gravity (buoyant force) to an inertial force. This
is often related to boundary layers or stratified flow. Richardson investigated flows and stability in
the atmosphere (cf. Richardson, 1922), and the parameter was formulated in that instance.
Schlichting (1979:512) defines the Richardson number for boundary layers and provides more ref-
erences. He expresses it as a function of the density gradient:

g dρ du 1 −2
Ri = − , (C.2)
ρ d x3 d x3
where x 3 is directed upwards against the gravitational acceleration g. This expression is more gen-
eral than expressing the stability criterion as a function of the temperature gradient, as it was done
originally.
Tennekes and Lumley (1972:98f) define and discuss a flux Richardson number for turbulence. This
is the ratio of the production of turbulence energy by buoyancy to the production by shear forces.
They find that a simpler gradient Richardson number is more conveniently used, which corresponds
to the Richardson number defined by Schlichting, however with a temperature gradient.
All Richardson numbers are a measure for stability of stratified flows with buoyancy. They can be
used as criteria to indicate whether the flow becomes turbulent due to buoyancy.
The reciprocal quantity is known as the Froude number, Fr. Hence, this is the ratio of an inertial
force to gravity (buoyancy force),
U2
Fr = . (C.3)
gL
The Froude number is more often used in a general sense, while the Richardson number is mainly
used in relation to boundary layers and stratified flows. The original dimensionless quantity is
U/(g L)1/2 , where U is the wave speed (at the sea) relative to a ship, and L was the length of
the waterline of the ship.
C.4 Prandtl number, Schmidt number, Lewis number

Ludwig Prandtl, 1875–1953, see page 244.
Ernst Heinrich Wilhelm Schmidt, 1892–1975, see page 244.
Warren Kendall Lewis, 1882–1978, see page 243.
C.5 Damköhler number 239
The Prandtl number is the ratio of momentum diffusivity (viscosity) to the thermal diffusivity,
µ ν
Pr = = . (C.4)
λ/C p α
The Schmidt number is the ratio of the momentum diffusivity (viscosity) to the mass diffusivity,
µ ν
Sc = = . (C.5)
ρD D
The Lewis number is the ratio of the thermal diffusivity to the mass diffusivity,
λ/C p α Sc
Le = = = . (C.6)
ρD D Pr
The Schmidt number can be introduced into Fick’s law, see Eqs. (A.18)–(A.19). We have also
introduced the Prandtl, Schmidt and Lewis numbers in the model for heat flux expressed in terms of
an enthalpy gradient, see Eq. (A.89) and that which follows.
In the gradient model for a turbulence flux, Eqs. (5.1)–(5.3), a turbulence diffusivity is used and we
can define the turbulence Prandtl number, (νt /αt ), and the turbulence Schmidt number, (νt /Dt ).
In other modeled equations for turbulent flow, quantities (numerical constants) similar to the Prandtl
and Schmidt numbers are used. For instance, in a k-ε model (page 53) we find the quantities σk and
σε . These do not have their own names but are known as turbulence Prandtl/Schmidt numbers.
C.5 Damköhler number

Gerhard Damköhler, 1908–1944, see page 241.
A total of five dimensionless groups are known as Damköhler numbers, Da, or Damköhler groups.
From their originator they are enumerated from 1 to 5, and the first four are related to chemical
reactions. Basically, reaction and transport are compared. This can be expressed as the ratio of a
chemical reaction rate to the rate of mass transport by flow or molecular diffusion. It can also be
expressed as the released reaction heat divided by the heat transfer by flow or diffusion.
Reactions and transport can be expressed in terms of timescales, such that the Damköhler numbers
becomes the ratio of a transport timescale to a reaction timescale (chemical timescale).
The turbulence Damköhler number, Dat , is the ratio of the chemical reaction rate to the transport
by turbulent motions. This is customarily expressed as a turbulence timescale divided by a chemical
timescale, see Sec. 10.2. Notice that in some cases, «large» timescales are used – for instance ℓ′ /u ′ or
k/ε, and in some other cases «small» timescales are used – for instance the Kolmogorov timescale.
Hopefully, it is seen from the context which is used. The ratio of the chemical timescale and the
Kolmogorov timescale, τc /τ , is also known as the Karlovitz number, Ka.
C.6 Courant number or CFL number

A flow has velocity U in the x direction and is computed numerically with a grid of mesh δx. Then,
a particle or a perturbation will use the time δx/U to pass the length δx. In order to capture this
motion, the numerical solution has to have a timestep δt less than δx/U .
Courant, Friedichs and Lewy (1928) showed that a solution can converge properly only when
δt < δx/U . This is known as the Courant-Friedichs-Lewy criterion, or, simply the CFL crite-
rion. The criterion is used in Eq. (10.32), when saying that the shortest time has to be τU = η/U or
less; cf. also Eq. (9.7).
The dimensionless group
δx
(C.7)
U δt
is known as the Courant number or the CFL number. (A variety of symbols are used.)
C.7 Other dimensionless parameters

The International Standard ISO 31-12 gives symbol, name and definition of 25 dimensionless pa-
rameters.
Catchpole and Fulford (1966,1987) give a schematic overview of 285 dimensionless groups with
symbol, definition, explanation and references.
C.8 More prominent people of the subject

The following is some words about persons that have influenced the science of turbulence, flow, heat
and combustion. It can be said about each individual that he was a clever guy, so this is not repeated.
They also have in common that they are all dead, and that all are males.
Most of them have got their name linked to models, theories, laws etc. Some have, as seen above, got
their name associated with dimensionless quantities. A few have become adjectives (e.g. Cartesian,
Newtonian) – this is supposed to be grander. In English (and Norwegian), at least one has become
a common noun: diesel, about fuel oil and engines. None of the people presented here has – to my
knowledge – become a verb (cf. to pasteurize, to boycott, to macadamize), which linguistically is
supposed to be the grandest.
Quite a few are presented in Dictionary of Scientific Bibliography (1970ff). Other sources are gen-
eral encyclopedia and textbooks of the subject. A couple of history books have also been useful;
Rouse and Ince (1957) on hydraulics and Truesdell (1980) on thermodynamics.
The selection may seem a bit incidental. Moreover, I have been somewhat biased in favour of Nor-
wegians, including some that have been working at the outskirts of the topic of this book. For natural
reasons, Norwegians have in particular been working with motions in atmosphere and ocean.
C.8 More prominent people of the subject 241
Abu Jafar Muhamad Ibn Musa al-Khwarizmî, born before 800, dead after 847, Arabic mathe-
matician and geographer. Particularly known for books on arithmetics and algebra, which came in
several issues in European languages and paved the road for Arabic/Indian numerals in Europe. The
word algorithm originates from his name.
Svante August Arrhenius, 1859–1927, Swedish physicist and chemist. Fundamental work on
electro-chemistry and on, among other, how chemical reactions vary with the temperature (cf.
Sec. 1.8). Received the Nobel price in chemistry 1903.
Daniel Bernoulli, 1700–1782, Swiss physicist etc. Gave a mathematical foundation for the fluid
mechanics in his work Hydrodynamica from 1738. Has given name to the Bernoulli equation, see
Sec. A.2.2.
Carl Anton Bjerknes, 1825–1903, Norwegian mathematician and physicist, father to Vilhelm
Bjerknes. Known for work on hydrodynamics, on the interactions between a liquid and several
spherical bodies moving in it. Rejected theoretically and experimentally the idea that there are hy-
drodynamic body forces similar to magnetic and electric forces.
Vilhelm Friman Koren Bjerknes, 1862–1951, Norwegian physicist and meteorologist, son of Carl
A. Bjerknes. One of the founders of modern meteorology.
Joseph Valentin Boussinesq, 1842–1929, French physicist and mathematician. Worked within
most areas of classical mathematics and physics. Comprehensive work on analytical and mathemat-
ical treatment of liquid flows; curved streamlines, velocity distribution, turbulence, etc. See also
Sec A.2.4 on different Boussinesq approximations.
Robert Wilhelm Eberhard Bunsen, 1811–1899, German chemist. Known for, among other, fun-
damental work in spectral analysis. Made use of, and has given name to, the Bunsen burner. This
burner gives a simple, laminar premixed flame. Michael Faraday is supposed to be the originator
of the concept, and it is somewhat unclear whether Bunsen contributed to the development of the
burner.
Chou Pei-Yuan (Zhou Peiyuan according to newer transcription) 1902–1993, Chinese physicist.
Worked with theoretical physics and relativity theory. At the the outbreak of the war, he turned
towards fluid mechanics. Among other, he developed the first modeled equations for the Reynolds
stresses and for the dissipation (see Sec. 3.1). Went on with mechanics after WWII. In spite of
political activity (among other, he was the only Chinese delegate at the first Pugwash conference in
1957 and member for the rest of his lifetime), he came safely through several upheavals, including
the Cultural Revolution. Lumley (1995) gives a biography.
Gerhard Damköhler, 1908–1944, German, did fundamental work on chemical reactors and com-
bustion (flames). Apart from the Damköhler number (Sec. C.5) and a couple of publications, it
seems that he is almost forgotten. It is, for instance, not easy to find reference books where he is
mentioned. A biography is given by Wicke (1984).
René du Perron Descartes (Latin: Renatus Cartesius), 1596–1650, French philosopher and mathe-
matician. Work on many branches of science. Has given name to Cartesian coordinates (page 266).
Rudolf Diesel, 1858–1913, German mechanical engineer. Designed (1893) and constructed (1897)
an engine with considerably higher thermal efficiency than any other heat engines at the time. The
Diesel engine is virtually the only engine used for heavy vehicles and construction machinery,
medium size vessels, and otherwise used in smaller power plants and similar.
Vagn Walfrid Ekman, 1874–1954, Swedish physicist, known for work on oceanography and fluid
mechanics. Student of Vilhelm Bjerknes and co-worker to Fritjof Nansen. Developed a theoretical
basis for Nansen’s observation that a wind current changed direction with the elevation over the
ocean (the Ekman spiral or Ekman boundary layer).
Ægidius Elling, 1861–1949, Norwegian engineer and inventor, «father of the gas turbine». Got the
first gas turbine patented in 1884, construction completed in 1903. This was the first gas turbine
with surplus power (positive thermal efficiency). The first successful industrial gas turbine fist came
in 1936 and for airplanes in 1939.
Leonhard Euler, 1707–1782, Swiss mathematician and physicist. Work on several branches of
mathematics. On Euler equations, see Sec. A.2.2.
Adolf Eugen Fick, 1829–1901, German physiologist and physician. Worked on problems on the
interface between physiology, medicine and physics – especially heat and mass transfer. Known for,
among other, the gradient model for (mass) diffusion, Fick’s law (see Sec. A.1.3), proposed in 1855.
Jean Baptiste Joseph Fourier, 1768–1830, French
mathematician and physicist. Active in the revolu-
tion, was imprisoned and close to execution in the ter-
ror regime in the mid-1790s. Fundamental work on
mathematical problems in heat transfer, 1822. The
mathematical methods are used in several branches of
physics. Fourier’s law of heat transfer, Sec. A.3.4.
William Froude, 1810–1879, English naval engineer,
best known for developing scaling laws for ship model
experiments. On Froude number, see C.3.
Galileo Galilei, 1564–1642, Italian.
Josiah Willard Gibbs, 1839–1903, American mathe-
matician, physicist and chemist. Founded thermody-
namics for chemical reactions. The Gibbs function or Figure C.1: Hagen’s device: At a
Gibbs free energy is named after him. certain limit, the jet ejected from
the pipe began oscillating and be-
Gotthilf Heinrich Ludwig Hagen, 1797–1884, Ger- came intermittent; we would say
man, civil engineer and physicist. Wrote a handbook turbulent (cf. page 16). From Ha-
on hydraulic engineering and a work on probability es- gen (1839).
timates. Studied the transition from laminar to turbu-
lent flow (see page 16). With the apparatus in Fig. C.1,
Hagen (1839) investigated the relation between the
pressure head (the liquid column E in the figure) and
the flow in a pipe (A in the figure). He found what is later known as the Poiseuille equation (also the
Hagen-Poiseuille equation) in 1839, one year before Poiseuille.
Hoyt Clarke Hottel, 1903–1998, American chemist. Work on flames, radiation, combustion mech-
anisms, turbulence, etc. One of the founders of the The Combustion Institute, which among other
arrange the biannual international symposium of combustion. The main plenary lecture there is
called the Hoyt Hottel lecture in his honour. See also page 14.
Theodore von Kármán, 1881–1963, Hungarian physicist, worked in Germany to 1930, then in the
USA. Best known for fundamental work on aerodynamics, and as a consultant for Nato on this area.
The Nato laboratory for aero and fluid dynamics outside Brussels is called the von Karman Institute
after him. On von Kármán vortex street, see Sec. 256. (The car name Karmann has a different
origin.)
William Thomson, lord Kelvin, 1824–1907, British (Ireland, Scotland) physicist. Best known
as one of the founders of thermodynamics. (And the subject has got its name from one of his
publications. «On the dynamical theory of heat . . . », 1851.) He was the consultant for the company
that laid the first telegraph cable across the Atlantic ocean, and was knighted (got the «Sir» title) for
this in 1866. Ennobled in 1892 for his work in thermodynamics. Studied several areas of physics,
among other eddying motion and vortex-free slow, and instabilities in viscous flow. I supposed
to have introduced (1887) the word turbulence about the flow regime above the critical Reynolds
number.
Martin Hans Christian Knudsen, 1871–1949, Danish physicist and hydrographer. Known, among
other, for work on gases at low pressure. The Knudsen number is named after him.
Andrej Nikolaievitsj Kolmogorov, 1903–1987, Russian mathematician, especially known for fun-
damental work on probability theory, also topology and other areas of mathematics. The study of
steady random processes he related to the analysis of isotropic turbulence (see Sec. 8.3).
Leopold Kronecker, 1823–1891, German mathematician, work on number theory and algebra,
solution of equations, etc. Has given name to theKronecker delta, page 220.
Henry-Louis Le Châtelier, 1850–1936, French chemist. Best known for the Le Châtelier principle,
originally proposed in 1884, on how a thermodynamic system in equilibrium reacts on perturbations
(see a book on physical chemistry); also known for Le Chatelier’s law for the lower flammability
limit (see Chomiak, 1990:59) and the theory for propagation of laminar flames (see e.g. Glass-
man, 1996:127f). The oxygen-acetylene burner used for welding and cutting was developed (1895)
according to Le Chatelier’s proposal.
Leonardo da Vinci, 1452–1519, Italian artist, engineer and scientist. Formulated a simple principle
for continuity in a flow of a liquid. Studied motions in water and air, constructed channels, etc., see
also page 16.
Gilbert Newton Lewis, 1875–1945, American chemist. Best known for work on acids and bases
(the Lewis theory) and for a textbook on thermodynamics for chemical substances (with Randall).
Warren Kendall Lewis, 1882–1978, American, professor in chemical engineering at MIT. Known
among other for work (1922) on simultaneous heat and mass transfer at evaporation, where i showed
how the two phenomena influenced each other. The Lewis number is named after him, Sec. C.4.
Henrik Mohn, 1835–1916, Norwegian meteorologist, regarded as the founder of Norwegian mete-
orology. Head of the Norwegian Meteorological Institute from the beginning in 1866.
Fritjof Nansen, 1861–1930, Norwegian scientist, diplomat, oceanographer. Nobel peace price
1922.
Claude Louis Marie Henri Navier, 1785–1836, French civil engineer. Founded the scientific struc-
tural analysis. Put up the equation of motion for viscous fluids.
Thomas Newcomen, 1663–1729, British engineer and inventor. Constructed a steam engine oper-
ating at atmospheric pressure in 1712 – mainly used for pumping in mines and to lift water driving
water wheels.
Isaac Newton, 1642–1727 (b. 1643 in the Gregorian calendar), English mathematician and physi-
cist.
Ernst Kraft Wilhelm Nusselt, 1882-1957, German physicist. Work on heat transfer, the first to
analyze heat transfer in a practical manner. Adopted dimensional analysis (1915) to investigate heat
transfer in flows. Other work on thermodynamics, mass- and heat transfer.
Lars Onsager, 1903–1976, Norwegian-American chemist and physicist, graduated from NTH 1925,
went to USA. Nobel price in chemistry 1968 for fundamental work on irreversible thermodynamics.
Did also work on other subjects, among those, turbulence (Onsager, 1949), page 123.
Nikolaus Otto, 1832–1891, German engineer, constructed a four-stroke reciprocating engine with
premixing (the Otto engine) in 1876. This was the precursor to the gasoline engine used in most
automobiles. It is estimated that a thousand million Otto engines exist, about half of these in cars.
Jean Léonard Marie Poiseuille, 1799–1869, French physician and experimental physiologist. Stud-
ied among other the flows in arteries and narrow tubes. Proposed the the Poiseuille equation in 1840
(also known as the Hagen-Poiseuille equation), independent of Hagen (see above), who found the
same relation the year before.
Ludvig Prandtl, 1875–1953, German, regarded as the father of fluid dynamics; worked out the
basis of a common treatment of both liquid flow (hydrodynamics, hydraulics) and gas flow (aero
dynamics). Developed the boundary layer theory (see page 66) and worked on turbulence modeling
(page 40 and 52) and wing profiles. On the Prandtl number, see Sec. C.4.
Osborne Reynolds, 1842–1912, British engineer and physicist, best known for experimental and
theoretical work in fluidmechanics, see more on page 16 and 34, and Fig. C.2.
Lewis Fry Richardson, 1881–1953, English meteorologist and physicist. Did fundamental work
on numerical analysis of fluids, especially for weather forecasting, see page 19 and 121. On the
Richardson number, see Sec. C.3. Later, he introduced mathematical modeling as a tool in phsycol-
ogy and in the study of conflicts and war. Hunt (1998) has given a biography.
Carl-Gustaf Arvid Rossby, 1898–1957, Swedish-American meteorologist. Studied large-scale mo-
tions in air and introduced equations for these motions.
Carl Ernst Heinrich Wilhelm Schmidt, 1892–1975, German, known for work on mass and heat
transfer. On the Schmidt number, see Sec. C.4.
Robert Stirling, 1790–1878, Scottish clergy and inventor, best known for the Stirling machine (or
engine), patented 1816. The motivation for the invention is said to be that pastor Stirling saw many
people from his parish been injured by poor high-pressure boilers exploding. From 1843 to about
1920, quite a few smaller Stirling engines were in operation. Due to their cost, these were mainly
Figure C.2: The test bench of of Reynolds, where the experiments on transition to turbulence
were made. From Reynolds (1883). See also page 16 and 34.
replaced by electric motors when electricity distribution became widely available.

George Gabriel Stokes, 1819–1903, British mathematician and physicist. Found and has given
name to several laws/relations. The bes known is the relation between motion and drag of a sphere
in a fluid flow.
Brook Taylor, 1685–1731, English mathematician. Worked out (1715) among other the Taylor
series – hence, the Taylor microscales in turbulence has got their name after him.
Geoffrey Ingram Taylor, 1886–1975, English physicist and meteorologist. Studied eddying mo-
tions in the atmosphere on a voyage with a meteorological vessel in 1913. The next decades he
delivered a large number of fundamental work on statistical analysis of turbulence. Developed the
theory of diffusion of continuous motions that linked diffusion and dissipation. Laid much of the
foundation of turbulence theory. See also page 19. (On the term Taylor micro scale, see Brook
Taylor above.)
William Thomson, see lord Kelvin.
Giovanni Battista Venturi, 1746–1822, Italian physicist. Studied effects of contractions in channels
with fluid flow. Gave name to the Venturi meter and the Venturi nozzle/pipe.
Felix Wankel, 1902–1988, German engineer and inventor. Designed (1954) and constructed (1957)
an engine with a rotating piston. Wankel worked at the development department of NSU and the
first car with a Wankel engine was the NSU Spider in 1964.
James Watt, 1736–1819, Scottish instrument maker and inventor. Designed and built the steam
engine (1769) with a separate condenser. This made the engine much more effective than previous
machines (cf. Newcomen). The steam engine was extremely important for the industrial develop-
ment, and the Parliament adopted special laws that prolonged the patent and ensured that the most
modern engines were available for British industry.
Adolf Watzinger, 1879–1959, German-Norwegian mechanical engineer, educated in Leipzig and
Darmstadt. Professor in thermal engineering at NTH from the beginning in 1909. Established and
led the Thermal engineering laboratory to 1950.
Appendix D
Nokre ord om ord

Ukjende ord gjer ein tekst mindre tilgjengeleg. Den som skjønar orda som
er brukte, han skjønar også meir av innhaldet i teksten. Difor ei ordliste her.
Skjønar du ikkje vitsen – så skal du ikkje bry deg.
Mange likar å hente inn framandlandske ord og vendingar, særleg frå engelsk. Nokre gjer det berre
av rein latskap eller vankunne. Men om vi låner ordet, så er det ikkje dermed sagt at vi skjønar det,
snarare tvert om. At lånte ord er så mykje meir presise, kjem berre av at vi ikkje veit betre. Dei
er berre presise så lenge vi held utanfor alle andre tydingar og nyansar. I dette kapitlet dyrkar eg ei
interesse på sida av faget, og eg gjev meg høve til å ri nokre kjepphestar, – og kanskje vert du litt
klokare.
Mange av dei som utan blygsel brukar framande lånord, har det vanskeleg med seg sjølve når dei
støyter på – eller støyter seg på – eit norsk ord. Forfattaren Ivar Eskeland (1981) seier det slik:
«. . . nordmenn, det einaste folk i verdi so vidt eg veit, som stivheld inntil dauden på den læra, at det
er språket det er noko i vegen med når dei sidan i livet møter eit norsk ord dei ikkje er fødde med
kunnskap om.» Hjå andre kulturfolk, meiner han, er det vanleg å lære seg det nye ordet, t.d. ved
hjelp av ei ordbok.
Eg har gjort som Eskeland seier nordmenn (til vanleg) ikkje gjer. Dersom ein språkvitar skulle sjå
lista og seie at forklaringane er mangelfulle, så er svaret berre at «Blant dei blinde er sjølv den
einøygde profet». Eg likar å tru at eg veit meir om språkfag enn det ein språkvitar veit om turbulens
og forbrenning.
Utvalet av ord i er nokså tilfeldig, bortsett frå at eg har samla dei gjennom nokre års arbeid med faget.
Kjelder for ordforklaringane og opphavet er først og fremst Nynorskordboka (1986), Concise Oxford
Dictionary (1982) og Oxford English Dictionary (1933). Eg har også nytta Norsk Ordbok (1966ff),
Norsk Ordbog (Aasen, 1873), Grunnmanuskriptet (u. år), Latinsk ordbok (Steinnes og Vandvik,
1958) og Oxford Latin Dictionary (1968). Mange ord har fleire ulike tydingar på opphavsspråket,
og eg har ikkje teke med alle.
248 Appendix D Nokre ord om ord
D.1 Ordforklaringar
Ad- latin til.
Adiabatisk som ikkje kan passerast; berre nytta i varmelære, dvs. som varme ikkje kan gå forbi,
etter gresk «αδιαβατ oς » (adiabatos), av «α-» (a-): ikkje og «διαβαινω» (diabaino): gå forbi,
passere.
Adveksjon frå latin «ad-»:til og «vehere», «vect-»: bere, føre. I varme- og strøymingslære er det
somme forfattarar som vil halde av ordet konveksjon til å gjelde berre varmetransport. Transport
med strøyminga kallar dei adveksjon.
Affin (eng: «affine») av latin «affinis»: nært knytt til, slekt med, bokstaveleg grense til, av «finis»:
grense. Jamfør avsnitt D.5.
Algebra om italiensk frå arabisk «al-jebr»: sameining av oppbrotne delar; rekning med symbol,
bokstavar, likningar. Algebraisk er nytta om likningar som ikkje er differensiallikningar. Ein
algebraisk modell har ikkje differensiallikningar, jamfør avsnitt 6.5.
Algoritme etter namnet al-Khwarizmî, arabisk matematikar frå 800-talet, sjå avsnitt C.8, med
innverknad frå gresk «arithmos»: tal. Han skreiv ei bok om algebra, og boka gjorde det arabiske
talsystemet kjent i Europa. Opphavleg var «algoritme» ei nemning for rekning med det arabiske
talsystemet. I matematikk (i dag) og programmering: fullstendig og nøyaktig framgangsmåte for
dei rekneoperasjonane som krevst for å løyse eit problem, eintydig sett med instruksar for å løyse ei
oppgåve.
Anergi etter gresk «αν-» (an-): ikkje og «εργ oν» (ergon): arbeid. Nylaga ord, kring 1960.
Termodynamisk eigenskap: Den delen av energien som ikkje kan gjerast om til arbeid. Jamfør
eksergi, energi.
Anisotrop ikkje isotrop etter gresk «αν-» (an-): ikkje. Sjå isotrop.
Aritmetikk, aritmetisk gresk «αριθ µητ ικη» (arithmetike): kunsten å telje, av «αριθ µoς » (arith-
mos): tal. Læra om tal og rekning med tal.
Cascade (fransk, engelsk), sjå kaskade.
Co-, col-, com-, con- sjå Ko-, kol-, kom-, kon-.
Compute, computer (engelsk) frå latin «con-»: med, «putare»: rekne.
Concept (engelsk, fransk), sjå konsept.
Conflagration (engelsk/latin) av «con-»: med, saman, og «flagrare»: eld, flamme, brann. Stor og
voldsom brann.
Data noko som er rekna som kjent eller gjeve, opplysning, kjensgjerning. Frå latin, eigentleg
fleirtal av «datum», av «dare»: gje. Bøying på norsk (nyn.): data - dataet - data - dataa el. data,
(bokm.): data - dataen el. dataet - data - dataene. På engelsk: datum (sg.) - data (pl.) eller (om
fakta, opplysning): data (sg.) - data (pl.)
Deflagrasjon (eng: «deflagration») frå latin, av «de-»: frå kvarandre, til ymse sider, bort, og
«flagrare»: eld, flamme, brann. Som fagord: flamme utan sjokkbølgje, subsonisk flamme, sjå avs-
D.1 Ordforklaringar 249
nitt D.2.4.
Detonasjon frå latin, av «de-»: frå kvarandre, til ymse sider, bort, og «tonare»: torden. I all-
mennspråket: eksplosjon med kraftig smell. Som fagord: flamme med sjokkbølgje, sjå avsnitt D.2.4.
Di-, dis-, dif-, de-, des- latin: frå kvarandre, til ymse sider, bort, u-, ikkje. Forma på forstavinga er
avhengig av lyden som følgjer. Til dømes diffundere, dispergere, dissipere.
Differensial i matematikk: uendeleg små skilnader, frå latin «differentia»: skilnad.
Diffundere, diffusjon frå latin «dif-»: frå kvarandre, til ymse sider og «fundere»: renne; spreie,
renne bort, renne ut.
Dilatation (engelsk) utviding frå latin «di-»: frå kvarandre, til ymse sider, bort og «latare», «la-
tus»: vid. Mest nytta om volumtøyingsfart («volumetric dilatation»), men også om lengdetøyingsfart
(«linear dilatation»); sjå om tøying, avsnitt D.2.2.
Dispergere, dispersjon spreie seg, fordele seg, frå latin «dis-»: frå kvarandre, til ymse sider og
«spergere, spers-»: spreie. Til dømes om finfordelte partiklar i gass eller væske.
Dissipere, dissipasjon frå latin «dis-»: bort frå eller mangel og «sipare»: kaste; «dissipare»: kaste
bort, tape. Brukt om nedbryting/overføring av rørsleenergi til varme i strøyming, sjå likning (3.5)
og (A.44). Ordet er også brukt om nedbrytingsledd i andre turbulenslikningar.
Divergere fjerne (seg), skiljast; frå latin «di-»: frå kvarandre, til ymse sider, «vergere»: halle,
skråne. Jamfør konvergere.
Drag (engelsk), sjå slepekraft.
Dynamikk, dynamisk frå gresk «δυναµικoς » (dynamikos), adjektiv av «δυναµις » (dynamis):
kraft, styrke. I fysikk: lære om samanhengen mellom rørsle og kraft; rørslelære (eigentleg: kraftlære).
Eddy (engelsk) krinsrørsle i vatn og luft, lånord frå norrønt «iða», norsk ide, sjå avsnitt D.4.
-eder (eng: «-hedron», fl. «-hedra», adj. «-hedral»), sideflate; geometrisk lekam avgrensa av eit
visst tal flater av gresk «έδρα» (hedra): sete, plass, flate. Til dømes tetraeder (fire), heksaeder (seks),
oktaeder (åtte), polyeder (mange).
Eks-, ekso- etter gresk «εξ -, εξ o-» (eks-, ekso-): ut, ut av, (bort) frå.
Eksergi etter gresk «εξ -» (eks-): ut, ut av, (bort) frå og «εργ oν» (ergon): arbeid. Jamfør energi,
anergi. Nylaga ord ved Zoran Rant i 1955 (sjå Rant, 1956). Termodynamisk eigenskap: Den delen
av energien som kan gjerast om til arbeid. Også kalla tilgjengeleg energi (eng: «available energy»,
«availability»), arbeidsevne (tysk: «Arbeitsfähigkeit») eller reversibelt arbeid. Rant peika m.a. på
at ordet høver i alle dei store europeiske språkgruppene (tysk: «Exergie», engelsk: «exergy», fransk:
«exergie», spansk: «exergia», italiensk: «essergia», slaviske språk: «eksergija»).
Eksoterm som avgjev varme, som varme går ut av; sjå eks- og termo-. (Merk eng: «exothermic»,
men «isothermal».)
Eksplisitt uttrykt, uttrykkjeleg framstilt, avgjort, frå latin «explicitus», av «ex-»: ut, ut av, frå og
«plicare», «plicat-»: falde, brette; utfalda, utbretta; jf. implisitt og avsnitt D.2.5.
Eksplodere, eksplosjon (eng: «explode, explosion») frå latin, av «ex-»: ut (av), (bort) frå og
«plaudere»: klappe. Rask eller brå utviding, auke, utviding, gjerne med sterk lyd eller uro. Opphav-
leg jage med bråk og klapping. Som fagord, sjå avsnitt D.2.4.
Endo- etter gresk «ενδoν» (endon): innafor.
Endoterm som tek opp (bind) varme, som varme går inn i; sjå endo- og termo-, jf. eksoterm.
Energi frå gresk «εν-» (en-): inn, inn i og «εργ oν» (ergon): arbeid. Jamfør eksergi, anergi.
Entalpi etter gresk «εν-» (en-): inn, inn i og «θ αλπǫιν» (thalpein): varme opp. Ordet er nylaga
omkring 1900. Termodynamisk eigenskap, sjå i avsnitt A.3.1.
Entropi etter gresk «εν-» (en-): inn, inn i og «τρoπη» (trope): snunad, omlaging. Laga etter
mønster av «energi» av Clausius i 1865; tysk originalform «Entropie» (eng: «entrophy»). Termody-
namisk eigenskap, sjå i ei bok om termodynamikk.
Fenomen, fenomenologisk frå latin «phænomenon», av gresk «φαινoµǫνoν» (fainomenon): vise
seg.
Flamelet (latin, engelsk) diminutiv av latin «flamma»: flamme, altså lita flamme.
Fluid frå latin «fluere»: flyte, renne, «fluidum»: noko flytande. I fysikk brukt som samnemning på
gass og væske. I allmennspråket (helst fluidum) mest om væske, ofte om alkohol.
Fluiditet av fluidum. Evne til å flyte, det motsette av viskositet. Lite nytta i fysikk.
Fluks fransk/engelsk «flux», frå latin «fluxus», av «fluere»: flyte. Sjå avsnitt D.2.2.
Fysikk latin «physica», gresk «φυσ ικη» (fysike): kunnskap om naturen, naturlære, «φυσ ις »
(fysis): natur; vitskapen om den uorganiske naturen og fenomena der.
Fysisk, fysikalsk Bruken av desse to orda går ein del over i kvarandre på norsk. Begge har same
opphav, og kan tyde som gjeld fysikk, dvs. som fag, læra om den uorganiske naturen og fenomena
der. Men fysisk vert også brukt i meininga som gjeld lekamen, kroppsleg (fysisk arbeid, fysisk
skade, fysisk person). Svensk har «fysisk» – «fysikalisk» med om lag same bruk som på norsk. På
engelsk har dei berre «physical», på dansk berre «fysisk» (for begge meiningane). Tysk skil mellom
«physisch» (om kroppen) og «physikalisch» (fysikk som fag), og sameleis nederlandsk «fysiek» –
«fysisch» (eller helst «natuurkundig» om faget).
Gass ordet skal vere laga av den flamske kjemikaren van Helmont (1577–1644), etter den flamske
uttalen av gresk «khaos» (sjå kaos). Han nytta det som namn for karbondioksid (CO2 ).
Generell, general- frå latin: allmenn, uavgrensa, som gjeld allment, av «gener-, genus»: klasse,
slag, sort, stamme, rase; dvs. som gjeld (heile) klassen, slaget, sorten.
Generere frå latin «generare»: avle, av «gener-, genus»: klasse, slag, sort, stamme, rase.
Generisk frå latin, same opphav og tyding som generell. (Moteord i akademiske miljø.)
-gon, -gonal frå gresk: vinkel. Til dømes polygon (mangekant, eigentleg mange-vinkel), ortogonal
(rettvinkla).
Grid (engelsk) av «gridiron»: gitter (brukt om grill–, fengsels–, m.m.), overført tyding: nett,
nettverk. (Norsk grid: trygd, tilflukt er noko heilt anna.)
Gust (norsk) vindpust; (engelsk) brått og kraftig vindkast, storm, frå norrønt «gustr», «gjosa»:
strøyme, velle, drive, blåse (fram, ut). Sjå avsnitt D.2.1.
-hedron (engelsk; fl. «-hedra», adj. «-hedral»), sjå -eder.
Heksaeder (eng: «hexahedron») lekam med seks sider (firkantar), av gresk «έξ » (heks): seks og
-eder (sjå dette).
Heterogen ueinsarta, ulik, av ulike delar, av ulike slag, skiftande, av gresk «ǫ́τ ǫρoς » (heteros):
annan, ulik, annleis, – ofte i motsetnad til «homo-», også til «auto-», «iso-», «orto-»; og «γ ǫνης »,
«γ ǫνoς » (genes, genos): slag, art. Om heterogene flammer, sjå side 260.
Homogen einsarta, av like delar, av same slag, av gresk «óµoς » (homos): eins, same og
«γ ǫνης », «γ ǫνoς » (genes, genos): slag, art. Om homogen turbulens, sjå avsnitt 4.1. Om ho-
mogene flammer, sjå side 260.
Hydro- (også hydr- før vokal) som har med vatn (eller anna væske) å gjere, frå gresk «ύδρo-,
ύδρ-» (hydr(o)-), av «ύδωρ» (hydor): vatn. Brukt i mange samansetjingar, anten brukte i gresk,
eller laga seinare.
Hydraulikk, hydraulisk som har med vatn (eller anna væske) å gjere, gjennom latin «hydrauli-
cus», frå gresk «ύδραυλικoς », (hydraulikos) av «ύδωρ» (hydor): vatn (sjå hydro-) og «αυλoε»
(aulos): røyr. I gresk nytta om eit vassdrive musikkinstrument, «ύδραυλικoν oργ ανoν» (hy-
draulikon organon); seinare nytta i latin om andre vass-maskiner.
Implement sjå avsnitt D.3.2.
Implisitt innebere, men ikkje klart uttrykt, frå latin «implicitus», av «im-»: inn, inn i og «plicare»,
«plicat-»: falde, brette; innfalda, innbretta; jf. eksplisitt og avsnitt D.2.5.
In-, il-, im-, ir- frå latin med to ulike tydingar: 1) forstaving frå preposisjon «in»: i, inn. Til
dømes influere: flyte inn, inkludere: lukke inn, invitere: by inn. Merk særleg engelsk «inflamable»:
brennbar (og ikkje det motsette), av «inflame»: (setje) i flamme. 2) nektande forstaving «in-»: ikkje,
u-, mis-, van-. Forma på forstavinga er avhengig av lyden som følgjer.
Incinerator (engelsk) forbrenningsomn frå latin «in-»: i, inn, «cinis»: oske; «incinerare»:
gjere, brenne til oske. Brukt om omnar der hovudføremålet er å brenne opp noko, t.d. bosbrenning,
likbrenning (kremering).
Indeks frå latin «index»: peikefinger, jamfør «indicare»: melde, vise, peike på (eng: «index»,
fleirtal «indices»; tysk: «Index», fleirtal «Indexes, Indizes», også «Zeiger»: visar, peikar). Matema-
tisk språkbruk: Merketeikn som stend for tal, til dømes koordinatretningar. Også brukt om liste over
namn, stikkord, bøker, osb., om avlesingsmerke og om høvetal.
«Subindex»: Låg indeks, (tysk: «tiefgestellter Index», også «Fußzeiger»), «Superindex»: Høg in-
deks, (tysk: «hochgestellter Index», også «Kopfzeiger»). Sjå merketeikn. Sjå også side xv.
Inert utan evne til handling, rørsle eller motstand, av latin «iners»: treg, av «in-»: u-, ikkje, «ers,
ars»: evne, kunst. I kjemi: stoff som ikkje reagerer.
Inertia (latin, engelsk) tregleik; motstand mot endring.
Inkompressibel frå latin «in-»: ikkje, u-. Sjå kompressibel.
Intermittere, intermittens frå latin «inter»: mellom, «mittere» av «miss-»: late fare, sende. Om
skifting, (rykkvis) veksling mellom to tilstandar, noko som tek opphald frå tid til tid. Sjå avsnitt 7.3.
Iso- (somtid is- framfor vokal) frå gresk «ισ oς » (isos): lik, eins. Forstaving nytta i mange ord,
nesten alle vitskaplege. Det andre leddet er vanlegvis frå gresk, sjeldnare latin. (Det latinske tilsvaret
er «Equi-», på norsk ekvi-). Dei fleste orda er nye og forma etter mønster av «isotherm» (eins temper-
atur), «isothere» (eins middel-sommartemperatur), «isochimenal» (eins middel-vintertemperatur),
som A. von Humbolt innførte på fransk i 1817.
Isentalp lik entalpi. Sjå iso- og entalpi.
Isentrop lik entropi. Sjå iso- og entropi.
Merk: Ein vanleg skrivefeil er å blande saman «isotrop» og «isentrop».
Isobar likt trykk, etter gresk «βαρoς » (baros): tyngd; sjå iso-.
Isokinetisk lik rørsle. Sjå iso- og kinetisk. Nytta om måleteknikkar som tek prøver av ein gass
utan å forstyrre rørsla.
Isokor likt volum, etter gresk «χωρα» (khora): rom; sjå iso-.
Isolere stengje ute, skilje frå, frå italiensk «isolare», av latin «insula»: øy.
Isometri, isometrisk likskap i mål, målt med same mål. Sjå iso- og metrisk.
Isoplet lik i verdi, tal, mengd, kurve gjennom punkt med same verdi av ein variabel, etter gresk
«πληθ oς » (plethos): tal, mengd; sjå iso-.
Isopykn lik tettleik, etter gresk «πυκν oς » (pyknos): tett; sjå iso-.
Isoterm lik temperatur, sjå iso- og termo-. (Merk eng: «isothermal», men «exothermic».)
Isotrop like eigenskapar i alle retningane, frå gresk «τρoπη» (trope): snunad, vending; sjå iso-.
Om isotrop turbulens, sjå avsnitt 4.1.
Merk: Ein vanleg skrivefeil er å blande saman «isotrop» og «isentrop».
Kaos gresk «χαoς » (khaos): gap, svelg; uorden, forvirring. Merk at i matematikken har kaos ei
særskild meining som nemning på likningssystem med visse eigenskapar. Kaosteori er ikkje teori
om uorden og kaos i allmenn forstand, men teorien om slike likningssystem.
Kaskade (turbulens-, energi-) «cascade» er fransk for eit stegvis, gjerne kunstig, fossefall; frå
italiensk «cascata»: fall, fossefall. Om energikaskade for turbulens, sjå avsnitt 8.1 og 11.2.
Kartesisk av namnet Renatus Cartesius, latinsk form av Réne Descartes, fransk filosof og matem-
atikar, 1596–1650. Eit kartesisk koordinatsystem har rette linjer og vanlegvis rette vinklar, men kan
ha skeive vinklar. Sjå avsnitt D.5.
Kinetikk, kinetisk som gjeld rørsle, rørslelære frå gresk «κινητ ικoς » (kinetikos): som har
med/kjem av rørsle.
Ko-, kol-, kom-, kon- av latin «co-, col-, com-, con-»: med, saman, felles, sam-. Forma på
forstavinga er avhengig av lyden som følgjer. Til dømes koordinat, konveksjon, kollokvium, kom-
binere.
Konflagrasjon sjå conflagration (engelsk/latin). Stor og voldsom brann. Lite eller inkje nytta i
norsk.
Koordinat av latin «co-»: med, «ordinaire»: ordne, av «ordo»: orden; altså koordinat: med orden.
Komprimere, kompressibel frå latin «compressare»: trykkje saman; «com-»: saman; «premere»:
trykkje. Om skilnaden mellom kompressibilitet og variabel tettleik, sjå side 26.
Konsept frå latin «conceptum»: noko ein ber på, er svanger med, har avla; førestelling, tanke,
omgrep; noko utforma i tanken; av «con-»: med, saman, «cipere-, cept-»: ta. Hovudtydinga i
engelsk «concept» er tankebilete, førestelling (om noko), (allmenn-)omgrep. Frå tysk «Konzept» og
fransk «concept» har vi fått den vanlegaste tydinga i norsk: utkast, kladd.
Konservere frå latin «conservare»: ta vare på, av «con-»: med og «servare»: halde, ta vare på.
Kontra- av latin «contra»: imot, andsynes (jamfør ko-).
Konveksjon av latin «con-»: med og «vehere», «vect-»: bere, føre. I varme- og strøymingslære:
ført med, flyte med straumen. Sjå også adveksjon.
Konvergere frå latin «converge»: samle (seg), møtast; av latin «con-»: med, saman, «vergere»:
halle, skråne. Jamfør divergere.
Korrelat, korrelere, korrelasjon samsvar, (innbyrdes) samanheng, vekselverknad.
Kurve av latin «(linea) curva»: bogen linje, krum linje, boge; «curvus»: bøygt.
Kurvelineær med krumme linjer, krumlinja. Jamfør avsnitt D.5.
Laminær (eng: «laminar») av latin «lamina», fl. «laminae»: tynt lag, flak, plate. I fluid-
mekanikken: strøyming som ikkje er turbulent. Merk særskilt at ordet til vanleg ikkje vert nytta
i tydinga sjikta eller lagdelt strøyming – sjølv om det språklege opphavet kunne ha gjeve denne
meininga.
Lateral side-, -sidig, som gjeld sida på noko, av latin «latus»: side.
-logi lære, vitskap, av gresk «λoγ ια» (logia): lære, vitskap, «λoγ oς » (logos): ord, tale, tanke.
Mekanikk læra om apparat og maskiner, læra eller vitskapen om rørsler og krefter som verkar,
gjennom latin frå gresk «µηχνικoς » (mekhanikos), av «µηχανη» (mekhane): verkty, reiskap.
Mekanisk som gjeld maskiner eller mekanikk gjennom tysk, frå latin, jf. mekanikk.
Merketeikn norsk ord for latin «subscript»: lågt merketeikn, «superscript»: høgt merketeikn.
Meteorologi læra eller vitskapen om atmosfæren, vêr og klima, frå gresk, av «µǫτ ǫωρoς » (mete-
oros): noko som er oppe i lufta; sjå -logi.
Metrisk, -metri av gresk «µǫτρoν» (metron): mål. Jamfør avsnitt D.5.
Metrologi læra om mål og vekt, sjå metrisk og -logi.
Modell avbilding, biletleggjering, frå italiensk «modello», av latin «modulus»: mål, målestokk.
Normal av latin «norm»: rettvinkla figur som snikkarar brukte, dvs. rett vinkel.
Numerisk frå latin «numerus»: tal.

Orto- frå gresk «oρθ oς » (orthos): rett, beint. Til dømes ortografi (rettskriving), ortodoks (rett-
truande), ortogon (rettvinkla).
Ortogonal rett-vinkla frå gresk «ortho-»: rett- og «γ ων» (gon): vinkel. Eit ortogonalt koordinat-
system har rette vinklar, men treng ikkje ha rette linjer (jamfør kurvelineær). Sjå avsnitt D.5.
Plume (engelsk) fjør – særleg til pynt, fjørdusk – t.d. på hjelm, overført: røykstrime/sky frå
skorstein. I fluidmekanikk nytta om strøymingar som stig til vers med oppdrift.
Pneumatisk drive av (trykk)luft; som gjeld luft og gassar (dvs. i teknisk samanheng, omgrepet vert
også nytta i teologi og filosofi), frå gresk «πνǫυµατ ικoς » (pneumatikos): som har med vind eller
ånd å gjere, åndeleg, av gresk «πνǫυµα» (pneuma): vind, luft, pust, ande, ånd.
Produsere, produksjon frå latin «pro»: for, fram og «ducere»: leie, føre; jamfør «duct»: leidning,
kanal. Altså: framførsle.
Rate frå latin «(pro) rata (parte)»: (etter) fastsett (del). Ofte: Del eller mengd av pengar (t.d.
avbetalingsrate, fraktrate). Også tal, mengd pr tidseiing (t.d. produksjonsrate, vekstrate). Sjå også
om reaksjonsrate, avsnitt D.3.3.
Reagere, reaksjon etter latin «re-»: att, attende eller om, om att, igjen og «agere»: gjere; verke
attende.
Simulere etterlikne, frå latin «simulare», av «similis»: lik.
Slepekraft krafta (i rørsleretninga) som verkar mellom eit fluid og ein lekam når dei rører seg i
høve til kvarandre (eng: «drag (force)»; tysk: «Wiederstand (-skraft)»). Også kalla luftmotstand når
dét høver. Sjå avsnitt D.2.2.
Statikk, statisk frå latin, av gresk «σ τ ατ ικoς » (statikos): stå i ro, vere i jamvekt. Lære om krefter
som held kroppar i jamvekt; jamvektslære.
Statistikk, statistisk frå fransk og tysk, av latin «status»: stode, tilstand. Lære som gjeld innsam-
ling, tilrettelegging og ordning av talmateriale som kastar lys over mengdtilhøve; som gjeld slikt
talmateriale.
Ein statistisk modell for turbulens er ein modell som gjev uttrykk/likningar for statistiske stor-
leikar, dvs. middelverdiar, korrelasjonar – til skilnad frå storevje-simulering (eng: «large-eddy sim-
ulation»), direkte simulering (eng: «direct (numerical) simulation»), og fleire.
Stokastisk frå gresk «σ τ oχασ τ ικoς » (stokhastikos), «σ τ oχαζ ǫσ θ αι» (stokhazesthai): sikte
mot eit mål, gisse, av «σ τ oχoς » (stokhos): sikte, gisse. I statistikk: tilfeldig, sjå i ei bok om
statistisk metodelære.
Støkiometri etter gresk «σ τ oιχǫιoν» (stoikheion): element, grunnstoff og «-metri» (sjå metrisk).
Nylaga ord, innført av J.B. Richter i 1792 på tysk: «Stöchiometrie» (eng: «stoichiometry», også
«stoicheiometry»). Læra om masse- og mengdehøve for kjemiske reaksjonar.
Subindex (latin, engelsk) låg indeks. «Sub-»: under, nedantil, tett innåt. Sjå indeks.
Subscript (latin, engelsk) lågt merketeikn. «Sub-»: under, nedantil, tett innåt og «scribere»,
«script-»: skrive. Noko som er skrive attåt eit anna teikn, og plassert lågt i høve til dette teiknet.
(I mellomalderen tok norsk opp lånorda skrift og skrive frå latin «script», «scribere». Islandsk har
halde på «rita», jamfør engelsk «write».) MERK: Indeks («index») er eit «subscript», men ikkje alle
«subscript» er ein indeks. Sjå merketeikn og indeks.
Superindex (latin, engelsk) høg indeks. «Super-»: over. Sjå indeks.
Superscript (latin, engelsk) høgt merketeikn. «Super-»: over og «scribere», «script-»: skrive.
Noko som er skrive attåt eit anna teikn, og plassert høgt i høve til dette teiknet. Sjå subscript,
merketeikn og indeks.
Svervel norsk dialektord (Østerdalen, Gauldalen, m.m.) for kvervel.
Swirl skotsk-engelsk dialektord for «whirl»: kvervel, svarar til norsk svervel.
Syklus, syklisk (engelsk: «cycle, cyclic») gjennom latin frå gresk «κυκλoς » (kyklos): krins,
ring. Brukt m.a. om varmeprosessar og om strøymingar, sjå avsnitt D.2.1.
Teknikk, teknologi frå gresk «τ ǫχνικoς » (tekhnikos), «τ ǫχνoλoγ ια» (tekhnologia), av «τ ǫχνǫ»
(tekhne): kunst, handtverk, middel; -ikk av «-ικoς , -ικη» (-ikos, -ike): utøving, tame, som har å
gjere med (føreleddet) og -logi (sjå dette): lære, vitskap. Desse to orda vert nytta noko om kvarandre
i meininga metodar og verksemd for å lage og bruke maskiner og apparat. Skilnaden mellom orda
er at teknikk peikar meir i praktisk lei, medan teknologi peikar i meir teoretisk og vitskapleg retning.
Sjå meir i Råd om språk (1983), og sidene 12 og 15.
Termo- frå gresk «θ ερµoς » (thermos): varm, heit, «θ ερµη» (therme): varme, hete.
Termodynamikk av termo- og dynamikk (sjå desse). Varmelære, læra om samanhengen mellom
varme og andre energiformer, og om endringar i dei fysiske eigenskapane til stoff. Nokon meiner at
termostatikk ville vere ei rettare nemning på den delen som gjeld jamvekt eller som føreset tilnærma
jamvekt (kvasi-jamvekt) – dvs. den «vanlege» termodynamikken. Historisk skriv nemninga termo-
dynamikk seg frå W. Thomson (Kelvin) sin artikkel i 1851: On the dynamical theory of heat. Etter
dette har nemninga dynamikk hange ved fagområdet. Året før nytta Clausius ordet «Wärmelehre»
på tysk.
Tetraeder (eng: «tetrahedron») lekam med fire sider (trekantar), frå gresk «τ ετρα» (tetra), av
«τ ετ τ αρα»: fire og -eder (sjå dette).
Trans- frå latin: på hi sida, gjennom, tvers over.
Transport av latin «trans-»: på hi sida, gjennom, tvers over og «portare»: bere, bringe, føre.
Turbin frå fransk, av latin «turbo»: kvervel, kvervelvind, snelle; kringsviv.
Turbulens, turbulent sterk og uregelrett rørsle i væske eller gass, og i overført tyding uroleg,
opprørsk. Oxford Latin Dictionary (1968) viser til kjelder som nytta ordet tilbake til tidleg litterær
tid i latin, ikring 200 f.Kr. «Turbulentus» har ei vidare tyding på latin enn i moderne norsk og
engelsk. Felles for alle tydingane er at noko er eller vert gjort uroleg. Grunnordet er «turba»:
ståk, oppstyr, uro. Det er nært knytt til «turbo»: kvervel, kvervelvind, snelle; kringsviv. Begge
skal ha kome frå gresk «τ ύρβη» (turbe): uro, leven. Nemninga turbulens skal vere innført om
strøymingstilstanden av William Thomson (lord Kelvin) i 1887 (Rouse og Ince, 1957:212).
Vake opphavleg kjølstripe, råk, til dømes etter ein båt. Utlånt til engelsk: «wake». Som fagord er
det brukt om eit strøymingsfenomen der strøyminga vert bremsa opp bak ei hindring, lestrøyming. I
ei vake kan det vere bakevjer (soner med resirkulasjon). Sjå avsnitt D.2.1.
-variant (adjektiv) som endrar seg, skiftar av latin «varius»: ulik, mangfelt, skiftande.
Void (engelsk, frå gammalfransk dialekt) tom, ledig, jf. avsnitt D.2.2.
Wake (engelsk) lånord frå norsk «vake»: kjølstripe, råk. Sjå vake.
D.2 Ord om strøymingar og forbrenning

D.2.1 Særtilfelle av strøyming; Inndeling etter geometri og anna
Bakevje område bak hindring, utviding, e.likn., der strøyminga bøyer attende, resirkulasjonsom-
råde eller -sone (eng: «recirculation zone», tysk: «Rückströhmung»).
Blandingssjikt (eng: «mixing layer», tysk: «Trennungsschicht») to strøymingar langsmed kvaran-
dre (t.d. etter ei skiljeplate) som utvekslar (blandar) rørslemengd, masse og energi. Dersom ei
strøyming går langs stilleståande fluid, kan blandingssjiktet også kallast halv-jet eller jetgrense (eng:
«half jet», «jet boundary»; tysk: «Strahlgrenze»). Sjå figur 4.4.
Grensesjikt (eng: «boundary layer», tysk: «Grenzschicht»), på norsk også grenselag. Sjå avs-
nitt 4.3.
Gust vindkast, i fluidmekanikk særleg nytta om kraftige strøymingar eller brå vindkast oppe i
luftlaga i atmosfæren. Jamfør avsnitt D.1.
Jet (frå engelsk; tysk: «Strahl», «Freistrahl», også «Jet»). På norsk er også stråle nytta, men det
kan vere høveleg å halde av dette ordet for elektromagnetisk stråling (viktig i varmetransport, jamfør
avsnitt 12.2). Etter forma skil vi mellom rund jet, plan jet og radiell jet. Ein veggjet (eng: «wall
jet») er ein jet som ligg heilt inn til ein vegg. Det vert ein kombinasjon av jet og grensesjikt. Sjå
figur 4.4.
Kanal (eng: «duct», «channel»; tysk: «Kanal») inneslutta strøyming med vilkårleg tverrsnitt;
kan ev. vere open på oversida. Vanlege særtilfelle er plan kanal (sjå plan strøyming), røyr og open
firkanta kanal.
Kvervelgate (eng: «vortex street»; tysk: «Wirbelstraße») særskild kvervelstrøyming i laminære
vaker, gjerne knytt til von Kármán sitt namn. Sjå Schlichting (1979:18).
Roterande strøyming (eng: «rotating flow»; tysk: «Drehströhmung»). Strøyming i noko som
roterer, t.d. kanal i turbinhjul. Kan også vere nytta i same meining som svervel (eng: «swirl», tysk:
«Drall»).
Røyr (eng: «pipe»; tysk: «Rohr») aksesymmetrisk særtilfelle av kanal.
Sjikting, lagdeling (eng: «stratification»; tysk: «Schichtung»)
Sjikta strøyming (eng: «stratified flow»; tysk: «geschichtete Ströhmung») lagdelt pga. skilje i
temperatur (tettleik), samansetjing (t.d. olje/vatn), m.m. Sjå om richardsontal, avsnitt C.3.
D.2 Ord om strøymingar og forbrenning 257
Skjerlag, skjersjikt (enkle –, tynne –) (eng: «(simple, thin) shear layer») Strøyming med skjer-
spenning (og fartsgradient, tøying) i berre éi retning – dvs. ei enkel strøyming, som grensesjikt, jet,
røyr, osb.
Skjerstrøyming (eng: «shear flow»; tysk: «Scherströhmung») kan vere det same som skjerlag,
men er helst nytta i vidare meining om strøymingar med skjerspenningar (og fartsgradientar, tøyin-
gar) i fleire retningar. Ei turbulent skjerstrøyming er ei strøyming med turbulens-skjerspenning(ar).
Svervel (eng: «swirl»; tysk: «Drall») strøyming med fartskomponentar både i aksial og tangential
retning (og gjerne i radiell retning). Sverveltal (eng: «swirl number», tysk: «drall-Zahl») er høvet
mellom rørslemengd (fart) i tangential retning og aksial retning.
Vake (opphavleg norsk ord, sjå avsnitt D.1; eng: «wake»; tysk: «Nachlauf(-ströhmung)», «Wind-
schatten») del av strøyming som vert bremsa opp på lesida av (dvs. etter) ei hindring. Dersom
oppbremsinga er sterk, kan det danne seg bakevjer i vaka. Sjå figur 4.4.
Veggjet sjå jet.
Fri konveksjon (eng: «free convection») Strøyming som er driven av oppdrift, skilnader i tettleik.
Også kalla naturleg konveksjon. Sjå drøfting t.d. i Tritton (1988:171). Om ei strøyming er fri eller
tvungen, kjem i noko mon an på synsstad og synsvidd. Når vi ser på varme- og massetransport frå
ei overflate i friluft, vil vi rekne ein jamn vindstraum som eit tilfelle med tvungen konveksjon. Men
ser vi den same vinden i eit større perspektiv, ser vi kanskje ein luftstraum styrd av oppdrift.
Naturleg konveksjon (eng: «natural convection») Det same som fri konveksjon.
Tvungen konveksjon (eng: «forced convection») Strøyming som er driven av andre krefter enn
oppdrift, i teknisk samanheng t.d. av vifter. Jamfør fri konveksjon.
Enkle (skjer-)strøymingar har éin hovudgradient og ei hovudretning, som grensesjikt, røyr, osb. I
slike strøymingar er tilstanden i eit tverrsnitt lite eller inkje påverka av det som hender nedstraums.
Strøyminga kan reknast ved å «marsjere» nedover i straumretninga.
Samansette eller komplekse strøymingar har gradientar i to eller fleire retningar, og har ikkje éi
typisk hovudretning.
Aksesymmetrisk, plan, radiell ei eigentleg 3-dimensjonal strøyming som kan handsamast i eit 2-
dimensjonalt (plant) koordinatsystem. Eit aksesymmetrisk eller rotasjonssymmetrisk (eng: «axially
symmetrical», «rotational symmetrical»; tysk: «rotationssymmetrisch») tilfelle er uendra dersom
vi dreiar koordinatsystemet kring symmetriaksen; det er ingen tangentiale gradientar. Eit plant
(eng: «plane», «planar»; tysk: «eben») eller plan-parallelt tilfelle er uendra om vi flyttar koordi-
natsystemet langs den tredje aksen; det er ingen gradientar normalt på planet. Eit plant tilfelle er i
prinsippet uendeleg i den tredje retninga. I praksis er det den delen av ei brei strøyming som ikkje er
påverka av kantane. Eit radielt tilfelle har ikkje gradientar i akseretninga; altså normalt på radiell
og tangential retning.
Legg merke til at eit turbulensfelt alltid er 3-dimensjonalt (u ′i 6= 0 og u ′i 2 > 0 for i = 1, 2, 3), men
at middelfeltet kan ha null gradient i ei retning, slik at vi reknar 2-dimensjonalt.
Syklisk, periodisk strøyming eller grense; ei 2- eller 3-dimensjonal strøyming der like stykke går
= 50 m/s
(a) Nettverk (b) Fart i eit tverrsnitt
Figure D.1: Døme på periodisk (syklisk) strøyming: Utsnitt på 18◦ av eit ringforma
brennkammer. Dyser for brensel og luft er fordelte jamt kring brennkammeret. Det gjer
at strøyminga ikkje er aksesymmetrisk. Men brennkammeret kan delast inn i 20 like utsnitt.
Det som strøymer inn på eine sida må vere lik det som strøymer ut på den andre sida, og inn
i neste utsnitt (frå Inge R. Gran, 1994:178,187).
att, slik at utstrøyminga på ei side er lik innstrøyminga på den motsette sida. Døme: Eit ringforma
(annulært, eng: «annular») brennkammer har 20 like innblåsingsrøyr likt fordelt langs sirkelen. Då
er aksesymmetrien broten, men kammeret kan delast inn i 20 like «kakestykke» som kjem etter kvar-
andre i ein krins (syklus); sjå figur D.1. Eit anna døme: Ein kanal med fullt utvikla strøyming kan
delast inn i like stykke i lengderetninga. Eit stykke kan reknast syklisk på det viset at innstrøyminga
i ein ende er lik utstrøyminga i hin enden. Denne teknikken vert m.a. nytta for å lage grensevilkår i
direktesimuleringar og storevje-simuleringar av turbulent strøyming.
D.2.2 Storleikar i strøyminga

Fluks summen av konveksjon og diffusjon. For eit kjemisk stoff k er fluksen i x j -retning Jk, j =
ρYk u j +ρYk Vk, j , jamfør likning (A.17). For indre energi e er fluksen i x j -retning Je, j = ρeu j +q j ,
jamfør likning (A.44).
Slepekraft krafta (i rørsleretninga) som verkar mellom eit fluid og ein lekam når dei rører seg i høve
til kvarandre (eng: «drag (force)»; tysk: «Wiederstand (-skraft)»). Også kalla luftmotstand når dét
høver. Kan uttrykkjast ved ein slepekraftskoeffisient, motstandskoeffisient (eng: «drag coefficient»;
tysk: «Wiederstandsbeiwert», «Wiederstandskoeffizient»).
Sverveltal (eng: «swirl number», tysk: «drall-Zahl») er høvet mellom rørslemengd (fart) i tangen-
tial retning og i aksial retning, jf. svervel, avsnitt D.2.1.
Turbulensgrad, -faktor det same som turbulensintensitet.
Turbulensintensitet karakteristisk fartsskala for turbulensen (t.d. standardavvik) delt på middel-
D.2 Ord om strøymingar og forbrenning 259
p
verdien: σ/u = u ′ 2 /u. Kan også vere delt på ei referansefart, t.d. fristraumsfart for grensesjikt,
senterlinjefart for kanal eller jet. Sjå meir side 143.
Tøying (eng: «strain»; tysk: «Dehnung»): dimensjonslaust mål for deformasjon;
∂ξ
tøyingstensoren er Ci j = 12 ∂∂ξx ij + ∂ xij , der ξi er deformasjon (lengd) i x i -retninga.
Tøyingsfart (eng: «strain rate», «rate of strain», også «strain velocity»; tysk: «Dehngeschwindigkeit»);
1 ∂u i ∂u j
tøyingsfartstensoren er Di j = 2 ∂ x j + ∂ xi . Komponentar er lengdetøyingsfart (i = j ) (eng:
«rate of linear strain», «linear dilatation»), og skjertøyingsfart (i 6= j ) (eng: «rate of shear strain»).
Tensorane kan også vere definert som 2 gongar uttrykka slik dei stend her.
Volumtøyingsfart (eng: «(volumetric) dilatation», «volumetric strain rate», «divergence (of ve-
locity)»; tysk: «Dilatationsgeschwindigkeit»): Dii = ∂u i /∂ x i (divergensen av farta).
Void fraction I tofase-strøyming, gass/væske: Den delen (fraksjonen) av eit volum eller eit areal
som i tidsmiddel er fyllt av gassfase, jf. avsnitt D.1.
D.2.3 Forbrenningsutstyr
Brennar (eng: «burner») gjerne inndelt etter brensel (olje-, gass-, kolstøv-, mfl.) eller bruksområde
(gassturbinar, fyrkjelar, mfl.).
Brennkammer (eng: «combustion chamber», til gassturbin: «combustor»), jf. fyrrom.
Fakkel (eng: «flare») stor brennar for avbrenning av gass. Av tryggingsomsyn må ein i krisetilfelle
på olje/gass-plattformer, raffineri og liknande kunne brenne av store mengder gass på ein trygg måte.
Nokre stader brenner dei av all gass som følgjer med oljeproduksjonen.
Fluidized bed (engelsk) svevebed, kvervelsjikt, fluidisert bed, sjå avsnitt D.3.4.
Fyrkjel (eng: «furnace»; tysk: «Kessel»; svensk: «(ugns-)panna») med varme direkte frå fyring/for-
brenning; i motsetnad til kjelar med varme frå røykgass (utanfor forbrenningsromet) eller frå t.d.
spillvarme, kjernereaksjonar eller jordvarme.
Fyrrom forbrenningsrom, brennkammer i fyrkjel (eng: «(furnace) combustion chamber»; tysk:
«Brennkammer», «Feuerraum»; svensk: «pannrum»).
Incinerator (engelsk) forbrenningsomn frå latin «incinerare»: gjere, brenne til oske (sjå avs-
nitt D.1). Brukt om omnar der hovudføremålet er å brenne opp noko, t.d. bosbrenning, likbrenning
(kremering).
Motor (mest nytta om) kraftmaskin som omformar energien til drivkraft direkte, inne i maskina, t.d.
forbrenningsmotorar (Otto-, Diesel-, Wankel-, mfl.) og elektromotorar; i motsetnad til gassturbinar,
dampmaskiner, o.a.
Omn til industrielle føremål, t.d. glasomn, masomn, sementomn, smelteomn (eng: «furnace»);
til oppbrenning, t.d. kremeringsomn, bosbrenningsomn (eng: «incinerator»); vedomn, kakkelomn
(eng: «stove»); eller til oppvarming, til steiking eller baking (eng: «oven»). Merk at engelsk «oven»
ikkje svarar heilt til norsk omn og svensk «ugn».
Oven (engelsk) omn til steiking eller baking; små omnar til metallurgiske/kjemiske/termiske pros-
essar, jf. omn.
D.2.4 Flammer
Conflagration (engelsk/latin) vert nytta meir allment om ein stor og voldsom brann, jf. defla-
grasjon. Sjå også avsnitt D.1. Lite eller ikkje nytta i forbrenningslære.
Deflagrasjon (eng: «deflagration») Forbrenningsbølgje (flamme) som breier seg i ei homogen
reaktantblanding (forblanda) med ei fart mindre enn lydfarta (subsonisk), jf. detonasjon.
I litteraturen kan vi finne at orda forbrenningsbølgje (eng: «combustion wave») eller flamme er nytta
i denne meininga. Sjå også avsnitt D.1.
Detonasjon (eng: «detonation») Forbrenningsbølgje (flamme) som breier seg i ei homogen reak-
tantblanding (forblanda) med ei fart større enn lydfarta (supersonisk). Ei slik bølgje (forbren-
ningssone) ligg alltid like bak eit sjokk (diskontinuitet i trykk, temperatur og tettleik). Temperatur-
og trykkauken i sjokket medverkar til å tenne eller halde reaksjonen ved like, medan sjokkbølgja
vert driven av varmeutviklinga i reaksjonen. Jamfør deflagrasjon.
Ordet vert også nytta i andre samanhengar der ei sjokkbølgje breier seg, t.d. frå sprengstoff og
atombomber, og i supernova-eksplosjonar. Allment (særleg i engelsk) vert ordet nytta i meininga
sprenging eller eksplosjon. Sjå også avsnitt D.1.
Eksplosjon (eng: «explosion») Allment er dette ordet knytt til støy og øydelegging når ei trykkbøl-
gje breier seg. Som fagord vert det nytta på ulike vis, og vi kan finne fleire definisjonar eller fork-
laringar.
I forbrenningslæra vert det definert eller forklara (ikkje alle kallar det ein definisjon) med svært rask
og aukande varmeutvikling på grunn av ein kjemisk reaksjon i tilnærma homogene reaktantblandin-
gar (sjå t.d. Chomiak, 1990:133; Glassman, 1996:222; Kuo, 1986:233; Williams, 1985:576). Det
vert skilt mellom termiske eksplosjonar og eksplosjonar med greinande kjedereaksjonar. Dei siste
kan i visse tilfelle vere isoterme.
Nokon tek med at det kan vere trykkoppbygging i staden for varmeutvikling. Andre tek med
at ei trykkbølgje ofte følgjer med, men at dette ikkje er eit krav i definisjonen. Endå andre tek
med at det ikkje treng vere ei flamme (forbrenningsbølgje) som breier seg. I det siste tilfellet kan
varmeutviklinga (eller trykkoppbygginga) t.d. vere driven av ein kjernereaksjon eller trykkavlasting
med brå fordamping (dampeksplosjon – slik som når høgtrykkskjelar rivnar). Sjå også avsnitt D.1.
Diffusjonsflamme (eng: «diffusion flame»), flamme der reaktantane vert blanda ved diffusjon
(molekylær og turbulent) inn mot reaksjonssona. Vanlegvis det same som uforblanda flamme (sjå
avsnitt 10.1).
Forblanda flamme (eng: «premixed flame») reaktantane er perfekt blanda før den kjemiske reak-
sjonen (sjå avsnitt 10.1).
Homogen flamme flamme med reaktantar i berre éin fase (gass). Om homogen, sjå avsnitt D.1.
Heterogen flamme flamme med reaktantar i ulike faser, eller med andre faste/flytande stoff som
verkar inn på reaksjonen. Det kan vere fast karbon (koks, eller koksrest etter olje eller ved) som
reagerer med oksygen i luft. Nemninga vert nytta når brenslet er fast eller flytande, sjølv om det
fordampar før sjølve reaksjonen. Det kan også vere ein reaksjon mellom gassar i ein katalysator.
D.3 Ymse ord i fag og lag 261
Om heterogen, sjå avsnitt D.1.

Slokne, sløkkje Ein grammatisk detalj: Ei flamme sloknar (intransitivt verb) når brenslet tek
slutt. Du sløkkjer ei flamme (transitivt) når du får ei flamme til å slokne. (Bokmål: slokne/slukne;
slokke/slukke). På engelsk heiter det «extinguish»: sløkkje, men dei har ikkje noko ord for slokne.
Uforblanda flamme (eng: «non-premixed flame») Reaktantane er ikkje blanda før den kjemiske
reaksjonen (sjå avsnitt 10.1). Vanlegvis det same som diffusjonsflamme, jf. forblanda flamme.
Uforblanda kan også vise til tilførsla av reaktantar. Då fortel nemninga at reaktantane er tilførte kvar
for seg, men vert (delvis) blanda i strøyminga fram mot reaksjonssona. Eit døme er lyfta flammer.
D.2.5 Ord om likningar

Eksplisitt likning Variabelen kjem uttrykkjeleg fram frå likninga, dvs. slik at han ikkje er uttrykt
som ein funksjon av seg sjølv.
Implisitt likning Variabelen kjem fram som ein funksjon av seg sjølv.
Konserveringslikning (eng: «conservation equation») Denne nemninga vert nytta på to måtar:
(1) generelt om balansar, særleg for masse, energi og rørslemengd; (2) om likningar på ei viss
form, også kalla likninga på konservert form, konserveringsform, konservativ form eller divergens-
form (eng. også: «conservation form», «conservation-law form», «conservative form», «divergence
form»). Dette er partielle differensiallikningar der koeffisientane til dei deriverte anten er konstantar,
eller den deriverte av dei ikkje er med i likninga (Anderson, Tannehill og Pletcher, 1984:50f). Denne
eigenskapen ved likningane er viktig for den numeriske løysinga. Til dømes er kontinuitetslikninga
i likning A.1 på konservert form, medan likning A.2 ikkje er det. Like fullt har likningane same
innhald. Begge kan kallast konserveringslikning for masse, med tyding (1) ovanfor.
Transportlikning Vanleg nemning på balanselikningar uttrykte som partielle differensiallikningar;
med transent ledd, konvektivt ledd og eventuelt diffusivt ledd og kjeldeledd. Sjå likning 1.1 side 8.
D.3 Ymse ord i fag og lag

D.3.1 Turbulent energi eller turbulensenergi?1
Det er vanleg ordbruk å skrive turbulent kinetisk energi, turbulent viskositet, og så bortetter. Men
språkleg sett er turbulent eit adjektiv som fortel at noko er uroleg, – eller i vårt tilfelle, at noko
fluktuerer. Dei to storleikane er middelverdiar, og skulle såleis heite (kinetisk) turbulensenergi og
turbulens-viskositet. Ordet kinetisk tyder rørsle og er overflødig i denne samanhengen. Turbulensen-
ergi må, per definisjon, vere rørsleenergi.
På engelsk var turbulent kinetic energy og turbulent viscosity det vanlege tidlegare. Desse nemningane
er framleis nytta, men mange skriv no turbulence kinetic energy og turbulence viscosity. Det ser ut
1 Avsnittet er i hovudsak henta frå avhandlinga, Ertesvåg (1991:7).
til at mange forfattarar skifta skrivemåte like etter 1980. Seinare har somme forfattarar har byrja å
utelate ordet kinetisk, dvs. dei skriv turbulence energy.
Eg har lese for lite litteratur på tysk til å seie så mykje om ordbruken på dette språket. Men til dømes
Prandtl (1945) skriv Turbulenzenergie.
D.3.2 Implementere – komplett: berre fyll

Dette er eit moteord som vert nytta om likt og ulikt. Det har ei mangslungen ordhistorie, og eg tek
det med mest for moro skuld. Opphavet er latin «in-,im-»: inn,i, «plere»: fylle. «Implere»: fylle i,
fylle inn er eit verb. I overført tyding bruke, ta i bruk. Latin lagar substantiv av verb ved å hekte på
endinga «-mentum», altså «implementum», fleirtal «implementa».
Engelsk har lånt substantivet i forma «implement» og med tydinga verkty, instrument, hjelpemiddel.
Engelsk lagar verb direkte av substantiv på «-ment». Verbet «implement» har tydinga fylle opp,
leggje til, setje i verk. Men verb kan substantiverast, altså «implementation».
Norsk har lånt verbet og verbalsubstantivet frå engelsk: «implementere», «implementasjon». Den
eigentlege tydinga vert «implementasjon»: det å gjere det å fylle inn. Tenk på det du neste gong du
implementerer eitt eller anna.
Høvelege norske uttrykk for «det å gjere det å fylle inn», kan til dømes vere leggje inn ein modell i
eit program, å setje i verk, setje ut livet ein organisasjonsplan, innarbeide nye reglar i eit eksisterande
lovverk.
Eit anna ord får vi også av latin «plere»: fylle, men med forstavinga «com-»: med, saman. «Com-
plere»: fylle opp. Substantivet «complementum».
Engelsk har substantiv og verb «complement» med tydinga den som fullfører, gjer fullstendig (sub-
stantiv) og fullfører, gjer fullstendig (verb).
Norsk har fått dette ordet, ikkje frå engelsk, men frå tysk og fransk: «komplement»: tillegg, utfylling,
«komplettere»: gjere fulltalig, fullstendig; og «komplett» (adjektiv): fulltalig, fullstendig. Det heiter
altså ikkje «komplementere» og «komplementasjon».
D.3.3 Reaksjonsrate og reaksjonsfart

I denne boka har eg nytta namnet reaksjonsrate for storleiken Rk , som er kjeldeleddet i transport-
likninga for massekonsentrasjonen ρk = ρYk ; sjå likning (1.5) eller (A.16). Utan transport (konvek-
sjon og diffusjon) kan vi skrive dρk /dt = d(ρYk )/dt = Rk .
På engelsk vert dette kalla «reaction rate» og på tysk (Warnatz og Maas, 1993) «Reactionsgesch-
windigkeit». Kjemikarane brukar ofte reaksjonsfart (bokmål også «reaksjonshastighet») som namn
for storleiken dck /dt. Såleis skulle det vere grunnlag både frå norsk og tysk ordbruk for å bruke
reaksjonsfart for storleiken Rk .
Når eg vel reaksjonsrate på norsk, er det utfrå ei fagleg grunngjeving. Kjeldeleddet for massen av
eit stoff fortel ikkje utan vidare noko om kva fart ein reaksjon har. Frå likning (1.12) ser vi at for ein
reaksjon A + B → P kan kjeldeleddet for stoffet A uttrykkjast R A ∼ k1 c A c B (med l = 1). Dersom
D.3 Ymse ord i fag og lag 263
reaksjonen er rask (har stor «fart»), er k1 stor. Likevel kan R A vere liten dersom c A og/eller c B er
liten. Omvendt treng ikkje R A vere liten, sjølv om reaksjonen er etter måten sein.
Storleiken Rk fortel kor stor mengd av stoff k som vert danna pr tidseining, og då er rate det mest
høvelege ordet. Skulle vi bruke ordet reaksjonsfart, måtte det vere om kl .
D.3.4 «Fluidized bed»: Kvervelsjikt, sandseng, eller ... svevebed?

Dette gjeld ein teknikk der ein bles gass (vanlegvis luft) opp i gjennom sand slik at sandkorna «flyg».
I sanden kan brensel – faste eller tungtflytande – tilsetjast og brennast. Visse tilsatsstoff kan reagere
med forureiningar; t.d. kan kalk tilsetjast og reagere med svovel til gips. Sanden oppfører seg på
mange måtar som eit fluid – difor «fluidized». Teknikken vert også nytta til tørking og gassreinsing.
Kva skal vi kalle barnet på norsk?
På engelsk heiter det «fluidized bed»; «bed» kan tyde både seng og lag, sjikt – t.d. eit lag kol på
ei fyringsrist kan kallast «fuel bed»; altså brensellag som er gjort flytande eller (sand/brensel)seng
som er gjort flytande. Direkte lånt vert det fluidisert bed – om ein då ikkje svelgjer heile kamelen
med hud og hår, og brukar den engelske skrivemåten i norsk tekst. Nokon vil meine at det ikkje vert
godt norsk på nokon måte. Om det fagleg og sakleg sett er eit godt nemne, kan også diskuterast.
I tysk ser det ut til at «Wirbelschicht» er innarbeidd. Den norske varianten er kvervelsjikt, som m.a.
Rådet for teknisk terminologi (RTT – ein privat organisasjon) tilrår. Haken ved dette ordet er at det
ofte slett ikkje er snakk om eit sjikt, men at den svevande sanden fyller heile forbrenningsromet og
meir til.
Lat oss ta med at ein bed(d) er eit norsk ord (finst i mange norske dialektar; norr: «beðr»). Det kan
tyde (under)dyne, pute; eller botnen under eldstaden. (Eit bed i tydinga grønsakseng, blomsterseng,
er innlånt frå tysk i nyare tid.)
Sjølv har eg mest sans for ei nemning som også er brukt i svensk: svevebed.
D.3.5 «Large-eddy simulation»: Storevje-simulering

Denne reknemetoden har vi vore innom side 142. På engelsk heiter det «large-eddy simulation»,
forkorta «LES». – Kva skal vi så kalle det på norsk?
Storskala-simulering er eit ord som kunne høve. Vi simulerer eller løyser ut dei store turbulensskalaene;
medan verknaden av små skalaer må modellerast. Men denne nemninga vert også nytta for å
fortelje at ein har rekna eit stort tilfelle, dvs. eit fullskala-tilfelle (t.d. oljeplattform), i motsetnad
til utrekningar for eit laboratorietilfelle (samanliknar med måledata for ein skala-modell). Stor-
kvervel-simulering er ei direkte omsetjing – men nokså krunglete å seie. Ho er også ulik på nynorsk
og bokmål.
Språkleg sett er engelsk «eddy» ikkje anna enn eit norsk utlån; ide, ida (trykk på i-en), norrønt «iða».
Den mest direkte omsetjinga vert såleis storide-simulering. Men, sidan ide ikkje er eit ord nordmenn
flest «er fødde med kunnskap om» (jamfør side 247), er det lita von om at dette skulle slå gjennom.
Då er evje eit ord fleire kjenner (t.d. bakevje) – difor kort, greitt og norsk: storevje.
D.4 Ord om rørsle i luft og vatn

Ord om krinsrørsle
Alle kjenner orda kvervel (bokm: «virvel»), bakevje og skypumpe. Ved samansetjing med vind
eller storm får vi kvervelvind og vindkvervel, kvervelstorm og vindknute. Men så er det gjerne
slutt med norsken for mange. Vi held fram med «sirkulasjon» (frå latin), «syklon» (frå gresk),
«tornado» (frå spansk), «tyfon» (frå kinesisk) , malstraum (frå nederlandsk). Opphavet til orkan
(spansk: «huracan») har også denne meininga. I fluidmekanikk låner vi «eddy» (opphavleg av
norrønt «iða», norsk ide; i Trøndelag «udu», «uddu») og endåtil «whirl» og «swirl» (= kvervel) frå
engelsk, «vortex» (latin, engelsk), «turbulens», «turbiditet» og fleire andre «turbo»-ord (gresk, latin,
engelsk).
Det norske språket har eit mangfelt ordfang om krinsrørsle i vatn og luft, – det skulle berre mangle i
eit land som er rikt på dialektar, har nok av strøymande vatn og nøgda vêr og vind. Hovudkjelda her
er Grunnmanuskriptet (u. år) og Norsk Maalbunad (Aasen, 1876). Dei fleste orda stend i Nynorsk-
ordboka (1986).
Verb: firle, grøyvle, ide, kvervle (kvirvle, kverle, kvirle, virvle), røyle, sverve, svirle, svirre, tvirle.
Substantiv: kvervel (kverel, kveril, kverle, kvirvel, kvorvel, kvørvel, vervel, virvel), kvelv (gvelv),
kverv (kvarv, gvarv, varv), røyl, sveivle, svivle, sverkel, sverv, svervel, tull, tunn, tvingle, tvirle.
Mest om vatn: evje, bakevje, oppevje, ide (ida, ile, ie, ea, odo, udu, uddu, og fleire), ave, atterbere,
attide, atterkippe, atterreke, atterrenning, bakide, bakstraum, oppide, ti, flatstraum, grøyvl, havsvelg,
hovade, keile, kjel, korgamme, kringide, kringvode, kringsviv, ringstraum, malstraum, snaraevje,
snaride, snatrapull, straumide, straumkvervel, sveiv, sveivide, tunnevje, tunnide, kvitbulle, paul,
pytt, vodekall, øse, revje, oppfar.
Mest om luft: firlevind, kvervebye, kvervelflage, kvervelvind (kverlevind), kvervelknut (-knute,
kverelknut), røyksvivle, skypumpe, skyrøyl, stormkvervel, snuring, tvinnaknut, tvirleflage, tvir-
levind, tvirleguste, tulleflage, tunnarknute (tynnaknute), vindfirle, vindkall, vindknute, vindkvervel.
Andre rørsler
Luftrørsle: ande (vindande), blekk, blåst (blåster), bris, drag, eim, gjot, gnæs, gone, gos, gov, gråe,
græle, gule (havgule), gust (dalgust, fjellgust, vindgust), hå, kjøl (morgonkjøl), kule, kuling, kåre,
luftning, lygne, musk, novtot, rak, rek, røne (nordrøne, landrøne), skjelle, sno (dalsno, elvesno,
fjellsno, kaldsno), snære, storm, trekk, vind (lettvind, skoddevind, vindande, vinddrag, vindpust,
vindkule).
Motvind, landvind, nordavind, osb., etter retning og tilhøve.
Brå rørsle: brodd, brose, brysje, byge, byl, flage (kasteflage, rokflage, vindflage), fuke, føykje, garde
(drivgarde, brågarde, elgarde), gjerde, gysje, gøyve, kast (bråkast, kastevind, vindkast, stormkast),
knut, kule (vindkule), kvervel, rose, skjør, spong, tunn.
Vassrørsle: Drag, flaum, foss, gir, gjot, sig (oppsig), straum (andstraum, elvestraum, havstraum,
motstraum, tverrstraum, tvestraum, understraum), straumgir, straumrand, streng, strind, vell, ågang,
åsog.
D.4 Ord om rørsle i luft og vatn 265
Figure D.2: «Maalstroom», den malande straumen, kalla nederlandske sjøfolk den kraftige
og farlege havstraumen mellom Lofotodden og Værøy, Moskstraumen. I forteljingane vart
straumen til eit skremande havsvelg, ei kverveltrekt som slukte skip og mannskap. Jules
Verne let kaptein Nemo og ubåten (nesten) gå under her. Edgar Allan Poe («A descent into
the Maelstrom», 1841) skildrar eit sluk meir enn ei halv mil i diameter som heller førtifem
grader innover mot svelget i midten. Figuren viser eit utsnitt frå «Carta marina», utgjeve av
den svenske (katolske) eksilbiskopen Olaus Magnus i Venezia i 1539. Straumen er ett så
stort svalg, ..., att det plötsligt kringsvallar och i ett ögonblick uppslukar de sjöfarande som
ovarsamt ... (Olaus Magnus, 1555:2-7).
Bekk, elv, flaum, flod, grov, keile, kvisl, log (jf. Lågen), lon, løk, sikl, silder, vassfar, veke, å.
Alde (langalde, tungalde), brim, brot (fallbrot, grunnbrot), brotsjø, bærling, bølgje, båre (småbåre,
storbåre, straumbåre, vindbåre), demlesjø, dønning, fall, havbåre, oppfar, revje, sjø (havsjø, krapp-
sjø, manglesjø, stoplesjø, straumsjø), sjøgang, skavl, skol (båreskol), skval, våg.
Driv (sjødriv), drivgarde, drivkast, foss, fræning, gos, rok (rokflage, rokfoss), skum (båreskum),
skvett, sprett, sprut.
Verb: Drive, flyte, fløyme, renne, sige, sikle, sildre, sive, skylje, strøyme, tyte.
D.5 Ord om koordinatsystem og nettverk

Det er sagt at «Gud skapte ikkje verda i kartesiske koordinatar.» Iallfall er det
slik at somtid ønskjer vi å nytte andre koordinatsystem.
Affin transformasjon lineær transformasjon (Sokolnikoff 1964:10).

Galileo-transformasjon eller galileisk transformasjon, etter Galileo Galilei, 1564–1642; også
kalla newtonsk transformasjon, er eit likningssett som fortel at det kjem an på kva du ser det i høve
til. Det er Galileos relativitetsprinsipp i klassisk mekanikk (i motsetnad til Einsteins relativitetsprin-
sipp i relativistisk mekanikk). Det knyter saman rom- og tidskoordinatar for to system som rører
seg med konstant fart i høve til kvarandre. Ei rørsle som er kjend i det eine koordinatsystemet, kan
uttrykkjast i det andre koordinatsystemet ved ein slik transformasjon. Sjå døme side 65.
Transformasjonen byggjer formelt på tanken om at tid og rom er absolutte; at lengde, tid og masse
er uavhengig av rørsla til observatøren, og at lysfarta er avhengig av relativ-rørsla til observatøren.
Motsvaret i relativistisk fysikk er Lorentz-transformasjonar. Dei byggjer på at lysfarta er konstant
og uavhengig av rørsla åt observatøren eller kjelda, og at lengde, tid og masse er avhengige av rela-
tivrørsla.
Galileo-transformasjonar kan nyttast når farta er mykje mindre enn lysfarta.
Ein galileisk invariant er ein storleik som er uendra ved ein Galileo-transformasjon, t.d. aksel-
erasjon.
Heksaeder (eng: «hexaheder», fl. «-hedra», adj.: «-hedral») lekam eller volum med seks sideflater
(firkantar), t.d. om celle i nettverk. Sjå avsnitt D.1.
Kartesisk av namnet Renatus Cartesius, latinsk form av Réne Descartes, fransk filosof og matem-
atikar, 1596–1650. Eit kartesisk koordinatsystem har rette linjer og vanlegvis rette vinklar, men kan
ha skeive vinklar.
Koordinatflate flate der éin koordinatverdi er konstant og dei to andre varierer (Sokolnikoff,
1964:113).
Koordinatkurve det same som koordinatlinje (Irgens, 1982:2).
Koordinatlinje linje der éin koordinatverdi varierer og dei to andre er konstante (Sokolnikoff,
1964:113). Koordinatkurve.
Kurvelineære koordinatar koordinatlinjer langs to ulike koordinatar kryssar kvarandre éin og
berre ein gong. Koordinatlinjer langs same koordinat kan ikkje krysse kvarandre. Ei koordinatlinje
D.6 Ein stad å møtast 267
kan ikkje dele seg (forgreining) (Sokolnikoff, 1964:113).

Metrisk koordinat koordinat der koordinatdifferansen mellom to punkt på koordinatlinja er lik
eller proporsjonal med lengda av kurvestykket mellom punkta. Døme: For sylinderkoordinatar (po-
lare koordinatar) (r ,θ ,z) er r og z metriske koordinat, men ikkje tangentialkoordinaten θ . Ein logar-
itmiske akse er ikkje metrisk.
Ortogonalt koordinatsystem har rette vinklar, men treng ikkje ha rette linjer (jamfør kurvelineær).
Ortonormal rett-rettvinkla. Ortonormale vektorar: Tre vektorar der kvar vektor stend normalt på
dei to andre, og kvar har lengd 1 (Sokolnikoff, 1964:7).
Quadrilateral (engelsk, latin) som gjeld fire sider, firkant. Om firkanta celler i eit todimensjonalt
nettverk.
Tetraeder (eng: «tetraheder», fl. «-hedra», adj.: «-hedral») lekam eller volum med fire sideflater
(trekantar), t.d. om celle i eit tredimensjonalt nettverk. Sjå avsnitt D.1.
D.6 Ein stad å møtast

kan vere forum, kollokvium, konferanse, seminar, symposium, utval, komité, førelesing, møte, ...
Forum frå latin torg, møtestad, samlingsstad, også brukt om tribunal, domstol (romersk).
Førelesing ordet har opphav frå den tida skulen/universitetet hadde berre éi bok av kvart slag.
Då måtte ein lesar (jf. «lektor», avsnitt D.7) lese føre og studentane høyre etter. Jamfør engelsk
«lecture»: (opp-)lesnad.
Gruppe ordet skal ha kome, via italiensk og fransk, frå germansk «kruppaz»: rund masse, klump.
Kollokvium frå latin «col-»: med, saman, i lag; «loquium» av «loqui»: snakke, tale; altså samtale,
snakke saman.
Komité frå fransk, tiltru, overlate; av latin «com-»: med, saman, i lag; «mittere» sende; sende
saman.
Konferanse frå latin «con-»: med, saman, i lag; «ferre»: bringe, bere, føre; «confer»: bere saman,
føre saman.
Seminar frå latin «semen»: frø; «seminarium»: drivbenk, planteskule, – stad der ein legg ned
(kunnskaps)frø. Merk: Ordet har litt ulike meiningar i amerikansk engelsk, britisk engelsk og norsk
(sjå i ei ordbok).
Symposium frå gresk, av «σ υν-» (syn-): med, saman, i lag; «πoτ ης » (potes): drikkar; altså
drikkelag.
D.7 Ord om folk

I denne lista er det ein del titlar lesarane har, eller som dei vonar å kunne
smykke seg med, eller som nokon i omgjevnadene har. Har du tenkt på kva det
er vi kallar kvarandre? Og burde ikkje fleire få greie på kva ingeniør og student
eigentleg tyder? Eller, kanskje ikkje?
Amanuensis frå latin «a manu»: med hand; «-ensis»: tilhøyre; skrivetræl, skrivekunnig træl,
skrivar.
Assistent frå latin «assistere»: stille seg bortåt, stå attmed; «assistens»: til stades; av «ad»: til,
«sistere»: stille.
Dekanus frå latin, av «decem»: ti (10); leiar for ei gruppe på ti (soldatar, munkar, prestar, m.m.).
Disponent frå latin «disponere»: setje (bort) på ymse stader; «dis-»: frå kvarandre, til ymse sider,
bort; «ponere»: setje.
Direktør frå latin «director»: styrar, av «di-»: frå kvarandre, til ymse sider, bort; «regere»: råde,
styre.
Doktor frå latin «docere, doct-»: undervise, lære (frå seg). Altså: lærar.
Dosent same opphav og tyding som «doktor».
Forskar ein som driv vitskapleg gransking, frå tysk.
Ingeniør om fransk frå mellomlatin «ingenium»: (krigs)maskin, latin «ingenium»: huglag, skarp-
sinn, oppfinningsevne.
Jus docendi (latin) rett til å undervise; ved promosjonen av ein ny doktorand vil han/ho få tilsagt
«jus docendi» av rektor.
Kandidat frå latin «candidatus»: kvitkledd, kledd i kvit kappe.
Kollega frå latin «col-»: med, saman, i lag; «legare»: velje, utpeike; den medvalde.
Konsulent frå latin «consultare, consulere»: ta råd, spørje.
Lektor frå latin «legere» samle, lese; opplesar, ein som les (for andre). Jamfør «lektie», «leksjon»,
«lekse»: lesnad. Sjå førelesing, avsnitt D.6.
Professor frå latin «pro»: for, fram; «fiteri, fess»: vedgå, seie, kunngjere. Altså: forkynnar, ein
som kunngjer/underviser.
Rektor frå latin «regere»: råde, styre.
Senior (-ingeniør, -forskar, mfl.) frå latin: eldre, gammal, av «senex»: gammal (mann).
Sivilingeniør frå latin «civilis»: borgarleg, «civis»: borgar. Merk at tilsvarande ord kan ha andre
tydingar på andre språk, t.d. engelsk «civil engineer»: bygningsingeniør, eller ordet er ikkje nytta.
Sjef (-ingeniør, forskings-, mfl.) førar, leiar, overhovud, styrar gjennom fransk «chef, chief», frå
latin «caput»: hovud.
D.8 Engelsk – norsk ordliste; forkortingar 269
Student frå latin «studere»: streve etter, vere ihuga for, ihuga og uthaldande strev mot noko.
D.8 Engelsk – norsk ordliste; forkortingar

Advection — adveksjon (konveksjon, jf. avs- Flammability limit (upper/lower) — flam-
nitt D.1) megrense (øvre/nedre)
Atomization — forstøving Flare — fakkel (jf. avsnitt D.2.3)
Atomizer — forstøvar Flow — strøyming
Boundary condition — grensevilkår Fluidized bed — svevebed(d), kvervelsjikt, flu-
idisert bed(d) (sjå avsnitt D.3.4)
Boundary layer — grensesjikt (også: grenselag)
Flux — fluks (gjennomstrøyming)
Boundary value — grenseverdi
Forced convection — tvungen konveksjon (jf.
Burner — brennar avsnitt D.2.1)
Calorific value (net –) — effektiv nedre bren- Free convection — fri konveksjon, naturleg kon-
nverdi, dvs. frårekna fordampingsvarme til vatn veksjon (jf. avsnitt D.2.1)
i brenslet.
Furnace — fyrkjel (i visse tilfelle: omn; jf. avs-
Channel — kanal nitt D.2.3)
Combustion — forbrenning Gauge (gage) pressure — manometertrykk, målt
Combustion chamber — brennkammer trykk, overtrykk
Combustor — brennkammer, brennar Grid — nettverk, nett (numerisk); gitter
Conflagration — stor og voldsom brann Gust — gust (jf. avsnitt D.2.1)
Density — tettleik (bokm: tetthet) Heating value (lower/higher) — brennverdi (ne-

dre/øvre)
Dilatation (volumetric) — Volumtøyingsfart (jf.
avsnitt D.2.2) Ideal gas — ideell gass, idealgass
Duct — leidning, kanal Incinerator — forbrenningsomn; ev. spesifisere

t.d. bosbrenningsomn, krematorieomn (jf. avs-
Drag — slepekraft nitt D.2.3).
Efficiency — verknadsgrad Indeks (sub-; super-) — indeks (låg –; høg –)
Enthalpy of formation — danningsentalpi Inertia — tregleik, inertia
Extinguish — sløkkje (jf. avsnitt D.2.4) Jet — jet (også: stråle)
Equilibrium — jamvekt, likevekt Laminar — laminær
Fire — brann Large-eddy simulation — storevje-simulering
Flame — flamme (avsn. D.3.5)
Magnitude — storleik, mål Strain rate, rate of strain — tøyingsfart (jf. avs-
nitt D.2.2)
Magnitude; order of – — storleiksorden
Strain velocity — tøyingsfart
Mean free path — middelfriveg
Stratification — lagdeling, sjikting
Mesh (grid mesh size) — (i nettverk) maskevidd;
(i gitter) gitteropning Stress (shear –; normal –) — spenning (skjer-;
normal-)
Mixing layer — blandingssjikt
Subscript — lågt merketeikn
Mixing length — blandingsveg
Subindex — låg indeks
Mixture fraction — blandingsfraksjon
Superscript — høgt merketeikn
Moment — moment
Superindex — høg indeks
Momentum — rørslemengd (bokm: bevegelses-
mengde) Swirl — svervel, drall (jf. avsnitt D.1, D.2.1)
Natural convection — naturleg konveksjon, fri Viscosity — viskositet
konveksjon (jf. avsnitt D.2.1)
Vortex — kvervel, evje
Non-premixed flame — uforblanda flamme
Vortex street — kvervelgate (von Kármán)
Oven – omn (til oppvarming, baking; jf. avs-
nitt D.2.3) Wake — vake, kjølvatn
Perfect gas — (oftast:) ideell gass, idealgass. Wall jet — veggjet

(Det finst lærebøker der det med «perfect gas» er
meint ideell gass som også har konstant spesifikk Vanlege forkortingar
varmekapasitet.)
ASM, ASEM: Algebraic (Reynolds) stress-
Pipe — røyr, leidning equation models — modell med algebraiske
Plume — røykstrime, røyksky, dampsky, varm likningar for reynoldsspenningane
gass som stig til vers (jf. avsnitt D.1) BVM: Boussinesq-viscosity model — modell
Premixed flame — forblanda flamme med Boussinesq-viskositet (turbulensviskositet)
Probability density function — sannsynstettleik DNS: Direct numerical simulation — direkte nu-
merisk simulering, direktesimulering
Radiation — stråling (elektromagnetisk)
DSM, DSEM: Differential (Reynolds) stress-
Recirculation zone — bakevje, resirkulasjon- equation model — modell med (differen-
ssone sial)likningar for reynoldsspenningane
Shear strain — skjertøying EDC: Eddy Dissipation Concept — Magnussens
Shear strain, rate of – — skjertøyingsfart forbrenningsmodell
Spray — sprut, spray EVM: Eddy-viscosity model — modell med tur-

bulensviskositet
Stochiometric — støkiometrisk
LDA: Laser-Doppler anemometry — laser-
Strain — tøying (jf. avsnitt D.2.2) doppler-anemometri
D.8 Engelsk – norsk ordliste; forkortingar 271
LES: Large-eddy simulation — storevje- stettleik(sfunksjon)

simulering (avsn. D.3.5)
RANS: Reynolds-averaged Navier-Stokes —
ODE: Ordinary differential equation — ordinær Reynolds-midla Navier-Stokes, Reynolds-midla
differensiallikning rørslelikningar, Reynoldslikningane
PDE: Partial differential equation — partiell dif- RSE, RSEM: Reynolds-stress-equation model
ferensiallikning — modell med likningar for reynoldsspen-
ningane, reynoldsspenningslikningsmodell
PDF: Probability-density function — sannsyn-
List of litterature
Anderson, Dale A., John C. Tannehill and Richard H. Pletcher 1984 Computational fluid mechanics
and heat transfer. McGraw-Hill, New York.
Annamalai, Kalyan and William Ryan 1992 Interactive processes in gasification and combustion
Part I: Liquid drop arrays and clouds. Prog. Energy Comb. Sci. 18: 221-295.
Annamalai, Kalyan and William Ryan 1993 Interactive processes in gasification and combustion
Part II: Isolated carbon, coal and porous char particles. Prog. Energy Comb. Sci. 19: 383-446.
Annamalai, Kalyan, William Ryan and Senthilvelan Dhanapalan 1994 Interactive processes in
gasification and combustion Part III: Coal/char particle arrays, streams and clouds. Prog. Energy
Comb. Sci. 20: 487-618.
Aris, R. 1962 Vectors, tensors, and the basic equations of fluid mechanics. Prentice Hall, Engle-
wood Cliffs, New Jersey. (Reprint: Dover, New York 1989.)
Aris, R. 1978 Mathematical modelling techniques. Pitman, London.
Ayyaswamy, Portonovo S. 1995 Direct-contact transfer processes with moving liquid droplets.
Advances in Heat Transfer 26: 1-104.
Bachalo, W.D. 1994 Injection, dispersion, and combustion of liquid fuels. Twenty-Fifth Symp. (Int.)
Comb.: 333-344. Comb. Inst., Pittsburgh, Pennsylvania.
Beér, J.M. and N.A. Chigier 1972 Combustion Aerodynamics. Applied Science Publishers, London.
Beér, J.M., J. Chomiak and L.D. Smoot 1984 Fluid dynamics of coal combustion: A review. Prog.
Energy Comb. Sci. 10: 177-208.
Bejan, Adrian 1993 Heat Transfer. John Wiley & Sons, New York.
Belsvik, Rune 1991 Dustefjerten. Cappelen, Oslo.
Besnard, D.C. and F.H. Harlow 1988 Turbulence in multiphase flow. Int. J. Multiphase Flow 14:
679-699.
Bilger, R.W. 1980 Turbulent flows with nonpremixed reactants. In: Libby, P.A., and F.A. Williams
(eds.): Turbulent reacting flows: 65-113. Springer, Berlin.
Bilger, R.W. 1988 The structure of turbulent nonpremixed flames. Twenty-second Symp. (Int.)
274 List of literature
Bird, R. Byron, Warren E. Stewart and Edwin N. Lightfoot 1960 Transport phenomena. John Wiley
& Sons, New York.
Bockhorn, Henning (ed.) 1994 Soot formation in combustion. Springer, Berlin.
Borghi, Roland 1974 Chemical reactions calculations in turbulent flows: Application to a CO-
containing turbojet plume. Advances in Geophysics 18B: 349-365.
Borghi, R. 1988 Turbulent combustion modelling. Prog. Energy Comb. Sci. 14: 245-292.
Boussinesq, J. 1877 Essai sur la théorie des eaux courantes. Mémoires a l’Académie des Sciences
de l’Institut National de France 23: 1-680.
Bradley, Derek 1992 How fast can we burn? Twenty-fourth Symp. (Int.) Comb.: 247-262. Comb.
Inst., Pittsburgh, Pennsylvania.
Bradshaw, P. 1971 An introduction to turbulence and its measurement. Pergamon Press, London.
Bradshaw, Peter and George P. Huang 1995 The law of the wall in turbulent flow. Proc. Roy. Soc.
London A451: 165-188.
Bray, K.N.C. 1980 Turbulent flows with premixed reactants. In: Libby, P.A., and F.A. Williams
(eds.): Turbulent reacting flows: 115-183. Springer, Berlin.
Byggstøyl, S. and W. Kollmann 1981 Closure model for intermittent turbulent flows. Int. J. Heat
Mass Transfer 24: 1811-1822.
Byggstøyl, S. and W. Kollmann 1986a Stress transport in the rotational and irrotational zones of
turbulent shear flows. Phys. Fluids 29: 1423-1429.
Byggstøyl, S. and W. Kollmann 1986b A closure model for conditioned stress equations and its
application to turbulent shear flows. Phys. Fluids 29: 1430-1440.
Byggstøyl, S. and B.F. Magnussen 1983 A model for flame extinction in turbulent flow. In: L.J.S.
Bradbury mfl. (ed.): Turbulent Shear Flows 4: 381-395. Springer, Berlin 1985.
Calcote, Haetwell F. and Robert J. Gill 1994 Comparison of the ionic mechanism of soot forma-
tion with a free radical mechanism. In: Bockhorn, Henning (ed.): Soot formation in combustion.
Springer, Berlin.
Catchpole, John P. and George Fulford 1966 Dimensionless groups. Ind. Eng. Chem. 58: 46-60.
Catchpole, John and George Fulford 1987 Dimensionless groups. Encyclopedia of science &
technology 5: 284-295, 6th ed. McGraw-Hill, New York.
Cebeci, Tuncer and Peter Bradshaw 1984 Physical and computational aspects of convective heat
transfer. Springer, New York.
Chomiak, Jerzy 1990 Combustion – A study in theory, fact and application. Abacus Press/Gordon
and Breach, New York.
Chorin, Alexandre J. (1995) On turbulence modeling. Unpublished manuscript.
List of literature 275
Chou, P.Y. 1940 On an extension of Reynolds’ method of finding apparent stress and the nature of
turbulence. Chin. J. Phys. 4: 1-33.
Chou, P.Y. 1945 On velocity correlations and the solutions of the equations of turbulent fluctuation.
Quart. of Appl. Math. 3: 38-54.
Coles, D.E. and E.A. Hirst 1968 Memorandum on data selection. In: Coles, D.E. and E.A. Hirst
(ed.): Proceedings – Computation of turbulent boundary layers – 1968 AFOSR-IFP-Stanford Con-
ference: 47-54. Stanford University, California.
Concise Oxford dictionary 1982 7th ed, (ed.: J.B. Sykes), Oxford Univ. Press, Oxford, England.
Correa, S.M. and W. Shyy 1987 Computational models and methods for continuous gaseous turbu-
lent combustion. Prog. Energy Comb. Sci. 13: 249-292.
Corrsin, S. 1961 Turbulent flow. American Scientist 49: 300-325.
Corrsin, S. 1962 Turbulent dissipation fluctuations. Phys. Fluids 5: 1301-1302.
Courant, R., K. Friedrichs and H. Lewy 1928 Über die partiellen Differenzengleichungen der
mathematischen Physik. Math. Annal. 100: 32-74.
Cox, Geoffrey (ed.) 1995 Combustion fundamentals of fire. Academic Press, London.
Daly, Bart J. and Francis H. Harlow 1970 Transport equations in turbulence. Phys. Fluids 13:
2634-2649.
Dictionary of Scientific Biography 1970–1980 (ed. C.C. Gillispie), Charles Scribner’s Sons, New
York.
Doebelin, Ernest O. 1990 Measurement systems : Application and design. 4th ed. McGraw-Hill,
New York.
Drysdale, Dougal 1985 An Introduction to fire dynamics. John Wiley & Sons, Chichester, England.
Eckbreth, A.C. 1988 Laser diagnostics for combustion temperature and species. Abacus Press,
Kent, England.
Ertesvåg, Ivar Ståle 1991 Utvikling av turbulensmodell for låge reynoldstal med likning for reynoldsspen-
ningane og likning for karakteristisk frekvens Dr.ing. thesis 1991:49. Institutt for teknisk varmelære,
NTH, Trondheim.
Ertesvåg, Ivar S. and Bjørn F. Magnussen 1997 The Eddy-Dissipation turbulence energy cascade
model. Report STF84 A97501, SINTEF Energy Department of Applied Thermodynamics and Fluid
Dynamics, Trondheim.
Eskeland, Ivar 1981 Språket i forfall. Fram daa, frendar 8: 93-109. Norsk Måldyrkingslag, Oslo.
Favre, A. 1965 Équations des gaz turbulents compressibles. J. Mécanique 4: 361-421.
Engelsk omsetjing: Statistical equations of turbulent gases, Block, J.E. (ed.): Problems of hydro-
dynamics and continuum mechanics : contributions in honor of the sixtieth birthday of L. I. Sedov,
14th November 1967., 231-266. Society for Industrial and Applied Mathematics, Philadelphia 1969.
Ferziger, J.H. 1980 Homogeneous turbulent flow. In: Kline, S.J., B.J. Cantwell and G.M. Lilley
(eds.): The 1980-81 AFOSR-HTTM-Stanford Conference on Complex turbulent flows: Comparison
of computation and experiment I: 405-433. Stanford, California 1981.
Ferziger, Joel H. and Milovan Perić 1996 Computational methods for fluid dynamics. Springer,
Berlin.
Fitzroy, R. 1863 The weather book: A manual of practical meteorology, 2nd ed. Longman, Green,
Longman, Roberts, & Green, London.
Furukawa, Junichi, Kyoko Okamoto and Toshisuke Hirano 1996 Turbulence characteristics within
the local reaction zone of a high intensity turbulent premixed flame. Twenty-sixth Symp. (Int.)
Comb.:, 405-412. Comb. Inst., Pittsburgh, Pennsylvania, 1997.
Gibson, Carl H. 1968a Fine structure of scalar fields mixed by turbulence. I. Zero-gradient points
and minimal gradient surfaces. Phys. Fluids 11: 2305-2315.
Gibson, Carl H. 1968b Fine structure of scalar fields mixed by turbulence. II. Spectral theory. Phys.
Fluids 11: 2316-2327.
Gibson, M.M. and B.E. Launder 1978 Ground effects on pressure fluctuations in the atmospheric
boundary layer. J. Fluid Mech. 86: 491-511.
Gibson, M.M. and B.A. Younis 1986 Calculation of swirling jets with a Reynolds stress closure.
Phys. Fluids 29: 38-48.
Glassman, Irvin 1996 Combustion, 3rd ed. Academic Press, San Diego, USA.
Goulard, R., A.M. Mellor and R.W. Bilger 1976 Combustion measurements in air breathing propul-
sion engines. Survey and research needs. Comb. Sci. Technol. 14: 195-219.
Gran, Inge Røinaas 1994 Mathematical modeling and numerical simulation of chemical kinetics
in turbulent combustion. Dr.ing. thesis 1994:49, Institutt for mekanikk, termo- og fluiddynamikk,
NTH, Trondheim.
Gran, Inge R. and Bjørn F. Magnussen 1996a A numerical study of a bluff-body stabilized diffusion
flame. Part 1. Influence of turbulence modeling and boundary conditions. Comb. Sci. Technol. 119:
171-190.
Gran, Inge R. and Bjørn F. Magnussen 1996b A numerical study of a bluff-body stabilized diffusion
flame. Part 2. Influence of combustion modeling and finite-rate chemistry. Comb. Sci. Technol. 119:
191-217.
Gran, Inge R., Tarek Echekki and Jacqueline H. Chen 1996 Negative flame speed in an unsteady
2-D premixed flame: A computational study. Twenty-sixth Symp. (Int.) Comb.:, 323-329. Comb.
Inst., Pittsburgh, Pennsylvania, 1997.
Gran, Inge R., Ivar S. Ertesvåg and Bjørn F. Magnussen 1997 Influence of turbulence modeling on
predictions of turbulent combustion. AIAA J. 35: 106-110.
Også: 10th Symp. Turb. Shear Flows: 19.1-19.6, Pennsylvania State University, University Park,
Pennsylvania, aug. 1995.
Grimsmo, Bård 1991 Numerical simulation of turbulent flow and combustion in a four stroke
homogeneous charge internal combustion engine Dr.ing. thesis 1991:83. Institutt for teknisk
varmelære, NTH, Trondheim.
Grunnmanuskriptet (without year) Unpublished manuscript for a large Norwegian dictionary. Ma-
terial from the collections after Aasen, Ross, Schjøtt, Vidsteen, Torp etc. Made available by Institutt
for nordistikk og litteraturvitenskap, Universitety of Oslo, at the web (“WWW” at internet).
Hagen, G. 1839 Ueber die Bewegung des Wassers in engen cylindrischen Röhren. Poggendorff’s
Annalen der Physik und Chemie 46: 423-442. (Også i: Ostwald’s Klassiker der exakten Wis-
senschaften, Nr. 237, Leipzig 1933.)
Han, Zhiyu and Rolf D. Reitz 1997 A temperature wall function formulation for variable-density
turbulent flows with application to engine convective heat transfer modeling. Int. J. Heat Mass
Transfer 40: 613-625.
Hanjalić, K. 1994 Advanced turbulence closure models: a view of current status and future
prospects. Int. J. Heat and Fluid Flow 15: 178-203.
Hanjalić, K. and B.E. Launder 1972 A Reynolds stress model of turbulence and its application to
thin shear flows. J. Fluid Mech. 52: 609-638.
Hanjalić, K. and B.E. Launder 1976 Contribution towards a Reynolds-stress closure for low-
Reynolds-number turbulence. J. Fluid Mech. 74: 593-610.
Härtel, Carlos 1996 Turbulent flows: direct numerical simulation and large-eddy simulation. In:
Roger Peyret (ed.): Handbook of computational fluid mechanics: 283-338. Academic Press, Lon-
don.
Hinze, J.O. 1975 Turbulence, 2nd ed. McGraw-Hill, New York.
Hottel, H.C. and W.R. Hawthorne 1949 Diffusion in laminar flame jets. Third Symposium on Com-
bustion and Flame and Explosion Phenomena: 254-266. Standing Committee on Comb. Symp.,
The Williams & Wilkins Company, Baltimore, Maryland.
Hunt, J.C.R 1998 Lewis Fry Richardson and his contributions to mathematics, meteorology, and
models of conflict. Adv. Appl. Mech. 30: xiii-xxxvi.
Incropera, Frank P. and David P. DeWitt 1990 Fundamentals of heat and mass transfer. John Wiley
& Sons, New York.
Irgens, Fridtjov 1982 Tensoranalyse. Tapir, Trondheim.
Jones, W.P. 1994 Turbulence modelling and numerical solution methods for variable density and
combusting flows. In: Libby, P.A., and F.A. Williams (eds.): Turbulent reacting flows: 309-374.
Academic Press, London.
Jones, W.P. and B.E. Launder 1972 The prediction of laminarization with a two-equation model of
turbulence. Int. J. Heat Mass Transfer 15: 301-314.
Jones, W.P. and J.H. Whitelaw 1982 Calculation methods for reacting turbulent flows: A Review.
Comb. Flame 48: 1-26.
Kader, B.A. 1991 Heat and mass transfer in pressure-gradient boundary layers. Int. J. Heat Mass
Transfer 34: 2837-2857.
Kays, W.M. and M.E. Crawford 1993 Convective heat and mass transfer, 3rd ed. McGraw-Hill,
New York.
Kennedy, Ian M. 1997 Models of soot formation and oxidation. Prog. Energy Comb. Sci. 23:
95-132.
Kerstein, Alan R. 1991 Linear-eddy modelling of turbulent transport. Part 6. Microstructure of
diffusive scalar mixing fields. J. Fluid Mech. 231: 361-394.
Kim, John, Parviz Moin and Robert Moser 1987 Turbulence statistics in fully developed channel
flow at low Reynolds number. J. Fluid Mech. 177: 133-166.
Klebanoff, P.S. 1955 Characteristics of turbulence in a boundary layer with zero pressure gradient.
Report 1247, NACA. (19 p.)
Kollmann, W. 1992 Pdf transport modelling. Modelling of combustion and turbulence. Lecture
Series 1992-03, von Karman Institute for Fluid Dynamics, Belgium.
Kolmogorov, A. 1941a The local structure of turbulence in incompressible viscous fluid for very
large Reynolds’ numbers. (Translation:) C. R. (Doklady) Acad. Sci. URSS 30: 301-305.
Kolmogorov, A. 1941b Dissipation of energy in the locally isotropic turbulence. (Translation:) C.
R. (Doklady) Acad. Sci. URSS 32: 16-18.
Kolmogorov, A.N. 1942 Equations of turbulent motion of an incompressible fluid. Translation by
D.B. Spalding, Report ON/6 Imperial College, London 1968. From: Izvestia Akademia Nauk SSSR,
Seria fizicheska VI 1942: 56-58.
Komori, Satoru, J.C.R Hunt, Takao Kanzaki and Y. Murakami 1991 The effects of turbulent mixing
on the correlation between two species and on concentration fluctuations in non-premixed reacting
flows. J. Fluid Mech. 228: 629-659.
Kreplin, Hans-Peter and Helmut Eckelmann 1979 Behavior of the three fluctuating velocity com-
ponents in the wall region of a turbulent channel flow. Phys. Fluids 22: 1233-1239.
Kuo, Kenneth Kuan-yun 1986 Principles of combustion. John Wiley & Sons, New York.
Lai, Yong G. 1996 Predictive capabilities of turbulence models for a confined swirling flow. AIAA
J. 34: 1743-1745.
Landahl, M.T. and E. Mollo-Christensen 1986 Turbulence and random processes in fluid mechan-
ics. Cambridge Univ. Press, Cambridge, Massachusetts.
Launder, B.E. 1984 Numerical computation of convective heat transfer in complex turbulent flows:
time to abandon wall functions? Int. J. Heat Mass Transfer 27: 1485-1491.
Launder, B.E. 1988 On the computation of convective heat transfer in complex turbulent flows.
ASME J. Heat Transfer 110: 1112-1128.
Launder, B.E. 1989a Second-moment closure and its use in modelling turbulent industrial flows.
Int. J. Numer. Meth. Fluids 9: 963-985.
Launder, Brian E. 1989b Second-moment closure: present... and future?. Int. J. Heat and Fluid
Flow 10: 282-300.
Launder, B.E., G.J. Reece and W. Rodi 1975 Progress in the development of a Reynolds-stress
turbulence closure. J. Fluid Mech. 68: 537-566.
Launder, B.E. and B.I. Sharma 1974 Application of the energy-dissipation model of turbulence to
the calculation of flow near a spinning disc. Letters in heat and mass transfer 1: 131-138.
Launder, B.E. and D.B. Spalding 1972 Lectures in mathematical models of turbulence. Academic
Press, London.
Launder, B.E. and D.B. Spalding 1974 The numerical computation of turbulent flows. Computer
Methods in Appl. Mech. and Eng. 3: 269-289.
Lesieur, Marcel 1990 Turbulence in Fluids, 2nd rev. ed. Kluver, Dordrecht, Netherlands.
Libby, Paul A. and Forman A. Williams (eds.) 1980 Turbulent reacting flows. Springer, Berlin.
Libby, Paul A. and Forman A. Williams (eds.) 1994 Turbulent reacting flows. Academic Press,
London.
Lockwood, F.C. and N.G. Shah 1981 A new radiation method for incorporation in general combus-
tion prediction procedures. Eighteenth Symp. (Int.) Comb.: 1405-1413. Comb. Inst., Pittsburgh,
Pennsylvania.
Lumley, J.L. 1978 Computational modeling of turbulent flows. Adv. Appl. Mech. 18: 123-176.
Lumley, J.L. 1995 Introduction. In: Chou, Pei-Yuan and Ru-Ling Chou: 50 years of turbulence
research in China. Adv. Appl. Mech. 27: 1-15.
Magnussen, B.F. 1971 The rate of combustion of soot in turbulent flames. Thirteenth Symp. (Int.)
Magnussen, B.F. 1973 Prediction of characteristics of enclosed turbulent jet flames. Fourteenth
Symp. (Int.) Comb.: 553-556. Comb. Inst., Pittsburgh, Pennsylvania.
Magnussen, B.F. 1975a An investigation into the behavior of soot in a turbulent C2 H2 -flame.
Fifteenth Symp. (Int.) Comb.: 1415-1425. Comb. Inst., Pittsburgh, Pennsylvania.
Magnussen, Bjørn F. 1975b Some features of the structure of a mathematical model of turbulence,
NTH-ITV, Aug.1975. (16 p.)
Magnussen, Bjørn F. 1981 On the structure of turbulence and a generalized eddy dissipation concept
for chemical reaction in turbulent flow. 19th. AIAA aerospace science meeting. Jan. 12-15, St.Louis,
Missouri.
Magnussen, Bjørn F. 1989 Modeling of NOx and soot formation by the Eddy Dissipation Concept.
Int. Flame Research Foundation, 1st Topic Oriented Technical Meeting. 17-19 Oct., Amsterdam,
Holland.
Magnussen, B.F. and B.H. Hjertager 1976 On mathematical modeling of turbulent combustion with
special emphasis on soot formation and combustion. Sixteenth Symp. (Int.) Comb.: 719-729. Comb.
Inst., Pittsburgh, Pennsylvania.
Magnussen, B.F., B.H. Hjertager, J.G. Olsen and D. Bhaduri 1978 Effects of turbulent structure
and local concentrations on soot formation and combustion in C2 H2 diffusion flames. Seventeenth
Symp. (Int.) Comb.: 1383-1393. Comb. Inst., Pittsburgh, Pennsylvania 1979.
Mansour, N.N., J. Kim and P. Moin 1988 Reynolds-stress and dissipation-rate budgets in a turbulent
channel flow. J. Fluid Mech. 194: 15-44.
McComb, W.D. 1990 The physics of fluid turbulence, Clarendon Press, Oxford, England.
Mills, Anthony F. 1995 Heat and mass transfer. Irwin, Chicago.
Moin, Parviz and Krishnan Mahesh 1998 Direct numerical simulation: A tool in turbulence re-
search. Annu. Rev. Fluid. Mech. 30: 539-578.
Moran, Michael J. and Howard N. Shapiro 1993,1998 Fundamentals of engineering thermodynam-
ics, 2nd ed. (1993), 3rd ed. (1998). John Wiley & Sons, New York.
Naot, D., A. Shavit and M. Wolfshtein 1970 Interactions between components of the turbulent
velocity correlation tensor due to pressure fluctuations. Israel J. Tech. 8: 259-269.
Norsk Ordbok 1966f (preliminarily 3 volumes, A-G.) Det Norske Samlaget, Oslo.
Nynorskordboka 1986 Det Norske Samlaget, Oslo.
Olaus Magnus 1539 Carta marina et descriptio septemtrionalium terrarum ac mirabilium in eis
contentarum diligentissime elaborata rerum. Venezia. (Faksimile: Lærdomshistoriska samfundet,
Uppsala 1949).
Olaus Magnus 1555 Historia om de nordiska folken, (på latin) Roma. (Svensk utgåve: Gidlunds,
Stockholm 1982)
Onsager, L. 1949 Statistical hydrodynamics. Nuovo Cimento, Suppl. 6: 279-287.
Oxford English Dictionary 1933 Oxford at The Clarendon Press, England.
Oxford Latin Dictionary 1968 Oxford University Press, Oxford, England.
Patankar, Suhas V. 1980 Numerical heat transfer and fluid flow. Hemisphere, Washington.
Patankar, S.V. and D.B. Spalding 1967 Heat and mass transfer in boundary layers. Morgan-
Grampian, London.
Patel, Virendra C. , Wolfgang Rodi and Georg Scheuerer 1985 Turbulence models for near-wall
and low-Reynolds number flows: A review. AIAA J. 23: 1308-1319.
Peters, N. 1984 Laminar diffusion flamelet models in non-premixed turbulent combustion. Prog.
Energy Comb. Sci. 10: 319-339.
Peters, N. 1986 Laminar flamelet concepts in turbulent combustion. Twenty-first Symp. (Int.)
Peyret, Roger (ed.) 1996 Handbook of computational fluid mechanics. Academic Press, London.
Pope, S.B. 1985 Pdf methods for turbulent reactive flows. Prog. Energy Comb. Sci. 11: 119-192.
Prandtl, Ludwig 1904 Über Flüssigkeitsbewegung bei sehr kleiner Reibung. Verhandlungen des
III. Internationalen Mathematiker-Kongresses, Heidelberg: 484-491. Teubner, Leipzig 1905. (Also
in: F.W. Riegels (Schriftleiter): Ludwig Prandtl Gesammelte Abhandlungen 2: 575-584. Springer,
Berlin 1961.)
Prandtl, Ludwig 1925 Bericht über Untersuchungen zur ausgebildeten Turbulenz. Z. angew. Math.
Mech. 5: 136-139. (Also in: F.W. Riegels (Schriftleiter): Ludwig Prandtl Gesammelte Abhandlun-
gen 2: 714-718. Springer, Berlin 1961.)
Prandtl, Ludwig 1945 Über ein neues Formelsystem für die ausgebildete Turbulenz. Nachrichten
Akad. Wiss. Göttingen, Matem.-phys. Klasse 1945: 6-19. (Also in: F.W. Riegels (Schriftleiter):
Ludwig Prandtl Gesammelte Abhandlungen 2: 874-887. Springer, Berlin 1961.)
Rant, Z. 1956 Exergie, ein neues Wort für technische Arbeitsfähigkeit . Forschung 22: 36-37.
Reynolds, Osborne 1883 An experimental investigation of the circumstances which determine
whether the motion of water shall be direct or sinuos, and of the law of resistance in parallel channels.
Phil. Trans. Roy. Soc. London A174: 935-982.
Reynolds, Osborne 1895 On the dynamical theory of incompressible viscous fluids and the deter-
mination of the criterion. Phil. Trans. Roy. Soc. London A186: 123-164.
Reynolds, W.C. 1989 The potential and limitations of direct and large eddy simulations. In: J.L.
Lumley (ed.): Whither turbulence? Turbulence at the crossroads: 313-343. Springer, Berlin 1990.
Richardson, John M., Henry C. Howard jr. and Robert W. Smith jr. 1953 The relation between
sampling-tube measurements and concentration fluctuations in a turbulent gas jet. Fourth Symp.(Int.)
Comb.: 814-817. Standing Committee on Comb. Symp., Williams & Wilkins, Baltimore, Maryland.
Richardson, Lewis F. 1922 Weather prediction by numerical process. Re-publication. Dover Publ.,
New York 1965. (Cambridge Univ. Press, London 1922).
Rodi, W. 1976 A new algebraic relation for calculating the Reynolds stresses. Z. Angew. Math.
Mech. 56: 219-221.
Rodi, Wolfgang 1980 Turbulence models and their application in hydraulics. IAHR, Delft, Nether-
lands. ([unchanged reprint:] Balkema, Rotterdam 1993.)
Rodi, Wolfgang 1991 Examples of turbulence model applications. Introduction to the modeling of
turbulence. Lecture Series 1991-02, von Karman Institute for Fluid Dynamics, Belgium.
Rotta, J. 1951a Statistische Theorie nichthomogener Turbulenz – 1. Mitteilung. Z. Physik 129:
547-572.
Rotta, J. 1951b Statistische Theorie nichthomogener Turbulenz – 2. Mitteilung. Z. Physik 131:
51-77.
Rotta, J.C. 1962 Turbulent boundary layers in incompressible flow. Prog. Aeron. Sci. 2: 1-219.
Rouse, Hunter and Simon Ince 1957 History of hydraulics. Dover Publications, New York.
Råd om språk 1983 fra Norsk språkråd. J.W. Cappelens Forlag, Oslo.
Schlichting, Hermann 1979 Boundary-Layer Theory, 7th ed. McGraw-Hill, New York.
Schumann, U. 1977 Realizability of Reynolds-stress turbulence models. Phys. Fluids 20: 721-725.
Sloan, D.G. and G.J. Sturgess 1996 Modeling of local extinction in turbulent flames. ASME J. Eng.
Gas Turbines Power 118: 292-307.
Smith, L.M. and W.C. Reynolds 1992 On the Yakhot-Orszag renormalization group method for
deriving turbulence statistics and models. Phys. Fluids A 2: 364.
Song, Ge 1996 Radiation heat transfer in a high temperature participating media. Dr.ing. thesis
1996:6, Institutt for mekanikk, termo- og fluiddynamikk, NTNU, Trondheim.
Sokolnikoff, I.S. 1964 Tensor analysis – Theory and applications to geometry and mechanics of
continua. John Wiley & Sons, New York.
Sonntag, Richard E. and Gordon J. Van Wylen 1982,1991 Introduction to thermodynamics – Clas-
sical and statistical, 2nd ed. (1982), 3rd ed. (1991). John Wiley & Sons, New York.
Spalart, Philippe R. 1988 Direct simulation of a turbulent boundary layer up to Rθ = 1410. J. Fluid
Mech. 187: 61-98.
Spalart, P.R. and S.R. Allmaras 1994 A one-equation turbulence model for aerodynamic flows.
Recherche Aerospatiale 1: 5-21.
Spalding, D. Brian 1976 Development of the Eddy-Break-up model of turbulent combustion. Six-
teenth Symp.(Int.) Comb.: 1657-1663. Comb. Inst., Pittsburgh, Pennsylvania.
Speziale, Charles G. 1998 Turbulence modeling for time-dependent RANS and VLES: A review.
AIAA J. 36: 173-184.
Sreenivasan, Katepalli R. 1995 On the universality of the Kolmogorov constant. Phys. Fluids 7:
2778-2784.
Steinnes, Asgaut and Eirik Vandvik 1958 Latinsk ordbok. Det Norske Samlaget, Oslo.
Tavoularis, Stavros and Stanley Corrsin 1981 Experiments in nearly homogeneous turbulent shear
flow with a uniform mean temperature gradient. Part 1. J. Fluid Mech. 104: 311-347.
Taylor, A.M.K.P. (ed.) 1993 Instrumentation for flows with combustion. Academic Press, London.
Taylor, G.I. 1927 Turbulence. Quart. J. Roy. Meteorol. Soc. 53: 201-211.
Tennekes, H. 1968 Simple model for the small-scale structure of turbulence. Phys. Fluids 11:
669-671.
Tennekes, H. and J.L.Lumley 1972 A first course in turbulence. 9th printing, MIT Press, Cam-
bridge, Massachusetts 1983.
Townsend, A.A. 1976 The structure of turbulent shear flow. 2nd ed. (1st ed. 1956). Cambridge
University Press, Cambridge, England.
Tritton, D.J. 1988 Physical fluid dynamics, 2nd ed. (1st ed. 1977). Clarendon Press, Oxford,
England.
Truesdell, C. 1952 The mechanical foundations of elasticity and fluid dynamics. J. Rat. Mech.
Anal. 1: 125-300.
Truesdell, Clifford Ambrose 1980 The tragicomical history of thermodynamics, 1822–1854. Springer,
New York.
Tyldesley, John R. 1975 An introduction to tensor analysis for engineers and applied scientists.
Longman, London.
Versteeg, H.K. and W. Malalasekera 1995 An introduction to computational fluid dynamics – The
finite volume method. Longman, London.
Vinje, Aasmund Olavson 1867 Om stil. Dølen 5.
Warnatz, J. 1983 The mechanism of high temperature combustion of propane and butane. Comb.
Sci. Technol. 34: 177.
Warnatz, J. 1990 NOx formation in high-temperature processes. In: Eurogas ’90: 303-320, Tapir,
Trondheim.
Warnatz, Jürgen and Ulrich Maas 1993 Technische Verbrennung. Springer, Berlin.
Warnatz, Jürgen, Ulrich Maas and Robert W. Dibble 1996,1999 Combustion – Physical and chem-
ical fundamentals, modeling and simulation, experiments, pollutant formation, 1st ed. (1996), 2nd
ed. (1999). Springer, Berlin.
Weinberg, Felix J. 1975 The first half-million years of combustion research and today’s burning
problems. Fifteenth Symp. (Int.) Comb.: 1-17. Comb. Inst., Pittsburgh, Pennsylvania.
White, Frank M. 1986,1994 Fluid mechanics, 2nd ed. (1986), 3rd ed. (1994). McGraw Hill, New
York.
Wicke, Ewald 1984 Gerhard Damköhler. In: VDI-Gesellschaft Verfahrenstechnik und Chemiein-
genieurwesen gestern – heute – morgen. Jubiläumsschrift, Zusammengestellt von Eckhart Blaß,
VDI-Gesellshaft/K.G.Saur, Düsseldorf.
Wilcox, David C. 1988 Reassessment of the scale-determining equation for advanced turbulence
models. AIAA J. 26: 1299-1310.
Wilcox, David C. 1993 Turbulence modeling for CFD. DCW Industries, La Cañada, California.
Wilcox, David C. and Morris W. Rubesin 1980 Progress in turbulence modeling for complex flow
fields including effects of compressibility. NASA Tech. Paper 1517.
Wilcox, D.C. and R.M. Traci 1976 A complete model of turbulence. AIAA 9th Fluid and Plasma
Dynamics Conference. July 14-16. San Diego, California. AIAA-Paper 76-351. (29 p.)
Williams, Forman A. 1985 Combustion theory, 2nd ed. Benjamin/Cummings, Menlo Park, Cali-
fornia.
Williams, F.A. 1992 The role of theory in combustion science. Twenty-fourth Symp. (Int.) Comb.:
1-17. Comb. Inst., Pittsburgh, Pennsylvania.
Wyngaard, J. 1989 The potential and limitations of direct and large eddy simulations. Comment
3 to W. C. Reynolds. In: Lumley, J.L. (ed.): Wither Turbulence? Turbulence at the crossroads:
369-373. Springer, Berlin 1990.
Yakhot, Viktor and Steven A. Orszag 1986 Renormalization group analysis of turbulence. I. Basic
Theory. J. Sci. Comput. 1: 3-51.
Yakhot, Viktor and Leslie M. Smith 1992 The renormalization group, the ε-expansion and deriva-
tion of turbulence models. J. Sci. Comput. 7: 35-61.
Zhou Peiyuan (Chou Pei-Yuan) and Chen Shiyi 1987 On the theory of turbulence for incompressible
fluids. Scientia Sinica A 30: 725-738.
Zhou, Ye 1993 Interacting scales and energy transfer in isotropic turbulence. Phys. Fluids A 5:
2511-2524.
Aasen, Ivar 1876 Norsk Maalbunad. Norsk bokreidingslag, Bergen 1975.
Aasen, Ivar 1873 Norsk Ordbog. (Photographical reprint, Fonna Forlag, Oslo 1977).

Ertesvag - Turbulent Flow and Combustion

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ertesvag - Turbulent Flow and Combustion

Uploaded by

Copyright:

Available Formats

Ivar S.

1 Introduction and basics 1

3 The k -εε turbulence model 45

4 Some simple flows 63

4.4 A logarithmic region in boundary layers – wall functions . . . . . . . . . . . . . . . 70

5 Energy and mass transfer (scalar transport) 79

6 Equations for the Reynolds stresses 93

7 Statistical functions in turbulence 107

8 Energy transfer in turbulent flow 121

9 Various topics in turbulence 135

10 Turbulent combustion 149

10.7 Prescribed probability density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

11 Magnussen’s combustion model, «Eddy Dissipation Concept» 171

Appendix: Equations that we love 195

A The fundamental equations – and something more 197

B Equations for turbulent flow 221

C Dimensionless parameters 237

D Nokre ord om ord 247

List of litterature 273

C1 , C2 constants in the redistribution model, Eq. (6.19)

f k,i acceleration of species k due to distant force in x i -direction [m/s2 ]

ReT = k 2 /νε; turbulence Reynolds number, Eq. (9.16)

α = λ/(C p ρ); thermal diffusivity, Eqs. (1.6), (A.87) [m2/s]

θ tangential coordinate in cylinder coordinates, Sec. B.8 [1]

∗ fine structure in EDC, Chapter 11

Introduction and basics

Purpose of the book:

Scientific foundations and relations

Relations and foundation:

Overview of the book

1.2 About reading this book

Many equations – little mathematics

1.3 This section you do not need to read

himself by eating marzipan.

Figure 1.2: Why do we put stones around the fireplace?

Combustion is a physio-chemical phenomenon. Improving the combustion requires knowledge of

Combustion as a physio-chemical phenomenon

(a) Schematic presentation (b) Experiments by Gaunce (1939, unpublished)

Models and methods for combustion

Figure 1.4: Sketches by Leonardo da Vinci.

Figure 1.5: Eddying motion in water. Drawings by Leonardo da Vinci.

1.6 Numerical tools

1.7 Basic equations

The stress tensor for a Newtonian fluid is expressed as

Compressibility and variable density

1.8 Chemical reactions

Turbulence and combustion

We will return to modeling of turbulent combustion in Chapters 10 and 11.

fuel + λ · amount of air → product + (λ − 1) · amount of air, (1.7)

CH4 + λ · (2O2 + 2 · 3.76N2)

8CH4 + 2O2 + 2 · 3.76N2

Many simultaneous reactions

′ and ν ′′ are stoichiometric coefficients for the species A in the reaction l.

Example: Equation (1.10) can best be explained by an example. We consider a stoichiometric

2H2 + O2 ⇄ 2H2 O. (1.11)

Chemical production term

Simple overall reaction

Why so much on turbulence in a book on combustion?

(f) Numerical model of a platform

Figure 2.1: Examples of models. Each model visualizes something;

Models, ideally and in practice1

• A model can be good without correct theory behind.

2.2 Reynolds-decomposition and averaging