Jeff Thesis

Direct numerical simulation of turbulent,
chemically reacting flows
A THESIS
SUBMITTED TO THE FACULTY OF THE GRADUATE SCHOOL
OF THE UNIVERSITY OF MINNESOTA
BY
Jeffrey Joseph Doom
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
Krishnan Mahesh, Advisor
January, 2009
c Jeffrey Joseph Doom 2009

Acknowledgments
The person I would like to thank first is my adviser Professor Krishnan Mahesh.
His experience, intelligence, and patience were instrumental. He provided a
great graduate experience for which that I am very grateful.
I am most indebted to my wife, Sarah for her endless support. I would like
to thank my Mother, Eileen Doom and Sarah’s Parents, Roger and Deb Haar
for their incredible patience and support. I would like to thank my children,
Kaitlyn and Jonathan. They bring so much joy and happiness to my life. Next,
I thank Dr. Ken Murphy. He was the one that encouraged me to study science
and mathematics. I would also like to thank the professors and students at the
Aerospace Engineering and Mechanics Department for providing an enjoyable

place to work. In particular, I would like to thank Mr. Michael Mattson,
Dr. Suman Muppidi and Dr. Yucheng Hou for their enjoyable conversations.
Finally, I thank my deceased father, Ronald Doom. My father taught me the
most important skills of all, good work ethic and patience.

This work was supported by the AFOSR under grant FA9550–04–1–0341
and by the Doctoral Dissertation fellowship awarded by the Graduate School
at the University of Minnesota. Computing resources were provided by the

Minnesota Supercomputing Institute, the San Diego Supercomputing Center
and the National Center for Supercomputing Applications.
i
To my wife, children, and parents
ii
Abstract
This dissertation: (i) develops a novel numerical method for DNS/LES of com-
pressible, turbulent reacting flows, (ii) performs several validation simulations,
(iii) studies auto–ignition of a hydrogen vortex ring in air and (iv) studies a
hydrogen/air turbulent diffusion flame.

The numerical method is spatially non-dissipative, implicit and applicable
over a range of Mach numbers. The compressible Navier–Stokes equations are

rescaled so that the zero Mach number equations are discretely recovered in
the limit of zero Mach number. The dependent variables are co–located in
space, and thermodynamic variables are staggered from velocity in time. The
algorithm discretely conserves kinetic energy in the incompressible, inviscid,

non–reacting limit. The chemical source terms are implicit in time to allow
for stiff chemical mechanisms. The algorithm is readily applicable to complex

chemical mechanisms. Good results are obtained for validation simulations.
The algorithm is used to study auto–ignition in laminar vortex rings. A

nine species, nineteen reaction mechanism for H2 /air combustion proposed
by Mueller et al. [37] is used. Diluted H2 at ambient temperature (300 K)
is injected into hot air. The simulations study the effect of fuel/air ratio,
oxidizer temperature, Lewis number and stroke ratio (ratio of piston stroke
length to diameter). Results show that auto–ignition occurs in fuel lean, high
iii
Abstract iv
temperature regions with low scalar dissipation at a ‘most reactive’ mixture
fraction, ζM R (Mastorakos et al. [32]). Subsequent evolution of the flame is

not predicted by ζM R ; a most reactive temperature TM R is defined and shown
to predict both the initial auto–ignition as well as subsequent evolution. For

stroke ratios less than the formation number, ignition in general occurs behind
the vortex ring and propagates into the core. At higher oxidizer temperatures,
ignition is almost instantaneous and occurs along the entire interface between
fuel and oxidizer. For stroke ratios greater than the formation number, ignition
initially occurs behind the leading vortex ring, then occurs along the length
of the trailing column and propagates towards the ring. Lewis number is
seen to affect both the initial ignition as well as subsequent flame evolution
significantly. Non–uniform Lewis number simulations provide faster ignition
and burnout time but a lower maximum temperature. The fuel rich reacting
vortex ring provides the highest maximum temperature and the higher oxidizer
temperature provides the fastest ignition time. The fuel lean reacting vortex
ring has little effect on the flow and behaves similar to a non–reacting vortex
ring.
We then study auto–ignition of turbulent H2 /air diffusion flames using the
Mueller et al. [37] mechanism. Isotropic turbulence is superimposed on an

unstrained diffusion flame where diluted H2 at ambient temperature interacts
with hot air. Both, unity and non–unity Lewis number are studied. The results
are contrasted to the homogeneous mixture problem and laminar diffusion
flames. Results show that auto–ignition occurs in fuel lean, low vorticity,
high temperature regions with low scalar dissipation around a most reactive
mixture fraction, ζM R (Mastorakos et al. [32]). However, unlike the laminar
flame where auto-ignition occurs at ζM R , the turbulent flame auto–ignites over

Abstract v
a very broad range of ζ around ζM R , which cannot completely predict the onset
of ignition. The simulations also study the effects of three-dimensionality. Past
two–dimensional simulations (Mastorakos et al. [32]) show that when flame

fronts collide, extinction occurs. However, our three dimensional results show
that when flame fronts collide; they can either increase in intensity, combine
without any appreciable change in intensity or extinguish. This behavior is
due to the three–dimensionality of the flow.

Table of contents
Acknowledgments i
Dedication ii
Abstract iii
Table of contents vi
List of tables ix
List of figures x
Nomenclature xiii
1 Introduction 1
1.1 Numerical method . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Auto–ignition of hydrogen vortex ring reacting with hot air . . 3
1.3 Auto–ignition in turbulent H2 /air diffusion flames . . . . . . . 6
1.4 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Numerical method 11
2.1 Governing equations . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.1 Implicit source term . . . . . . . . . . . . . . . . . . . 18
2.2.2 Zero Mach number limit . . . . . . . . . . . . . . . . . 23
3 Validation simulations 25
3.1 Simple chemistry . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.1 Laminar premixed flame . . . . . . . . . . . . . . . . . 26
vi
TABLE OF CONTENTS vii
3.1.2 Laminar diffusion flame . . . . . . . . . . . . . . . . . 28

3.1.3 Laminar two–dimensional jet flame . . . . . . . . . . . 30
3.1.4 Turbulent diffusion flame with one-step chemistry . . . 32
3.2 Complex chemistry . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.1 Well stirred reactor . . . . . . . . . . . . . . . . . . . . 36
3.2.3 Laminar jet flame . . . . . . . . . . . . . . . . . . . . . 37
3.2.4 Sandia lifted jet flame . . . . . . . . . . . . . . . . . . 39
4 Auto–ignition of a hydrogen vortex ring reacting with hot air 42

4.1 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2 Validation and grid–convergence study . . . . . . . . . . . . . 45
4.3 Homogeneous problem . . . . . . . . . . . . . . . . . . . . . . 48
4.4 Results for reacting vortex ring . . . . . . . . . . . . . . . . . 51
4.4.1 Ignition and flame evolution . . . . . . . . . . . . . . . 52
4.4.2 Evolution of heat release . . . . . . . . . . . . . . . . . 57
4.4.3 The importance of Lewis number . . . . . . . . . . . . 61
4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5 Turbulent H2 /air diffusion flames 67

5.1 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . 67
5.1.2 Turbulent diffusion flame . . . . . . . . . . . . . . . . . 70
5.1.3 Validation . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.2.1 Auto–ignition of diffusion flame . . . . . . . . . . . . . 70
5.2.2 The effect of Lewis number . . . . . . . . . . . . . . . 77
5.2.3 Effect of grid resolution . . . . . . . . . . . . . . . . . 80
5.2.4 Evolution of ignition fronts . . . . . . . . . . . . . . . . 80
5.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Appendix 84
A Non–Dimensionalizing the Navier–Stokes equations 84

A.1 Non–dimensionalizing the governing equations . . . . . . . . . 86
A.1.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
TABLE OF CONTENTS viii
A.1.2 Momentum . . . . . . . . . . . . . . . . . . . . . . . . 87
A.1.3 Species . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
A.1.4 Equation of state . . . . . . . . . . . . . . . . . . . . . 88
A.1.5 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
B Flamelet modeling 92
Bibliography 97
List of Tables
3.1 Parameters for laminar diffusion flame with one–step chemistry. 29

3.2 Parameters for turbulent diffusion flame with one–step chemistry. 34
3.3 Parameters for diffusion flame with one–step chemistry. . . . . 35
3.4 Parameters for laminar diffusion flame. . . . . . . . . . . . . . 37
3.5 Simulation parameters for laminar jet flame. . . . . . . . . . . 38
3.6 Simulations parameters for lifted jet flame. . . . . . . . . . . . 39
4.1 Inlet boundary conditions for reacting vortex ring. . . . . . . . 44

4.2 Mueller Mechanism. . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3 Global results for homogeneous mixture problem. . . . . . . . 51
4.4 Parameters used in reacting vortex ring simulations. . . . . . 51
4.5 Global behavior of reacting vortex ring. . . . . . . . . . . . . . 56
5.1 References variables for laminar H2 /air diffusion flame. . . . . 69
ix
List of Figures
2.1 Storage of variables. . . . . . . . . . . . . . . . . . . . . . . . 15

2.2 Variation of acoustic CFL and number of outer loop iterations
with reference Mach number. . . . . . . . . . . . . . . . . . . 22
3.1 Comparison of computed temperature to analytical solution for

a laminar premixed flame and spatial order of accuracy. . . . . 26
3.2 Comparison of asymptotic solution to numerical solution for a
laminar diffusion flame. . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Contour plot of temperature for a jet flame and center line pro-
file of mixture fraction. . . . . . . . . . . . . . . . . . . . . . . 31
3.4 Scalar dissipation rate and reaction rate for turbulent diffusion
flame. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.5 Iso–surface plot of the reaction rate for turbulent diffusion flame
and contour plot of temperature showing entire computational
domain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.6 Scatter plot of fuel mass fraction and oxidizer mass fraction. . 34
3.7 Comparison between Chemkin and present algorithm for a well–
stirred reactor. . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.8 Grid convergence study showing profiles of a major species and
a minor species. . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.9 Time versus maximum temperature and heat release for a lam-
inar diffusion jet flame. . . . . . . . . . . . . . . . . . . . . . . 39
3.10 Contours of a laminar diffusion jet flame. . . . . . . . . . . . . 40
3.11 Contour plots of a lifted jet flame. . . . . . . . . . . . . . . . . 41
4.1 Schematic of reacting vortex ring. . . . . . . . . . . . . . . . . 43
x
LIST OF FIGURES xi
4.2 Comparison using one–step chemistry to Hewett and Madnia

and a comparison of a major species & temperature between
Chemkin and present algorithm for a well–stirred reactor using
the Mueller mechanism. . . . . . . . . . . . . . . . . . . . . . 45
4.3 Grid convergence study for a two–dimensional laminar reacting
vortex ring. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.4 Profile of reaction rates for the species H versus time and tem-
poral evolution of temperature with all reactions, and reactions
1, 2, 3, 9, 10 & 11. . . . . . . . . . . . . . . . . . . . . . . . . 48
4.5 Mixture fraction versus τchem for varying temperature and mix-
ture fraction versus τchem for varying fuel/air ratio for the ho-
mogeneous mixture problem. . . . . . . . . . . . . . . . . . . . 49
4.6 Domain average of rates of reactions 1,2,3,9,10 and 11 for the
species H in our reacting vortex ring. . . . . . . . . . . . . . . 52
4.7 Evolution of auto–ignition for the reacting vortex ring. . . . . 53
4.8 Comparison between ζM R and TM R . . . . . . . . . . . . . . . . 54
4.9 Time evolution of heat release. . . . . . . . . . . . . . . . . . . 58
4.10 Contours of heat release and velocity vectors for non–uniform
and uniform Lewis number. . . . . . . . . . . . . . . . . . . . 59
4.11 Contours of heat release, temperature, and velocity vectors for
L/D = 2 and L/D = 6. . . . . . . . . . . . . . . . . . . . . . . 60
4.12 Time evolution of temperature. . . . . . . . . . . . . . . . . . 62
4.13 Maximum profiles of heat release, temperature and vorticity
magnitude versus time. . . . . . . . . . . . . . . . . . . . . . . 64
5.1 Schematic of initial conditions for the turbulent diffusion flame. 68

5.2 Comparison between Chemkin and present algorithm for a well–
stirred reactor. Also, grid convergence study of laminar diffu-
sion flame. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.3 Scatter plot of the laminar diffusion flame with unity and non–
unity Lewis number mixture fractions versus passive scalar. . . 71
5.4 Evolution of mixture fraction versus heat release and temporal
evolution of maximum heat release and temperature for the
laminar diffusion flame. . . . . . . . . . . . . . . . . . . . . . . 72
5.5 Evolution of temperature from a non–dimensional time of 0 to 4. 73
5.6 Evolution of YHO2 from a non–dimensional time of 0 to 4. . . . 74
LIST OF FIGURES xii
5.7 Evolution of YOH from a non–dimensional time of 0 to 4. . . . 74

5.8 Contours of ω̇n , T , ζ, χ, YHO2 , and |ω| for the turbulent H2 /air
diffusion flame. . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.9 Plot of mixture fraction (ζ) versus temperature in Kelvin (T ). 76
5.10 Contour lines of mixture fraction and temperature superposed
on shaded contours of reaction rate at different instants of time. 78
5.11 Comparison of temperature for unity Lewis number and non–
unity Lewis number. . . . . . . . . . . . . . . . . . . . . . . . 79
5.12 Effect of grid resolution for a turbulent diffusion flame. . . . . 80
5.13 Time evolution of flame fronts. . . . . . . . . . . . . . . . . . . 82
B.1 Comparison of a major species & temperature between Chemkin

and flamelet code for a well–stirred reactor. . . . . . . . . . . 95
B.2 Temporal evolution of mixture fraction versus T , χ, YOH , and
YN O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
B.3 Contours of |ω|, T , YH2 O , and YH . . . . . . . . . . . . . . . . 97
Nomenclature
x, y, z Coordinate axes, coordinates
ρ Density
u, v, w Velocity in the x, y and z directions
Yk Mass fraction of species k
p Pressure
T Temperature
t Time
V Volume
A Area
E Total energy per unit mass
Re Reynolds number
Sc Schmidt number
Le Lewis number
Pr Prandtl number
µ Dynamic viscosity
W Mean molecular weight
xiii
Nomenclature xiv
Ru Universal gas constant
Dk Diffusion coefficient of the k th species

cp Specific heat at constant pressure
cv Specific heat at constant volume

Qdk Heat of reaction per unit mass
ω̇k Mass reaction rate for the k th species

ω̇n Heat release
Subscripts and Superscript

( )k k th species
( )r Reference variable
( )f ace Faces of control volume
( )cv Control volume
( )d Dimensional value
Abbreviations
DNS Direct Numerical Simulation
LES Large–eddy Simulation
RANS Reynolds–averaged Navier–Stokes

Chapter 1
Introduction
1.1 Numerical method
The direct numerical simulation (DNS) and large-eddy simulation (LES) of
turbulent reacting flows are extremely challenging. Combustion involves a

large number of chemical species, and associated chemical reactions. Different
chemical reactions possess different time-scales, and different reaction zone

thicknesses. Turbulence introduces its own range of length and time-scales.
Furthermore, the nonlinear nature of turbulence can cause errors at the smaller
scales to corrupt the large scales of the solution. Turbulent combustion can
occur at very low Mach numbers (e.g. gas-turbine combustors at low pressures)
or very high Mach numbers (e.g. scramjets). Very low Mach numbers imply
numerical stiffness because of a large difference between the speed of sound

and flow velocities, while very high Mach numbers result in shock waves, and
their attendant problems. Desirable requirements for an algorithm to perform

DNS/LES of turbulent reacting flows are therefore: (i) to handle high Reynolds
numbers robustly and accurately, without the use of numerical dissipation,
1
1.1. Numerical method 2
(ii) to handle acoustic stiffness efficiently, (iii) to deal with chemical stiffness,
and combustion-induced large spatial gradients, and (iv) to deal with shock
waves/detonation. This dissertation proposes an approach that deals with

requirements (i) through (iii).
Most DNS/LES of turbulent reacting flows appear to either use the com-
pressible Navier–Stokes equations with Pade spatial discretization, and explicit
time-advancement (e.g. Poinsot [11], Trouve [50], Lele [27], & Pantano [39]),
or the zero Mach number equations along with a pressure-projection approach
(e.g. Majda & Sethian [31], Montgomery & Riley [36], Rutland & Ferziger [43]–
[44], Pember et al [40], and Mahesh et al. [30]). A second–order, staggered,

implicit compressible algorithm was proposed by Wall et al. [53]. However, the
Pade schemes become unstable at high Reynolds numbers; when explicit time-
advancement is used, they require very small time-step at low Mach numbers,
and for stiff chemical mechanisms. The zero Mach number equations are very
efficient at low Mach numbers because they analytically project acoustic waves
out; also along with implicit time-advancement they can resolve chemical stiff-
ness efficiently. However, due to the complete absence of acoustic effects, they
are not applicable to finite Mach number flows.

The approach proposed in this dissertation is an extension of the algorithm
developed by Hou & Mahesh [23] for compressible flows without chemical re-
action. The Hou & Mahesh algorithm is co–located in space, symmetric in
both space and time, and hence non-dissipative. Motivated by the theoretical
work of Thompson [49], and similar to the non-dimensionalization used by Bijl
& Wesseling [1], and van der Heul et al. [52]. The Navier–Stokes equations
are non-dimensionalized using an incompressible scaling for pressure, and the

energy equation is interpreted as an equation for the divergence of velocity. A
1.2. Auto–ignition of hydrogen vortex ring reacting with hot air 3
pressure–correction approach is used to constrain the divergence of the velocity

field to satisfy the energy equation. The resulting system of equations ana-
lytically projects acoustic waves out, in the limit of zero Mach number. The
discrete equations are fully implicit, and are constrained to discretely conserve
kinetic energy in the limit of incompressible, inviscid flow. These features

make the algorithm stable and accurate at high Reynolds numbers, and effi-
cient and accurate at very low Mach numbers. Hou & Mahesh [23] show results
for one-dimensional acoustic waves, a periodic shock tube, the incompressible
Taylor problem, and inviscid isotropic turbulence on very coarse grids. We
extend the Hou & Mahesh [23] algorithm to include chemical reaction.
1.2 Auto–ignition of hydrogen vortex ring re-
acting with hot air
The motion of a fluid column of length L through an orifice of diameter D
produces a vortex ring. Non–reacting vortex rings have been studied by many
researchers, and a large volume of work exists (see e.g. the review by Shariff
and Leonard [47]). Less is known about reacting vortex rings. This dissertation
is motivated by the relevance of vortex rings to fuel injection, and uses direct
numerical simulation (DNS) to study the flow generated by a piston pushing
a ring of gaseous fuel into hot oxidizer.

Past work on non–reacting vortex rings (Maxworthy [33], [34]; Didden [12];
Glezer [18]) has studied the process of vortex ring formation, their kinematics,
and the temporal evolution of vortex circulation. A notable finding is that
of Gharib et al [17] who show that a vortex ring achieves its maximum cir-
culation at the critical stroke ratio (termed ‘formation number’) across which
the flow transitions from a toroidal vortex ring to a vortex ring with trailing
column of vorticity. Here, the stroke length ratio is defined as the ratio of the
piston stroke length to the nozzle exit diameter. Sau and Mahesh [46] show
how the stroke length affects entrainment and mixing. They show that en-
trainment is highest at the formation number, and decreases when the trailing
column forms. They explain this behavior by showing that a toroidal vor-
tex ring entrains ambient fluid efficiently, but the trailing column does not.
As the stroke length increases towards the formation number, the vortex ring
remains toroidal, but increases in size and circulation, which increases entrain-
ment. When the stroke ratio exceeds the formation number, the leading vortex
ring remains the same, and the trailing column increases in length. The frac-
tional contribution of the leading vortex ring to overall entrainment therefore
decreases. Since the trailing column does not entrain ambient fluid efficiently,
the overall entrainment drops.
Chen et al. [5]–[7] have performed experiments on reacting vortex rings of

(pure/ nitrogen diluted) propane and ethane issuing into air. The fuel and oxi-
dizer are initially at the same temperature (300 K) in their experiments, and a
pilot flame is used to initiate combustion. Luminosity measurements are used
to provide data on flame shape and height. The fuel volume and ring circu-
lation are varied in a controlled manner. The nozzle exit–velocity is assumed
top–hat (U0 ), and ring circulation is estimated from hot–wire measurements

as Γ = U02 ∆t where ∆t is the duration of the pulse. The fuel volume is esti-
mated as VF = U0 ∆tA where A is the nozzle cross–sectional area. Chen et al.
also perform simulations of methane/air vortex rings using a steady laminar

flamelet approach, where the chemical species and temperature are functions
of the conserved scalar and scalar dissipation rate. The speed of the vortex ring
is seen to initially increase, and then decrease at later times. Dilatation due
to heat release is shown to be responsible for this behavior. The experiments

show that as the fuel volume or circulation is changed, the flame structure and
burnout time change significantly. The formation number (which they term
the overfill limit) does not significantly change from the non–reacting case.
Fuel volume which exceeds the overfill limit is seen to decrease heat release
and ring speed. Also, the burnout time and flame shapes are quite different
for ethane and propane under similar conditions although the stoichiometry
and diffusivity are nearly identical. Nitrogen dilution (for propane) decreases
burnout time and flame luminosity, but does not appreciably change flame
structure.
Simulations of reacting vortex rings in a slightly different configuration
were performed by Hewett & Madnia [19] and Safta & Madnia [45]. Hewett
& Madnia solved the axisymmetric compressible Navier–Stokes equations and
used a one–step Arrhenius mechanism to model combustion of a hydrocarbon

in air. No pilot flame exists; instead the vortex ring of fuel issues into hot
oxidizer. The stroke ratio of the vortex ring is 2 for all cases considered. The
simulations study the effect of oxidizer temperature which is controlled such
that ignition occurs either when the vortex rings forms, or occurs after the
vortex ring has fully formed. They find that at higher oxidizer temperatures,
most of the reaction occurs in front of the vortex ring, while at lower temper-
atures, most of the reaction occurs inside the ring. The heat released during
reaction is shown to affect the ring vorticity and strain rate. Safta & Madnia
[45] consider the same problem, but use a methane-hydrogen mixture as fuel.
They contrast three chemical mechanisms - GRI–Mech v3.0, and two reduced
1.3. Auto–ignition in turbulent H2 /air diffusion flames 6
mechanisms with 11 step, 15 species and 12 step, 16 species respectively.

The objective of this dissertation is to study the coupling between the
fluid flow and chemistry with as few simplifying assumptions as possible. We

therefore use direct numerical simulation to study the flow when a round nozzle
injects H2 + N2 fuel at room temperature into air at higher temperatures. The

amount of fuel injected is controlled by varying the stroke length or stroke ratio.
The governing equations are the three–dimensional, unsteady, compressible,

Navier–Stokes equations along with conservation equations for the nine species
described by the Mueller et al. [37] chemical mechanism. The simulations
consider the effects of Lewis number, stroke length ratio and fuel/air mixtures
(equivalence ratio) on auto–ignition and the flow.
1.3 Auto–ignition in turbulent H2/air diffu-
sion flames
The auto–ignition of turbulent diffusion flames is central to applications such

as diesel engines and scramjet engines where fuel is injected into hot oxidizer.
Fuel and oxidizer mix through convection and diffusion, then auto–ignite due
to the high temperatures of the oxidizer. Direct numerical simulation (DNS) is
used to study a diffusion flame where fuel (hydrogen/nitrogen) reacts with air
(oxygen/nitrogen). For laminar flames, fuel and oxidizer diffuse, then react,
yielding products. In turbulent flames, one might see auto–ignition, extinction
and even re–ignition due to the turbulence.

Mahalingam et al. [29] performed DNS of three-dimensional turbulent
non–premixed flames with finite rate chemistry. They incorporated one–step

1.3. Auto–ignition in turbulent H2 /air diffusion flames 7
and two–step reaction models for the chemistry. Their results show that the
intermediate species concentration overshoots the experiments. They suggest
that this behavior is due to time–dependent strain rates in the turbulent flow.
Montgomery et al. [36] also performed DNS of a three-dimensional turbulent
hydrogen–oxygen non–premixed flame using a reduced mechanism (7–species,

ten–reactions). The simulations were used to suggest a mixture fraction and
progress variable model for the chemical reactions. Im et al. [24] performed
DNS of two–dimensional turbulent non–premixed hydrogen–air flames to study
auto–ignition. They show that peak values of HO2 align with maximum scalar
dissipation during ignition. Im et al. also noted that weak and moderate tur-
bulence enhanced auto–ignition while stronger turbulence delayed ignition.
Echekki & Chen [15] used DNS to study auto–ignition of a hydrogen/air mix-
ture in two dimensional turbulence. Their results show spatially localized
sites where auto–ignition begins, which they define as a “kernel”. These ker-
nels have a build up of radicals at high temperatures, fuel–lean mixtures, and
low dissipation rates. Mastorakos et al. performed two–dimensional simula-

tion of auto–ignition of laminar and turbulent diffusion flames with one–step
chemistry. They found that ignition always occurs at a well–defined mix-

ture fraction (ζM R ). Hilbert & Thevenin [20] performed similar simulations to
Mastorakos et al. [32] and Im et al. [24]. The difference between Hilbert &
Thevenin [21] and others is that the simulation uses multicomponent diffusion
velocities.
A significant difference between the past work cited above, and our DNS is
that we perform fully three dimensional simulations using finite rate chemistry.
Also, we incorporate the Mueller mechanism for hydrogen/air (nine species

and nineteen reaction) and not a reduced mechanism. The objectives of our
1.4. Contributions 8
simulations are: (i) to study major differences between two dimensional and
three dimensional simulations of turbulent diffusion flames. In particular,
what happens to the flame front in three dimensions? (ii) How do the ignition
kernels evolve in time? (iii) What is the effect of unity Lewis number and
non–unity Lewis number on ignition, and (iv) How does the mixture fraction
compare to a passive scalar?
1.4 Contributions
The principal contributions of this work are:
• We have developed a novel numerical method for DNS/LES of reacting
flow. The algorithm is spatially non–dissipative, implicit and applicable

over a range of Mach number. It discretely conserves kinetic energy in
the incompressble, invisid, non–reacting limit. The algorithm is readily

applied to complex chemistry.
• DNS is used to study auto–ignition of a hydrogen vortex ring in hot air.

The most reactive mixture fraction (Mastorakos et al. [32]) is found to
predict the onset of initial ignition but not the subsequent evolution. A
most reactive temperature TM R is defined and shown to predict both the
initial auto–ignition as well as subsequent evolution.
• The effect of stroke ratio and oxidizer temperature is studied. For stroke
ratios less than the formation number, ignition in general occurs behind
the vortex ring and propagates into the core. At higher oxidizer tem-
peratures, ignition is almost instantaneous and occurs along the entire
interface between fuel and oxidizer. For stroke ratios greater than the
formation number, ignition initially occurs behind the leading vortex

ring, then occurs along the length of the trailing column and propagates
towards the ring.
• Lewis number is shown to affect both the initial ignition as well as sub-
sequent flame evolution significantly. Non–uniform Lewis number simu-

lations provide faster ignition and burnout time but a lower maximum
temperature.
• The effect of fuel/air ratio was studied. The fuel rich reacting vortex
ring provides the highest maximum temperature and the higher oxidizer
temperature provides the fastest ignition time. The fuel lean reacting
vortex ring has little effect on the flow and behaves similar to a non–
reacting vortex ring.
• DNS is used to study a turbulent hydrogen/air diffusion flame. Auto-
ignition is shown to occur in fuel lean, high temperature regions with

low scalar dissipation and low vorticity.
• Unlike laminar diffusion flames where auto-ignition occurs at ζM R , the

turbulent flame auto–ignites over a very broad range of ζ around ζM R .
• The simulations study the effects of three-dimensionality. Two–dimensional

simulations show that when flame fronts collide, extinction occurs (Mas-
torakos et al. [32]). However, our three dimensional results show that
when flame fronts collide; they can either increase in intensity or extin-
guish.
• Comparison between unity Lewis number and non–unity Lewis number

simulations show that viscous diffusion is important even in the turbulent

regime.
• It is shown that inadequate grid resolution causes auto–ignition to occur

in fuel–rich regions at low temperature. This is a non physical effect of
grid resolution.
This dissertation is organized as follows. The numerical method is de-

scribed in chapter 2. Chapter 3 discusses validation cases of varying com-
plexity. Chapter 4 considers auto–ignition of a hydrogen vortex ring reacting

with hot air. Direct numerical simulation of auto–ignition in turbulent non–
premixed H2 /air flames is discussed in chapter 5.

Chapter 2
Numerical method
This chapter extends the Hou & Mahesh [23] algorithm to include the effects
of chemical reaction. The resulting algorithm treats the chemical source terms
implicitly, solves the species equations in a segregated manner, which allows
easy extension to multiple species and chemical reactions, and reduces to the
zero Mach number equations in the limit of very small Mach number. The
chapter is organized as follows. Section 2.1 describes the non-dimensional gov-

erning equations, and their behavior in the limit of very small Mach number.
The discrete scheme is described in section 2.2. The positioning of variables,

and details of the pressure–correction approach, discretization of the chemical
source terms are discussed. Also, the behavior of the algorithm in the zero
Mach number limit is illustrated.
2.1 Governing equations
The governing equations are the compressible, reacting Navier-Stokes equation

for an ideal gas:
11
2.1. Governing equations 12
∂ρd ∂ρudj
+ = 0, (2.1)
∂td ∂xdj
!
∂ρd Ykd ∂ρYkd udj ∂ ∂Y d
+ d
= ρd Dkd kd + ω̇kd , (2.2)
∂t d ∂xj ∂xdj ∂xj
∂ρd udi ∂ρd udi udj ∂pd ∂τijd
+ = − + d (2.3)
∂td ∂xdj ∂xdi ∂xj
d d d
d
d d
∂ρ E ∂ ρ E + p uj
+
∂td ∂xdj
! N
∂τijd udi ∂ c
d p
d
∂T d X
= + d µ + Qdk ω̇kd (2.4)
∂xdj ∂xj P r ∂xdj k=1
Ru d
pd = ρ d R d T d = ρ d T . (2.5)
Wd
where the superscript ‘d’ denotes the dimensional value. The variables ρ, p, Yk
and ui denote the density, pressure, mass fraction of species k and velocities
respectively. E = cv T d + udi udi /2 denotes the total energy per unit mass and
d ∂ud
∂u ∂ud
τij = µd ∂xdi + ∂xdj − 23 ∂xdk δij is the viscous stress tensor. Dkd , cp and P r
j i k
denote the diffusion coefficient of the k th species, specific heat at constant
pressure and the Prandtl number. For the source term, Qdk is the heat of
reaction per unit mass and Qdk ω̇kd is the heat release due to combustion for
the ‘k th ’ species. ω̇k is the mass reaction rate for the k th species. Ru is the
universal gas constant and W d is the mean molecular weight of the mixture
P
defined as: W1d = N Yk
k=1 Wk . The source term is modeled using the Arrhenius
law in this paper.
The reacting governing equation are non-dimensionalized as follows. Let

ρr , Yr , L, & Tr denote the reference density, mass fraction, length and tem-
perature respectively. The reference velocity, dynamic viscosity and pressure

2.1. Governing equations 13
cdp
are denoted by ur , µr and pr respectively. The ratio γ = cdv
is assumed to
be constant (e.g. Poinsot & Veynante [42]). This yields the following non-
dimensional variables:
ρd ud td µd pd − pr
ρ= , ui = i , t = , µ= , p= , (2.6)
ρr ur L/ur µr ρ r ur 2
Td ur ur
T = , Mr = =√ , pr = ρr Rr Tr , (2.7)
Tr ar γRr Tr
Yd Dd Lω̇kd Yr Qdk
Yk = k , Dk = k , ω̇k = , Qk = , (2.8)
Yr ur L ur ρr Yr cp,r Tr
Ru Wd
Rr = , and W = . (2.9)
Wr Wr
Note that pressure is non–dimensionalized using an incompressible scaling.

This non-dimensionalization is motivated by Thompson [49], Bijl & Wesseling
[1], Van der Heul et al. [52] and Hou & Mahesh [23]. Therefore, the non-
dimensional governing equations for reacting flows are:
∂ρ ∂ρuj
+ = 0, (2.10)
∂t ∂xj

∂ρYk ∂ρYk uj 1 ∂ ∂Yk
+ = µ + ω̇k , (2.11)
∂t ∂xj ReSck ∂xj ∂xj
∂ρui ∂ρui uj ∂p 1 ∂τij
+ =− + , (2.12)
∂t ∂xj ∂xi Re ∂xj

∂ γ−1 ∂ γ−1 ∂uj
Mr2 p+ ρui ui + γp + ρui ui uj +
∂t 2 ∂xj 2 ∂xj
N
(γ − 1)Mr2 ∂τij ui 1 ∂ µ ∂T X
= + + Qk ω̇k (2.13)
Re ∂xj ReP r ∂xj W ∂xj k=1
ρT
= γMr2 p + 1. (2.14)
W
2.2. Discretization 14
where Sck is the Schmidt number for the k th species. When the reference Mach
number is zero, the non-dimensional reacting governing equations reduce to:
∂ρ ∂ρuj
+ = 0, (2.15)
∂t ∂xj
+ =− + , (2.16)

+ = µ + ω̇k , (2.17)
X N
∂uj 1 ∂ µ ∂T
= + Qk ω̇k , (2.18)
∂xj ReP r ∂xj W ∂xj k=1
ρT
= 1. (2.19)
W
Notice that the divergence of velocity equals the sum of the terms involving
thermal conduction and heat release. If the density is constant and there is
no heat release, the energy equation reduces to the incompressible continuity
equation. In the presence of heat release, the zero Mach number reacting
equations (Majda & Sethian [31]) are obtained. Most projection methods for
the zero Mach number equations project the momentum ρui to satisfy the
momentum equation. Here, the velocity is projected to satisfy the energy

equation. The reaction source term can be quite complicated for multiple
species, and is discussed in more detail in section 2.2.1.
2.2 Discretization
Density, pressure, and temperature are staggered in time from velocity by
Hou and Mahesh [23]. The mass fraction of species k are similarly staggered
in time here. The thermodynamic variables and mass fraction of species k are
VNt
VNt
VNt VNt
VNt
uti , pt+ −, Tt+ −, ρt+ −, Ykt+ −

1
2
1
2
1
2
1
2
VNt
Figure 2.1: Storage of variables.
advanced in time from t + 12 to t + 23 , illustrated in figure 1. The variables are
co–located in space, to allow easy application to unstructured grids.

Integrating over the control volume and applying Gauss’s theorem yields the
discrete governing equations. The discrete continuity and species equations

are
3 1
t+ t+
ρcv 2 − ρcv 2 1 X t+1 t+1
+ ρ V Af ace = 0 (2.20)
∆t V f aces f aces N
and
3
t+ 12
(Sk )t+
cv − (Sk )cv
2 1 X
+ (Sk )t+1 t+1
f aces VN Af ace
∆t V f aces
t+1
1 1 X 1 ∂Sk Sk ∂ρ
= µ − 2 Nj Af ace + ω̇kt+1 . (2.21)
ReSck V f aces ρ ∂xj ρ ∂xj f aces
P
where ρcv denotes ρi,j,k and f aces denotes summation over all the faces of the
control volume. ρf aces and vN denotes the density and normal face velocity at
each face. Note that Sk = ρYk . The variables (Sk )f aces denotes the species at
the face and Nj is the outward normal vector at the face. The variables V and
Af ace denote the volume of the control volume and the area of the face. The
discrete momentum equation is:
(gi )t+1
cv − (gi )cv
t
1 X t+ 21 t+ 1
+ (gi )f aces VN 2 Af ace
∆t V f aces
1 X t 1 1 X t+ 21
=− pf aces Nj Af ace + (τij )f ace Nj Af ace . (2.22)
V f aces Re V f aces
Here, gi = ρui denotes the momentum in the i direction and τij is the stress
tensor. The discrete energy equation is given by:

" t+ 21 t+ 21 #
∂ γ−1 1 X γ−1 t+ 21
Mr2 pcv + ρui ui + γpcv + ρui ui · vN Af ace
∂t 2 V f aces 2 f ace
1 X t+ 21 (γ − 1)Mr2 1 X t+ 21
+ vN Af ace = (τij ui )f ace Nj Af ace
V f ace Re V f aces
1
! N
1 1 X µ ∂T t+ 2 X t+ 1
+ Af ace + Qk ω̇k 2 . (2.23)
ReP r V f aces W ∂N k=1
The algorithm solves each equation separately. This allows one to add multiple
species with relative ease, which allows easy extension to complex chemistry.
Note that the chemical source term is handled implicitly. Details of the implicit
procedure are described in section 2.2.1. The algorithm is a pressure-correction

method. An important feature is that the face-normal velocities are projected
to satisfy the constraint on the divergence which is determined by the energy

equation. At small Mach number, this feature ensures that the velocity field is
discretely divergence free. This is in contrast to most approaches that project

the momentum which is constrained by the continuity equation. A predictor-
corrector approach is used to solve the momentum and energy equation:
3 3
pt+ 2 ,k+1 = pt+ 2 ,k + δp. (2.24)
The corrector step is the difference between the predictor equation and the
exact equation which is defined as:

t+1,k+1 ∗ ∆t ∂δp
gi,cv = gi,cv − , (2.25)
4 ∂xi cv

∆t ∂δp
ut+1,k+1
i,cv = u∗i,cv − t+1 . (2.26)
4ρcv ∂xi cv
Substituting equation (2.26) into the nonlinear term ui ui converts kinetic en-
ergy into an equation for δp which is:

t+1,k+1 ∗ ∆t ∂δp ∗ ∆t ∂δp
(ui ui ) = ui − t+1 ui − t+1
4ρcv ∂xi 4ρcv ∂xi
∆t ∂δp
= u∗i u∗i − t+1 u∗i + O(δp2 ). (2.27)
2ρ ∂xi
Equation (2.27) is then substituted into equation (2.23), and yields a discrete
energy equation in terms of δp. Note that δp is the difference between itera-
tions, which implies that δp converges to zero at each time-step. This allows
the high order terms in equation (2.27) to be neglected.
The implementation to solve the following discrete equations are to initial-
ize the outer loop:
3 1 3 1 t+ 23 ,0 t+ 12
ρt+ 2 ,0 = ρt+ 2 , ut+1,0
i
t+1,0
= uti , T t+ 2 ,0 = T t+ 2 , vN t
= vN , Sk = Sk .
(2.28)
The next procedure is to advance the continuity equation (2.20), then advance
the species equation (2.21). Once the species is advance, advance the momen-
⋆ ⋆
tum equation (2.22), then obtain vN by interpolation. After obtaining vN , the
next step is to solve the pressure correction equation (2.23). Use the corrector
steps to update pressure (2.24), momentum (2.25) and the velocities (2.26),
then check the convergence for the pressure, momentum, density, and species
between outer loop iterations.
2.2.1 Implicit source term
Consider a chemical system of N species reacting through M reactions denoted
as (Poinsot & Veynante [42]):
N
X N
X
′ ′′
νkj µk ⇋ νkj µk (2.29)
k=1 k=1
′ ′′
for j = 1, M where µk is a symbol for species k. νkj and νkj are the stoichio-
metric coefficients of species k for j reactions. The reaction term is defined
as:
M
X
ω̇k = Wk νkj Q̂j (2.30)
j=1
where
N ν ′ N ν ′′
Y ρYk kj Y ρYk kj
Q̂j = Kf j − Krj . (2.31)
k=1
Wk k=1
Wk
| {z } | {z }
forward reaction reverse reaction
Using the empirical Arrhenius law,

βj Ej βj Ta
Kf j = Af j T exp − = Aij T exp − j . (2.32)
RT T
Therefore, the source term is:
M
" N ν ′ #
X Y Sk kj
ω̇k = Wk νkj Kf j
j=1 k=1
Wk
M
" N νkj
′′
#
X Y Sk
−Wk νkj Krj . (2.33)
j=1 k=1
Wk
The backward rates Krj are computed through the equilibrium constants:
Kf j
Krj =
pa
PNk=1 ∆S 0 ∆Hj0
(2.34)
RT
exp R j − RT
where ∆Sj0 and ∆Hj0 are entropy and enthalpy, respectively. The discrete form
of equation B.11 is:

  νkj
′

M N t+ 3 t+ 1
X Y Sk,cv2 + Sk,cv2
ω̇kt+1 = Wk νkj Kft+1
   
j,cv 
j=1 k=1
2Wk
  νkj
′′

M N t+ 3 t+ 1
X  t+1 Y Sk,cv2 + Sk,cv2
−Wk νkj Krj,cv    . (2.35)
j=1 k=1
2Wk
Substituting equation (2.35) into equation (2.21) for ω̇kt+1 yields:
3
t+ 21
(Sk )t+
cv − (Sk )cv
2 1 X
+ (Sk )t+1 t+1
f aces VN Af ace
∆t V f aces
t+1
1 1 X 1 ∂Sk Sk ∂ρ
= µ − 2 Nj Af ace
ReSck V f aces ρ ∂xj ρ ∂xj f aces
  νkj
′

M N t+ 3 t+ 1
X Y Sk,cv2 + Sk,cv2
νkj Kft+1
  
+Wk j,cv
 
j=1 k=1
2Wk
  νkj
′′

M N t+ 3 t+ 1
X  t+1 Y Sk,cv2 + Sk,cv2
−Wk νkj Krj,cv    . (2.36)
j=1 k=1
2Wk
Equation (2.36) can be represented as:
t+ 3
X t+ 3
ap Scv 2 + anb Snb 2 = RHS. (2.37)
nb
where nb are the neighbors of the cv. A parallel, algebraic multi-grid approach
is used to solve the system of algebraic equations. The structured grid interface
of the Hypre library (Lawrence Livermore National Laboratory 2003) is used.

Three different examples of the implicit source term are described below.
If N = 1, M = 1 and only forward reaction is assumed, this implies that
ν11
′
β1 Ta ρY1
Q̂1 = A11 T exp − 1 (2.38)
T W1
which implies
ν11
′
β1 Ta ρY1
ω̇1 = W1 A11 ν11 T exp − 1 . (2.39)
T W1
Note that for a one-step premixed flame:

−Ta
ω̇ = BρY exp (2.40)
Td
which upon discretization yields:
t+ 3 t+ 21
t+1 Scv 2 + Scv −Ta
ω̇ =B exp t+1
. (2.41)
2 Tcv
Equation (2.41) is the source term used in section 3.1.1.
Consider a two-step reaction from Chen [4] and Mahalingam [29] given by:
A+B →I
A + I → P. (2.42)
Note that A is the fuel, B is the oxidizer, I is the intermediate step, and P is
the product. The source term for species A is defined as:

Ta Ta
ω̇k = B1 ρYA ρYB exp − 1 + B2 ρYA ρYI exp − 2 . (2.43)
T T
The discrete form is:
t+ 3 t+ 1
SA,cv2 + SA,cv2 t+1 Ta1
ω̇k = B1 SB,cv exp − t+1
2 Tcv
3 1
t+ t+
SA,cv2 + SA,cv2 t+1 Ta2
+B2 SI,cv exp − t+1 . (2.44)
2 Tcv
This implicit source term is used in section 3.1.4. Note that the other species
are handled in a similar manner.

The final example is a nonlinear source term which is used in 3.1.2 and
3.1.3:

Ta
ω̇k = −Aρ νF +νO
YFνF YOνO exp − . (2.45)
T
10
103
9
8
102
7
CF L iter 6
1
10
10
0 4
-1
10 -7 -6 -5 -4 -3 -2 -1 2 -7 -6 -5 -4 -3 -2 -1
10 10 10 10 10 10 10 10 10 10 10 10 10 10
(a) Mr (b) Mr
Figure 2.2: Variation of (a) acoustic CFL and (b) number of outer loop itera-
tions with Mr . Taylor problem , Reacting problem △.
Let νF = 2 and νO = 1. This implies:

2+1 Ta Ta
ω̇k = −Aρ YF2 YO1 exp − 2 1
= −ASF SO exp − (2.46)
T T
where F is the fuel and O is the oxidizer. The discrete equation for the fuel
species is:
 
t+ 3 t+ 1
SF,cv2 + SF,cv2 Ta
ω̇kt+1 = −A   t+1
SF,cv t+1
SO,cv exp − t+1 . (2.47)
2 |{z} Tcv
| {z } (2)
(1)
The nonlinear source term is linearized by only solving for term (1) and term
(2) is from previous iterations. The outer loop ensures that the nonlinear
corrections converge to zero.
2.2.2 Zero Mach number limit
The governing equations analytically reduce to the zero Mach number equa-
tions, as shown in section 2.1. Since the dependent variables are spatially
colocated, it appears that the pressure–projection step might result in odd–

even decoupling. It is well known that the incompressible equations require
staggering, temporal dissipation, or low order basis functions for pressure, to

avoid odd–even decoupling. However, note that the inner loop in the proposed
algorithm uses nearest neighbors for the pressure equation; it therefore does
not suffer from odd–even decoupling. The tolerance assigned to the outer loop
can in practice, be controlled to obtain low Mach number solutions. However,

a more reliable approach which explicitly introduces temporal biasing is that
proposed by [53], which is now used here.

Recall that the pressure gradient term in the momentum equation is defined
as:
1 1 1 3
∂p t+ 2 1 ∂p t− 2 1 ∂p t+ 2 1 ∂p t+ 2
= + + . (2.48)
∂xi 4 ∂xi 2 ∂xi 4 ∂xi
Temporal biasing can be introduced by computing the pressure gradient as:
1 1 1 3
∂p t+ 2 1 ∂p t− 2 1 ∂p t+ 2 1 ∂p t+ 2
= −ǫ + + +ǫ . (2.49)
∂xi 4 ∂xi 2 ∂xi 4 ∂xi
1
Taylor expansion of the pressure at time level t + 2
yields:
1 1 3
1 ∂p t− 2 1 ∂p t+ 2 1 ∂p t+ 2
−ǫ + + +ǫ =
4 ∂xi 2 ∂xi 4 ∂xi
1
!
∂ 1 ∂p t+ 2
pt+ 2 + 2ǫ∆t + O ∆t2 . (2.50)
∂xi ∂t
i.e.; the leading order term is first order in time, for a given ǫ in equation 2.50.
Alternate biased formulations of the pressure gradient can be used, but as the
Taylor series expansion shows, using ǫ of the order of ∆t would be the most
time–accurate.
Two examples are used to illustrate the low Mach number behavior of the
algorithm. The first example is the non–reacting Taylor problem discussed in
more detail in Hou & Mahesh [23], while the second example corresponds to
a synthetic premixed flame for which an analytical solution is obtained: i.e.
4
T = sin x + 2, Y = sin x + 2, u = sin x + 2, p = cos x − sin x,
3
ρ = 1/T, ω̇ = − cos x − sin x. (2.51)
For the Taylor problem, ∆t was set to 0.001 and for the reacting case ∆t
was 0.0001. Figures 2.2 (a) show the variation of acoustic CFL, and number
of outer loop iterations with Mr respectively. Mr varies from 10−2 to 10−5 for
the Taylor problem, and from 10−3 to 10−6 for the reacting case. Solutions
were also obtained for Mr equal to zero, but are not shown due to the use of
logarithmic axes. Here, ǫ was chosen to be 0.05. Note that the acoustic CFL
varies inversely with Mr ; i.e. low Mach numbers can be computed at fixed
time–step. Also, note that the number of outer loop iterations at low Mach
number is only twice that at finite Mach number, and remains constant at
very low Mach numbers.

Chapter 3
Validation simulations
The algorithm is applied to simulate flames using both simple (3.1) and com-
plex (3.2) chemical mechanisms. Section 3.1.1 considers a one–dimensional
steady laminar premixed flame while an unsteady laminar diffusion flame is
considered in section 3.1.2. These two examples illustrate the ability to han-
dle large heat release at low Mach number. Section 3.1.3 considers a steady
two–dimensional laminar reacting jet flame with large heat release at low Mach
number. Three–dimensional numerical simulations of turbulent non–premixed
flames with finite rate chemistry are considered in section 3.1.4. This problem
illustrates application to turbulence and finite Mach number. The final ex-
amples in section 3.2 illustrates application to complex chemistry for H2 /air

combustion using a 9 species, 19 reaction mechanism from Mueller et al. [37].
The well–stirred reactor problem in section 3.2.1 is used to study chemical

stiffness. A grid convergence study of a laminar diffusion flame is shown in
section 3.2.2. Section 3.2.3 and 3.2.4 discuss simulations of a laminar jet flame
and a lifted jet flame, respectively.
25
3.1. Simple chemistry 26
10-1
T Error -2
3 10
2
10-3
1 -1 0
-10 -5 0 5 10 10 10
(a) x (b) dx
Figure 3.1: (a) Comparison of computed temperature to analytical solution

for a laminar premixed flame. analytic solution, numerical solution.
Sc = Re = P r = 1, β = 10, α = 0.8, Mr = 0.01, (b) spatial order of
accuracy. Algorithm, second order, Sc = Re = P r = 1, α = 0.5, β = 2,
Mr = 0.01.
3.1 Simple chemistry
3.1.1 Laminar premixed flame
An irreversible, one–step, laminar premixed flame is considered as an example
of a low Mach number reacting flow. The reaction is given by R → P where

R is the reactant and P is the product. The reaction source term is that
proposed by Echekki & Ferziger [16]. It approximates the Arrhenius source

term, and is useful for validation, in that it permits analytical solution. The
reaction model is defined as:



 0 for Θ < Θc ,
ω̇ = (3.1)

 β (β − 1) (Θ − 1) otherwise .
The corresponding analytical solution is given by:



 (1 − β −1 ) exp (x) for x ≤ 0,
Θ= (3.2)
 1 − β −1 exp [(1 − β) x] for x ≥ 0 .

Generally, β is the order of 10 for hydrocarbon combustion (Echekki & Ferziger

[16]).
Figure 3.1 (a) shows computed results for a premixed flame where β = 10,
α = 0.8, and the inflow Mach number is equal to 0.01. The Echekki & Ferziger
source term is used. The inflow velocity is set equal to the flame speed, and the
solution is initialized using a hyperbolic tangent profile for the temperature
distribution. The inlet boundary condition for temperature, density, species

and velocity are set to one. Zero derivative boundary conditions are specified
at the outflow. Boundary conditions based on characteristic analysis are not

specified, due the simplicity of the solution at the boundaries, and the use of
‘sponge’ boundary conditions at both inflow and outflow boundaries to absorb

acoustic waves that are generated by the initial transient. A cooling term,
−σ (U − Uref ) is added to the right hand side of the governing equations over
the sponge zone, whose length is 10 percent of the domain. Here U and
Uref denote the vector of conservative variables and the ‘reference’ solution
respectively. The coefficient σ is a polynomial function defined as:
(x − xs )n
σ (x) = As , (3.3)
(Lx − xs )n
where xs and Lx denotes the start of the sponge and the length of the domain.
n and As are equal to three. After a brief transient, the solution settles into
a steady state which is shown in figure 3.1 (a). The computed and analytical
solutions seem to agree well.

1 3.5
0.8 3
0.6 2.5
YO,64 T64
0.4 2
0.2 1.5
0 1
0 1 2 3 4 0 1 2 3 4
(a) x (b) x
Figure 3.2: Comparison of asymptotic solution to numerical solution for a

laminar diffusion flame. (a) mass fraction of oxidizer and (b) temperature
profile at a non-dimensional time of 64. asymptotic solution, numerical
solution, Mr = 0.001, Sck = 1, P r = 1.
Figure 3.1 (b) shows results from a grid–convergence study. The time–step
is fixed at dt = 0.001, and the grid size is equal to 32, 64, and 128. Note that
second–order accuracy is obtained. The computation cost is as follows. For
N x = 128, ten outer loop iterations, one continuity equation iteration, five
iterations for each momentum equation, one iteration for species and three
iterations for the pressure–correction equation are required. The corresponding

residual for the outer loop is 10−8 and 10−15 for the inner loop.
3.1.2 Laminar diffusion flame
A one-dimensional, unstrained diffusion flame with one-step chemistry is com-

puted. This problem applies the algorithm to a high Damkohler number (de-
fined as the ratio of flow time scale to chemical time scale, Da = 50 × 106 )
chemical mechanism and very low Mach number. The chemistry model is a
one-step reaction defined as:
νF YF + νO YO → νP YP (3.4)
where YF , YO and YP are the mass fractions of the fuel, oxidizer and products.
The computed solution is compared to an asymptotic solution developed by

Cuenot & Poinsot [11]. Their analysis is applicable to diffusion flames with
variable density, non-uniform Lewis number, and finite rate chemistry. Cuenot
& Poinsot studied unsteady unstrained, steady strained and unsteady strained
H2 −O2 flames. Case 2 from their paper corresponds to an unsteady, unstrained

flame, and was chosen for comparison. Table 3.1 lists the conditions for case
2.
Table 3.1: Test condition.
case LeF TF,0 LeO TO,0 Φ νF νO A Ta Q/cp Re

2 1 300 K 1 300 K 8 2 1 108 3600 K 6000 K 10000
Note that LeF and LeO denote the Lewis number for fuel and oxidizer.
TF,0 and TO,0 are the initial temperatures of the fuel and oxidizer. Φ is the
equivalence ratio which is defined as:
YF,0 νO WO YF,0
Φ=s = (3.5)
YO,0 νF WF YO,0
where s is the stoichiometric ratio. The equivalence ratio is used to determine if
the mixture is rich (Φ > 1), lean (Φ < 1) or stoichiometric (Φ = 1). νF and νO
denote the stoichiometric coefficients of the species. A is the pre–exponential
factor and Ta is the activation temperature. The Arrhenius reaction rate is

defined by Cuenot & Poinsot as:
νF νO
ρYF ρYO Ta
ω̇k = −νk Wk A exp − . (3.6)
WF WO T
Figure 3.2 is a comparison of the computed solution to the asymptotic so-
lution from Cuenot & Poinsot. The asymptotic solution was used to initialize
the temperature, fuel mass fraction, and oxidizer mass fraction. The initial
velocity profile was calculated from the energy equation. The initial pressure
remains uniform in the flame and the initial density is obtained from tem-
perature. Initially the reaction rate is zero because the asymptotic solution
assumes infinitely fast chemistry. However, the simulation involves (fast) finite
rate chemistry. Therefore, the reaction rate rapidly recovers in the simulation
(Cuenot & Poinsot). The inlet boundary condition for temperature, density,
and mass fraction of fuel are set to one. Mass fraction of oxidizer and velocity
are set to zero. Zero derivative boundary conditions are specified at the out-
flow. The initial time was set at ct/L = 20 and the solution was advanced to
a time of ct/L = 64. Again, the ‘sponge’ boundary conditions were used at
both inflow and outflow boundaries to absorb acoustic waves generated by the
initial transient. The sponge zone is 10 percent of the domain. Also, the flow
Mach number is 0.001 and the grid used 512 points. Reasonable agreement
between computed and asymptotic solution is obtained.
3.1.3 Laminar two–dimensional jet flame
A steady two dimensional reacting laminar jet flame from Poinsot & Veynante
[42] is considered. This problem illustrates the ability to handle large heat
release and nearly incompressible flow for a one-step diffusion flame with a
0.8
y ζ 0.6
0.4
0.2
0 20 40 60 80 100 120 140
(a) x (b) x
Figure 3.3: (a) Contour plot of temperature for a jet flame. (b) Center line pro-
file of the mixture fraction ζ. asymptotic solution from Poinsot & Veynante
[11], numerical solution. Mr = 0.01, Re = 500, Sck = P r = 1.
Damkohler number of 50 × 106 . From Poinsot & Veynante [42], if one assumes
constant mass flow rate (ρu = constant), v = w = 0 and ρD = constant, then
the mixture fraction ζ satisfies:
∂ζ ∂2ζ
ρ F uf = ρ F DF 2 .
∂x ∂x
Note that ζ is defined as:

ΦYF − YO Φ−1
ζ = −2 + (3.7)
Φ+1 Φ+1
where Φ is the equivalence ratio. A similarity solution is obtained:

1 y2
ζ (x, y) = √ exp − (3.8)
2 παx 4αx
DF
where α = uF
. This solution assumes infinitely fast chemistry. Since the
Damkohler number in the computed solution is high (fast chemistry), the sim-
ulation results should show reasonable agreement with the similarity solution.
The initial conditions for temperature, mass fuel fraction and mass oxidizer
fraction are the similarity solution. The non-dimensional pressure is set to zero
and from the equation of state yields the density profile. The inlet boundary
condition for temperature, density, species and velocity are set to the similarity
solution. Zero derivative boundary conditions are specified at the outflow. In
the y–direction, the boundary conditions are set to a constant. Since the
mass flow rate is constant, this yields the u velocity profile. Again, sponge
boundary condition are used and a steady state solution is obtained after
the initial transient. Figure 3.3 (a) shows the temperature contours of the
simulation. Figure 3.3 (b) compares the similarity solution to the simulation
results where the Mach number is 0.01 and the grid is 128 by 128. Note that
reasonable agreement is obtained.
3.1.4 Turbulent diffusion flame with one-step chemistry
A one-step diffusion flame interacting with isotropic turbulence is considered.

The purpose of this calculation is to evaluate turbulence interacting with a
diffusion flame at finite Mach number. The simulations are compared to Ma-
halingam [29] and Chen [3]. The simulation uses a cubic domain with inflow
and outflow boundary condition in the x–direction (same as section 3.1.2) and
periodic boundary condition in the y and z direction. The grid uses 64 by 64
by 64 points and the Mach number is 0.1. The initial condition and parameters
for the simulation are given in the table 3.2. From table 2, lt is the turbulent
(a) (b)
Figure 3.4: (a) Scalar dissipation rate χ and (b) reaction rate ω̇ at z = 4
for turbulent diffusion flame. Notice that scalar dissipation is high where the
reaction rate is low. Mr = 0.1, Re = 935, Sck = P r = 1.
(a) (b)
Figure 3.5: (a) Iso–surface plot of the reaction rate for turbulent diffusion
flame. (b) Contour plot of temperature showing entire computational domain.
Mr = 0.1, Re = 935, Sck = P r = 1.
1 1
0.8 0.8
0.6 0.6
YF YO
0.4 0.4
0.2 0.2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(a) x (b) x
Figure 3.6: Scatter plot of (a) fuel mass fraction and (b) oxidizer mass fraction
from DNS of one-step, turbulent diffusion flame. Mr = 0.1, Re = 935, Sck =
P r = 1.
Table 3.2: Parameters for turbulence.

L lt η
Reλ lt δf l
Da
δf l
50 6 2 0.2 1
integral length scale and L is the length of the computational domain. δf l is

the laminar flame thickness and Reλ is the initial Taylor Reynolds number. η
is the Kolmogorov scale and Da is the initial global Damkohler number defined
as: " #
Z
lt 1
Da = ω̇T dx . (3.9)
u0 δf l δf l
The initial turbulence field is a three-dimensional turbulent flow where the

kinetic energy spectrum is defined as:
4 " 2 #
u2 k k
E (k) = C0 0 exp −2 (3.10)
k0 k0 k0
Table 3.3: Parameters for diffusion flame.
zst β α νA νB A (m3 mol−1 s−1 )

0.5 8 0.8 1 1 1012
where k is the wave number, k0 is the wave number at which E(k) is maximum,
and u0 is the rms velocity.

The initial diffusion flame is obtained by asymptotic solution (Cuenot &
Poinsot [11]); table 3 shows the relevant parameters. From table 3, zst is
1
defined as zst = 1+Φ
where Φ is the equivalence ratio. νA and νB are the sto-
ichiometric coefficients. A is the pre–exponential factor. The diffusion flame
is advanced in time to remove any acoustic transients. After the initial tran-
sients are removed, the initial turbulent velocity field is superimposed on the
diffusion flame. The solution is advanced from 0 to 2.0 eddy turnover times
and analyzed at each 0.1 eddy turnover time. Local extinction is observed to
occur in the reaction rate. Holes occur in the reaction zones where pure mix-
ing exists and the hole location corresponds to a high rate of scalar dissipation
rate as predicted by laminar flamelet theory. This is shown in figures 3.4 and
3.5 (a). Figure 3.5 (b) is a contour plot of temperature showing the entire
computational domain. Figure 3.6 shows two-dimensional mass concentration

distribution of fuel mass fraction and oxidizer mass fraction where good agree-
ment is obtained with Chen [3] (figure 2 (a) in their paper). The lower bound
on the species is the location of the flame and the upper bound is influenced
by extinction. The distribution between the lower and upper bounds is the
turbulence induced mixing (Mahalingam [29] & Chen [3]).

3.2. Complex chemistry 36
4E-06
3E-06 YHO2
YH2 T
2E-06
1E-06
0
0 0.02 0.04 0.06 0.08 0.1
(a) time (sec) (b) time (sec)
Figure 3.7: (a) Comparison of a major species & temperature between

Chemkin and present algorithm for a well–stirred reactor. ([YH2 ] Algo-
rithm, ▽ Chemkin & [T ] in kelvin, Algorithm, Chemkin) (b) Illustra-
tion of the effect of time–step by comparing the explicit Euler (dt = 1.0e − 8)
to present implicit source term (dt = 0.001 & dt = 0.0001). implicit
dt = 1.0e − 3, implicit dt = 1.0d − 4 explicit dt = 1.0d − 8.
3.2 Complex chemistry
This section illustrates application of the algorithm to complex chemistry.

Combustion of H2 and air is considered, using a 9 species and 19 reaction
mechanism [37].
3.2.1 Well stirred reactor
Figure 3.7 illustrates the ability of the implicit procedure described in section
2.2 to handle chemical stiffness. A perfectly–stirred reactor problem is consid-

ered, the time–step is varied, and the results are compared to Chemkin [26]
for accuracy. The implicit treatment of the chemical source terms allows the
time step to be five orders of magnitude higher (1.0e−3) than that possible
using explicit Euler time advancement (1.0e−8). The maximum error of the
Table 3.4: Parameters for laminar diffusion flame.
YH2 YO2 YN2 ,F YN2 ,O TF TO p u Mr pr Tr

0.029 0.233 0.971 0.767 1 4 0 0 0.4 1 atm 300 K
implicit solution (computed using YHO2 ) compared to the explicit solution was
around 5 percent, for dt = 1.0e−3. For dt = 0.0001, the percent error was
around 0.01 percent.
A laminar diffusion flame with complex chemistry was simulated. The param-
eters are listed in table 3.4. Figure 3.8 shows results from a grid convergence
study. Uniform grids ranging from 64 to 2048 points were used to obtain
a grid–converged solution. A nonuniform grid of 128 points that clustered

points in the flame was also used. The resolution was estimated to accurately
compute the instant when the flame was thinnest and the heat release was
highest. Figures 3.8 (a) and (b) show that the solution is stable and qual-
itatively correct even for the 64 points which only has 2 to 3 points in the
flame.
3.2.3 Laminar jet flame
A three dimensional laminar hydrogen/air round jet flame in heated co–flow is

considered. The jet Reynolds number is 1000, fuel jet width is 1 cm and other
relevant parameters are listed in table 3.5. The inflow boundary conditions
are shown in table 4.1. The first derivative is set to zero for thermodynamic
variables, species, and velocities at the outflow. At the y and z boundaries, the
YH2 O YH
(a) x (b) x
Figure 3.8: Grid convergence study showing profiles of a major species and
a minor species. (a) H2 O (b) H. N=64, △ N=128 ,
∇ N=128 nonuniform grid, ⊲ N=256, ⋄ N=512, ◦ N=1028, N=2048.
Table 3.5: Simulation parameters
YH2 YO2 YN2 T U p

Fuel 0.029 0 0.971 300 (K) 187 m/s 101325 Pa
Oxidizer 0 0.233 0.767 1200 (K) 75 m/s 101325 Pa
thermodynamic variables, velocities and species are set to a constant. ‘Sponge’

boundary conditions are used at the transverse boundaries to absorb acoustic
waves that are generated by the flame. The computational grid is 128 by 64
by 64 and the domain is 30 by 10 by 10. P r is 0.7, µ is equal to T 0.7 and unity
Lewis number (Lek = 1) is assumed.

Figure 3.9 shows the temporal variation of maximum ω̇n , T , YOH , YO , YH ,
YH2 O , YHO2 and YH2 O2 . A steady state solution is obtained at 170 units of
time and the final maximum temperature is 2370 K. Figure 3.10 illustrates
the steady state contours of T (K), YOH , YHO2 and YH2 O2 .

ω̇n T
(a) time (b) time
Figure 3.9: (a) time versus maximum ω̇n and T . (b) time versus
maximum species of YOH , YO , YH , YH2 O , YHO2 and YH2 O2 .
Table 3.6: Simulation parameters
YH2 YO2 YN2 T U p

Fuel 0.11 0 0.89 400 (K) 347 m/s 101325 Pa
Oxidizer 0 0.233 0.767 1100 (K) 100 m/s 101325 Pa
3.2.4 Sandia lifted jet flame
Experiments on a lifted turbulent hydrogen/air slot jet flame in a heated coflow

were performed at Sandia National Laboratories. The jet Reynolds number
is 11,000 and the simulation parameters are listed in table 3.6. The fuel jet
width is 2 mm. We perform preliminary two dimensions simulations under
the same conditions as the Sandia jet flame. The boundary conditions are
similar to that described in section 3.2.3. The computational grid is 1024 by
512 by 2 and the domain is 30 by 14 by 1. P r is 0.7, µ is equal to T 0.7 and

the nonuniform Lewis numbers are from Im et al. [24].
Figure 3.11 (a) is a contour plot of temperature; note the temperature in-
(a) (b)
(c) (d)
Figure 3.10: Contours of (a) T (K), (b) YOH , (c) YHO2 and (d) YH2 O2 for
laminar diffusion jet flame.
(a) (b)
(c) (d)
Figure 3.11: Contour plots of (a) T (K), (b) YHO2 , (c) ωz and (d) divergence.
crease down stream of the jet as expected for a ‘lifted’ jet. The precursor to
auto–ignition is adequate accumulation of the species YHO2 . Figure 3.11 (b)
shows the species YHO2 exists down stream in the jet. Figure 3.11 (c) shows the
roll–up of vorticity as the jet evolves, while the acoustic waves generated by the
combustion are illustrated in figure 3.11 (d). Note that the acoustic waves ap-
pear to emanate from the ignition regions. Future simulations should consider
three–dimensionality, larger domain and non–reflecting boundary conditions
to better represent the acoustic field.

Chapter 4
Auto–ignition of a hydrogen
vortex ring reacting with hot air
The objective of this chapter is to study auto–ignition of fuel injected into

hot air. We therefore use direct numerical simulation to study the flow when
a round nozzle injects H2 + N2 fuel at room temperature into air at higher

temperatures. The amount of fuel injected is controlled by varying the stroke
length or stroke ratio. The governing equations are the three–dimensional, un-
steady, compressible, Navier–Stokes equations along with conservation equa-
tions for the nine species described by the Mueller et al. [37] chemical mecha-
nism. The algorithm proposed in chapter 2 is used. The simulations consider
the effects of Lewis number, stroke length ratio and fuel/air mixtures (equiv-
alence ratio) on auto–ignition and the flow.
Chapter 4 is organized as follows. Section 4.1 describes relevant details

of the simulation. It provides a brief description of the initial and boundary
conditions. Results from a validation and grid convergence study are shown
in section 4.2. The homogeneous mixture problem is discussed in section 4.3
42
4.1. Problem statement 43
Figure 4.1: Schematic of the problem.
and is used to obtain a chemical time scale and ζM R . Section 4.4 discusses
results for the reacting vortex ring. A brief summary in section 4.5 concludes
this chapter.
4.1 Problem statement
Figure 4.1 shows a schematic of the problem and illustrates the ring vortex
and the trailing column which forms at large stroke ratios. The dimensional
variables for the reacting vortex ring are chosen such that the equivalence ratio
(φ) is one, and a stoichiometric mixture is obtained where YH2 ,0 equals 0.02936
and YO2 ,0 equals 0.233. For fuel lean mixtures, φ equals 0.25 (YH2 = 0.0074)
and for fuel rich mixtures φ equals 4 (YH2 = 0.1185). For all simulations, Re is
1000, P r is 0.7 and µ is equal to T 0.7 . The non–uniform Lewis numbers ([24]):
LeH2 = 0.3, LeO2 = 1.11, LeO = 0.7, LeOH = 0.73, (4.1)
LeH2 O = 0.83, LeH = 0.18, LeHO2 = 1.10, LeH2 O2 = 1.12 (4.2)
Sck
where Lek = Pr
. Reference pressure, length, Mach number and time are
1.0132e5 P a, 0.01 m, 0.4, and 6.6e − 5 sec, respectively.
Table 4.1: Non–dimensional inlet boundary conditions.

h hp ii
Tin (TO − TF ) 1 − 12 − 21 tanh 10 y2 + z 2 − 1 + TF
h p i
uin uF 12 − 12 tanh 10 y2 + z 2 − 1 Utime
h p i
YH2 ,in YF0 12 − 12 tanh 10 y2 + z 2 − 1 Utime
h hp ii
1 1 2 2
YO2 ,in YO0 1 − 2 − 2 tanh 10 y +z −1
pin 0
M Win
ρin Tin
The inflow boundary conditions are shown in table (4.1). TO and TF denote
the oxidizer and fuel temperatures respectively. uF is the fuel velocity at the
nozzle exit. YF0 and YO0 denote the mass fractions of fuel and oxidizer respec-
tively. Utime controls the stroke length (L/D). Note that all reference variables
are non-dimensionalized with respect to the fuel side. The first derivative is
set to zero for thermodynamic variables, species, and velocities at the outflow.
At the y and z boundaries, the thermodynamic variables and species are set
to a constant (same as initial conditions) and the velocities are set to zero.
‘Sponge’ boundary conditions are used at the outflow and transverse bound-
aries to absorb acoustic waves that are generated by the flame. A cooling term,
−σ (U − Uref ) is added to the right hand side of the governing equations over
4.2. Validation and grid–convergence study 45
0.8
0.6
YF T YH2
0.4
0.2
0
-3 -2.5 -2 -1.5 -1 -0.5 0
(a) r (b) time
Figure 4.2: (a) Comparison using one–step chemistry to Hewett and Madnia
( [19]). Present, Hewett & Madnia. (b) Comparison of a major species
& temperature between Chemkin and present algorithm for a well–stirred re-
actor using the Mueller mechanism. ([YH2 ] Algorithm, △ Chemkin & [T ]
Algorithm, Chemkin).
the sponge zone whose length is 10 percent of the domain. Here U and Uref
denote the vector of conservative variables and the ‘reference’ solution respec-
tively. The coefficient σ for the sponge at the outflow, is a polynomial function
defined as:
(x − xs )n
σ (x) = As ,
(Lx − xs )n
where xs and Lx denotes the start of the sponge and the length of the domain.
n and As are equal to three.
4.2 Validation and grid–convergence study
The simulations are first validated using a one–step chemical mechanism, and
compared to the one-step simulations of Hewett & Madnia [19]. Our results
YH2 O
YH
(a) y (b) y
Figure 4.3: Grid convergence study for a two–dimensional laminar reacting

vortex ring. (a) YH (b) YH2 O . N=64, N=128, Nx=128, Ny=Nz=128
nonuniform in y, Nx = 256, Ny=Nz=128 nonuniform in y, N=
256, N = 512.
are compared to case 5 in their paper where the temperature ratio is 6:1 and
the stroke length is two. Figure 4.2 (a) shows that good agreement is obtained.
Next, complex chemistry is validated using the homogeneous mixture or
well–stirred reactor problem. We use the 9 species (H2 , O2 , OH, O, H, H2 O,

HO2 , H2 O2 , and N2 ), 19 reaction Mueller mechanism (Mueller et al. [37])
which is shown in table 4.2. A detailed experiment and simulation of a lam-

inar hydrogen jet diffusion flame was performed by Cheng et al. [8]. Their
work shows that chemical mechanisms from Miller & Bowman [35] , GRI–Mech
3.0 [48], Mueller et al [37], Maas & Warnatz [28], and O’ Conaire et al. [38]
compare well with experimental results. Therefore, the Mueller mechanism

should perform adequately for our hydrogen/air reacting vortex ring simula-
tion. Figure 4.2 (b) is a comparison between Chemkin [26] and our algorithm
incorporating the Mueller mechanism. Good agreement is observed for the
homogeneous mixture.
Table 4.2: Mueller Mechanism: (cm3 –mole–sec–kcal–K)
Af j βj Ej
H2 /O2 Chain Reactions
1 H + O2 ⇋ O + OH 1.910E14 0.0 16.44
2 O + H2 ⇋ H + OH 5.080E04 2.67 6.29
3 H2 + OH ⇋ H2 O + H 2.160E08 1.51 3.43
4 O + H2 O ⇋ OH + OH 2.970E06 2.02 13.4
H2 /O2 Dissociation/Recombination Reactions
5 H2 + M ⇋ H + H + M 4.580E19 −1.40 104.38
6 O + O + M ⇋ O2 + M 6.160E15 −0.5 0
7 O + H + M ⇋ OH + M 4.710E18 −1.0 0
8 H + OH + M ⇋ H2 O + M 2.210E22 −2.0 0
Formation and Consumption of HO2
9 H + O2 + M ⇋ HO2 + M 3.50E16 −0.41 −1.12
10 HO2 + H ⇋ H2 + O2 1.660E13 0.0 0.82
11 HO2 + H ⇋ OH + OH 7.080E13 0.0 0.3
12 HO2 + O ⇋ OH + O2 3.250E13 0.0 0
13 HO2 + OH ⇋ H2 O + O2 2.890E13 0.0 −0.50
Formation and Consumption of H2 O2
14 HO2 + HO2 ⇋ H2 O2 + O2 4.200E14 0.0 11.98
15 H2 O2 + M ⇋ OH + OH + M 1.20E17 0.0 45.5
16 H2 O2 + H ⇋ H2 O + OH 2.410E13 0.0 3.97
17 H2 O2 + H ⇋ H2 + HO2 4.820E13 0.0 7.95
18 H2 O2 + O ⇋ OH + HO2 9.550E06 2.0 3.97
19 H2 O2 + OH ⇋ H2 O + HO2 1.000E12 0.0 0
The three dimensional computational grid used in the simulations is based
on a grid refinement study of a two–dimensional H2 /air reacting vortex ring.

The computational domain was 10 by 10 in x and y where the nozzle is in the
y direction. Uniform grids ranging from 64 to 512 points were used to obtain
a grid–converged solution. Also, a non–uniform grid of 128 points in the y
direction (to cluster points around the nozzle) and 128 uniform points in the
x direction was used. Figures 4.3 (a) and (b) show profiles in the middle of
the reacting vortex ring. Note that grid convergence is obtained for the non–
4.3. Homogeneous problem 48
H
ω̇n T
(a) time (b) time
Figure 4.4: (a) Rates of reactions 1,2,3,9,10, and 11 for species H. (b) Temporal
evolution of temperature with all reactions, and reactions 1, 2, 3, 9, 10 & 11.
All reactions: (T ) , (ω̇n ) reactions 1, 2, 3, 9, 10, & 11: (T ) ,
(ω̇n ) .
uniform grid of 1282 and uniform 2562 and 5122 grids. To ensure adequate
resolution for the three dimensional reacting vortex ring, a non–uniform 128
grid in y, z directions and 256 in the x direction was chosen, and the size of the
computational domain was 10 by 10 by 10 in the three coordinate directions.
4.3 Homogeneous problem
We first consider the homogeneous mixture or well–stirred reactor problem.
Our objectives are to: (i) obtain an ignition time scale; (ii) obtain an opti-
mum mixture fraction for auto–ignition (Mastorakos et al. [32]); (iii) identify
which reactions contribute to auto–ignition and which reactions contribute af-

ter auto–ignition. Echekki & Chen [15] used DNS to study auto–ignition of
a hydrogen/air mixture in two dimensional turbulence. They found that the

intermediate species O, H, OH, and HO2 play an important role in the auto–
τchem τchem
(a) ζ (b) ζ
Figure 4.5: (a) Mixture fraction versus τchem for varying temperature in kelvin.
φ = 1, T = 1100, φ = 1, T = 1200, φ = 1, T = 1300, and
φ = 1, T = 1400. (b) Mixture fraction versus τchem for varying fuel/air
ratio, φ = 1/4, T = 1200, φ = 1, T = 1200, φ = 4, T = 1200
and φ = 8, T = 1200.
ignition process. In the Mueller mechanism, reactions 1,2,3,9,10 and 11 have
a significant contribution to the minor species O, H, OH, and HO2 . Figure

4.4 (a) shows reaction rates of species H due to reactions 1,2,3,9,10 and 11.
Figure 4.4 (b) shows temperature and heat release when all the reactions are
simulated, and when only the auto–ignition reaction (1,2,3,9,10, and 11) are
simulated. Note that the maximum heat release for the full reaction set and
the auto–ignition reaction set occurs at about the same time. Up until this
instant of maximum heat release, the auto–ignition reactions are dominant;
after this instant, the other reactions are also significant. This instant is de-
fined as τchem and will be used to obtain the most reactive mixture fraction,
ζM R .
Mastorakos et al. [32] used one–step chemistry and the homogeneous mix-
ture problem to obtain τchem and the mixture fraction most likely to auto–
ignite (ζM R ). The following equations are used to set up the well–stirred
reactor problem in terms of the mixture fraction defined as:
YH2 = YH2 ,0 ζ, YO2 = YO2 ,0 (1 − ζ) , YN2 = 1 − YH2 − YO2 , (4.3)
T = TO − ζ (TO − TF ) .
For example, if ζ = 0.5, YH2 ,0 = 0.029, YO2 ,0 = 0.233, TF = 300 (K) and
TO = 1200 (K), then YH2 , YO2 , YN2 , & T would be 0.5, 0.0145, 0.1165, 0.869, &
750 (K), respectively. One can now consider different initial mixture fractions,
obtain the minimum ignition time (τchem ) and thereby obtain the most reactive
mixture fraction (ζM R ).

We obtain ζM R for the Mueller mechanism. Figure 4.5 (a) shows mixture
fraction versus τchem for initial temperatures that range from 1100 to 1400
Kelvin and figure 4.5 (b) shows mixture fraction versus τchem for fuel/air ratios
(φ) that range from 0.25 to 8. Note in figure 4.5 that the minimum τchem
corresponds to ζM R . Figure 4.5 also shows the effect of fuel/air mixture and
temperature on τchem . As temperature increases, τchem becomes smaller and

ζM R shifts to the right. As φ increases, τchem becomes smaller and ζM R shifts
to the left. These results illustrate the balance between adequate fuel and
high enough temperature for auto–ignition to occur. Table 4.3 lists the most
reactive mixture fraction and corresponding values of the initial temperature,
fuel/air ratio and τchem .

4.4. Results for reacting vortex ring 51
Table 4.3: Global results for homogeneous mixture problem.
T0 φ ζM R τchem TM R
1100 1 0.06 3.4e − 4 1052
1200 1 0.08 1.4e − 4 1128
1300 1 0.09 8.5e − 5 1210
1400 1 0.10 5.6e − 5 1290
1800 1 0.13 1.8e − 5 1605
1200 1/4 0.1 3.5e − 4 1110
1200 4 0.06 8.0e − 5 1146
1200 8 0.04 6.4e − 5 1164
4.4 Results for reacting vortex ring
Table 4.4 lists the different cases studied. The simulations consider the effect
of Lewis number, stroke length ratio (L/D), fuel/air ratio and oxidizer tem-
perature. The non–uniform Lewis numbers are obtained from Im et al [24].

Two stroke ratios of 2 and 6 are considered. Note that for non–reacting vortex
rings, a stroke ratio of 2 yields a coherent vortex ring, while a stroke ratio of 6
yields a vortex ring followed by a trailing column of vorticity (Sau & Mahesh
[46]). Oxidizer temperatures (TO ) of 1200 and 1800 Kelvin are considered. Fi-
nally, fuel/air ratios were chosen such that the equivalence ratios (φ) are equal
to 1/4, 1 and 4 yielding lean, stoichiometic and rich mixtures, respectively.
Table 4.4: Parameters used in reacting vortex ring simulations.
Lewis number φ (L/D) TO

non–uniform 1 2 1200
unity 1 2 1200
non–uniform 0.25 2 1200
YH
time
Figure 4.6: Domain average of rates of reactions 1,2,3,9,10 and 11 for the
species H. non–uniform Lewis reaction 1, reaction 2, reaction
3, reaction 9, reaction 10, reaction 11.
4.4.1 Ignition and flame evolution
We first consider the non–uniform Lewis number simulations at stroke ratio
of 2 and oxidizer temperature of 1200 K. Figure 4.6 shows reactions 1,2,3,9,10

and 11 for the species H. Time is normalized by τchem obtained from table
4.3. Comparison to a corresponding plot for the homogeneous mixture (figure

4.4 a) shows qualitative similarities along with some differences. The reaction
rates are seen to peak at time greater than τchem for the reacting vortex ring,
reflecting the finite amount of time taken for fuel and oxidizer to be advected
and then diffuse to yield regions of ζM R . Also, the relative importance of
reactions 9 and 11 is higher for the vortex ring,. The time scale on which
the reaction rates decay to zero following the initial ignition, is longer for the
reacting vortex ring, reflecting the propagation of the flame through the vortex
ring.
ζ T χ ζ T χ
ω̇n ω̇n
(a) ζ (b) ζ
ζ T χ ζ T χ
ω̇n ω̇n
(c) ζ (d) ζ
ζ T χ ζ T χ
ω̇n ω̇n
(e) ζ (f) ζ
Figure 4.7: Two–dimensional slices showing variation in time of ζ versus ω̇n .

The color contours correspond to ω̇n . On each color contour, a colored contour
line is plotted for ζ, T and χ. The color contour of ω̇n ranges from 0 to 1.
For the contour lines, purple is ζ (contour lines range from 0.04 to 0.12), T is
black (contour lines range from 1120 to 1140 K) and χ is red (contour lines
range from 0.0 to 0.15). Non–dimensional time: (a) 2.5, (b) 3.0, (c) 3.5, (d)
4.0, (e) 4.5 and (f) 5.0.
ζM R
L
D
=2
TM R
t : 2.5 3.0 3.5 4.0 4.5 5.0 5.5
ζM R
L
D
=6
TM R
t : 2.5 3.0 3.5 4.0 4.5 5.0 5.5
ζM R
TO = 1800
TM R
t : 0.25 0.75 1.25 1.75 2.25 2.75 3.25
Figure 4.8: Time evolution of heat release (ω̇n ), ζM R and TM R . On each color
contour of heat release, a contour line is superimposed for ζ and T . The
contour of ω̇n ranges from 0 to 1. The contour lines of ζ range from 0.04 to
0.12 (ζM R ) and temperature range from 1092 to 1164 K (TM R ) for L/D = 2
and L/D = 6. For oxidizer temperature of 1800 K, ζ ranges from 0.10 to 0.18
(ζM R ) and temperature ranges from 1590 to 1650 K (TM R ).
Figure 4.7 illustrates the process of auto–ignition in the vortex ring using
scatter plots of mixture fraction versus heat release at different instances in
time. At each time instant, heat release contours in a radial cross–section are
shown. Superposed on the heat release contours is a colored contour line of
mixture fraction (ζ), temperature (T ) and scalar dissipation (χ), respectively.

Note that the color contours of heat release (ω̇n ) range from 0 to 1. Mixture
fraction and scalar dissipation are defined as (Poinsot and Veynante [?]):

1 YF YO
ζ= φ 0 − 0 +1 , (4.4)
1+φ YF YO

2 ∂ζ ∂ζ
χ= , (4.5)
Re ∂xj ∂xj
respectively. For the contour lines, mixture fraction is purple (contour lines
range from 0.04 to 0.12), temperature is black (contour lines range from 1120
to 1140 K) and scalar dissipation is red (contour lines vary over the entire range
of 0.0 to 0.15). Note that the ranges for the contour lines of mixture fraction
and temperature were chosen based on the results of the homogeneous mixture
problem. Recall ζM R is 0.08 from the homogeneous problem when T0 = 1200
K and φ = 1 (table 4.3).

Figure 4.7 shows that auto–ignition occurs in regions of high temperature,
fuel-lean mixtures, and low scalar dissipation where ζ = ζM R . Ignition occurs

at a ζM R of 0.08 and propagates towards the stoichiometric side (ζ = 0.5).
Most of the ignition occurs behind the vortex ring and then moves rapidly
into the core. The contour line of mixture fraction shows that auto–ignition
occurs in fuel lean regions (ζ = ζM R = 0.08). Also, the contour lines of scalar
dissipation show that auto–ignition occurs in regions of low scalar dissipation.
The superposed contour lines of temperature are chosen to lie around TM R

which is 1130 K (TM R = 1200 − ζM R [1200 − 300]). Interestingly, maximum

heat release follows TM R at every instant of time in figure 4.7. Figure 4.8 is a
side by side comparison of ζM R and TM R . This figure shows that the onset of
ignition occurs at ζM R and TM R . As ignition continues from behind the vortex
ring into the core, ζM R is in the very fuel lean region and does not follow the
path of ignition. TM R on the other hand, follows the path of ignition all the
way into the core. This suggests that TM R instead of ζM R might be better
in predicting the evolution of auto–ignition. This behavior is observed in all
cases simulated (figure 4.8); i.e. in the pre–ignition phase, ζM R predicts the
onset of ignition fairly well, however TM R predicts both the onset of ignition
as well as its subsequent evolution.
Table 4.5 lists most reactive mixture fraction, burnout time in seconds,
maximum temperature in Kelvin and ignition time in seconds obtained from
all the simulations. Note that the onset of ignition largely depends on fuel/air
ratio and oxidizer temperature. However, in the post–ignition phase, the evo-
lution of the flame is greatly affected by other parameters. This behavior is

discussed in the following sections.
Table 4.5: Global behavior of reacting vortex ring.
φ Lek 1200 L/D ζM R burnout (sec) Tmax ignition time

1 Lek 6= 1 1200 2 0.08 5.6e − 4 1887 1.7e − 4
1 Lek = 1 1200 2 0.08 7.1e − 4 2105 2.3e − 4
1/4 Lek 6= 1 1200 2 0.1 7.4e − 4 1227 2.8e − 4
4 Lek 6= 1 1200 2 0.055 5.5e − 4 2890 1.3e − 4
1 Lek 6= 1 1800 2 0.14 5.0e − 4 2263 1.7e − 5
1 Lek 6= 1 1200 6 0.09 1.0e − 3 2054 1.7e − 4
4.4.2 Evolution of heat release
Figure 4.9 shows the effects of Lewis number, fuel/air ratio, oxidizer temper-
ature and stroke ratio on heat release. An ignition delay is noticeable for all
cases except the higher oxidizer temperature simulation. The ignition times
for all cases are listed in table 4.5. For non–uniform and uniform Lewis number
at stroke ratio of 2 (figures 4.9 a and b), ignition occurs behind the vortex ring,
and then propagates rapidly into the core. Note that the heat release for the
non–uniform Lewis number simulation is clearly different from that obtained

using unity Lewis number. At non–uniform Lewis numbers, the lighter fuel
diffuses faster towards the oxidizer, thereby decreasing ignition time.

It is interesting to examine why auto–ignition occurs behind the vortex
ring. Figure 4.10 shows velocity vectors in a radial cross–section superposed

on contours of mixture fraction. Note that the ring entrains oxidizer in the
rear; the region of ζM R is therefore ‘thicker’ behind the vortex ring compared
τr 6.6e−5
to the front. Since the global Damkohler number is low (Da = τchem
= 1.4e−4
=
4.7e − 01), the effects of this entrainment will be felt on the time–scale of the
chemistry. Auto–ignition therefore occurs behind the ring in this region where
ζM R covers a larger area. Figure 4.10 also shows the effects of Lewis number.
Note that the regions of ζM R in the non–uniform Lewis number simulations are
quite different from that obtained at unity Lewis number (figure 4.10 b). In
particular non–uniform Lewis numbers yield thicker regions where the mixture
fraction is equal to its most reactive value. These results show the importance
of both, flow–chemistry coupling and Lewis number in the evolution of the

flame.
The effect of fuel/air ratio is similar to that observed in the homogeneous

(a)
(b)
(c)
(d)
(e)
(f)
t: 1 2 3 4 5 6 7
Figure 4.9: Time evolution of heat release (ω̇n ). The contours are separated
by one unit of time for a total of seven units of time. (a) non–uniform Lewis
number at L/D of 2 and φ of 1, (b) uniform Lewis number at L/D of 2 and
φ of 1, (c) non–uniform Lewis number at L/D of 2 and φ of 0.25, (d) non–
uniform Lewis number at L/D of 2 and φ of 4, (e) non–uniform Lewis number
at L/D of 2 and oxidizer temperature of 1800 K and (f) non–uniform Lewis
number of L/D of 6 and φ of 1.
(a) (b)
Figure 4.10: Contours of heat release (ω̇n ) and velocity vectors. On each color
contour, a contour line is superimposed for ζ. (a) non–uniform Lewis number
(time = 2.25) and (b) unity Lewis number (time = 3.0).
mixture problem. Recall from table 4.5 that τchem for the fuel–lean and rich
cases was 3.5e−4 and 8.0e−5 seconds, respectively. The fuel lean vortex ring
therefore has minimal heat release and is the last to auto–ignite (figure 4.9 c).
The fuel rich reacting vortex ring (figure 4.9 d) auto–ignites behind the vortex
ring due to entrainment there, but note that the heat release in the fuel rich
vortex ring (figure 4.9 d) is much more intense compared to the stoichiometic
case (figure 4.9 a).
For the higher oxidizer temperature of 1800K, ignition occurs nearly in-
stantaneously (figures 4.9 e). This is because τchem is an order of magnitude
smaller at the higher oxidizer temperature (table 4.5). Ignition now occurs
along the entire interface between fuel and oxidizer and not just behind the
ring. As the flame evolves, heat release is mostly in the front of the vortex
ring. The resulting products are then shed into the ring wake and therefore
no reaction occurs there.

(a) (b)
Figure 4.11: Contours of heat release (ω̇n ), temperature (T ) and velocity vec-
tors at L/D of 6 (time = 4). On each color contour, a contour line is super-
imposed for ζ. (a) ω̇n and (b) T in Kelvin.
The stroke ratio changes the spatial behavior of the initial ignition, but not
the time taken for initial auto–ignition. Recall from the introduction section
that the vortex ring of stroke length 6 is identical to the vortex ring of stroke
length of 2 until a non–dimensional time of 2. The effect of stroke length is
therefore apparent after non–dimensional time of 2. For the fuel/air ratios

and oxidizer temperatures considered, the initial auto–ignition occurs at non–
dimensional time (Upiston t/D) around 2 as seen in figure 4.10. The fuel that
is injected before Upiston t/D = 2 therefore auto–ignites in the same manner

as the vortex ring with stroke ratio = 2. However fuel that is injected after
the Upiston t/D = formation number (approximately 4, [17]) exits the nozzle
in a trailing column. As shown in Sau & Mahesh ([46]) and as visible in
figure 4.11 (a) the trailing column does not entrain surrounding oxidizer very
efficiently. Fuel and oxidizer now mix due to the shear at the edges of the
trailing column; auto–ignition therefore occurs in regions of ζM R there. The

flame then propagates towards the vortex ring. Also, note that the expansion
associated with heat release in the trailing column forces oxidizer away from
the ring (figure 4.11 b).
4.4.3 The importance of Lewis number
The temperature field (figure 4.12) may be qualitatively anticipated from the
behavior of heat release discussed above. Also, temperature shows the im-
portance of accounting for different rates of diffusion of the different chemical
species. Note that for the non–uniform Lewis number simulations, the tem-
peratures are high both in front and rear of the vortex ring. However, when
all Lewis numbers are assumed to be unity, temperature rises only behind the
vortex ring. The fuel lean vortex ring shows very little temperature rise due to
low levels of heat release, and almost behaves like a non–reacting vortex ring.
The fuel rich vortex ring on the other hand shows high temperatures. At the
higher stroke ratio the temperature rise is along the the trailing column. Ta-
ble 4.5 lists maximum temperature and burnout time for all simulations. Note
that the maximum temperature is higher in the uniform Lewis number flame
compared to the non–uniform Lewis number. This behavior has also been
observed in axisymmetric simulations of a laminar hydrogen–air jet diffusion
flame (Katta et al. [25], figure 1 of their paper).

The Lewis number has an interesting effect on the impact of higher oxidizer
temperature. The one–step chemistry simulations of Hewett & Madnia [19]

and our simulations using the Mueller mechanism at unity Lewis number ([14])
show that lower oxidizer temperature provides a faster burnout time than the
higher oxidizer temperature. This counterintuitive behavior may be explained
(a)
(b)
(c)
(d)
(e)
(f)
t: 1 2 3 4 5 6 7
Figure 4.12: Time evolution (seven units of time) of temperature (T ) in Kelvin.

(a) non–uniform Lewis number at L/D of 2 and φ of 1, (b) uniform Lewis
number at L/D of 2 and φ of 1, (c) non–uniform Lewis number at L/D of 2
and φ of 0.25, (d) non–uniform Lewis number at L/D of 2 and φ of 4, and (e)
non–uniform Lewis number of L/D of 6 and φ of 1.
by noting that at high oxidizer temperature, ignition is nearly instantaneous.

The flow immediately expands, which decreases the vorticity, and therefore
decreases further mixing. The expansion also affects the entrainment of ox-
idizer. Instead of pushing fresh oxidizer behind the core, oxidizer is pushed
into the higher temperature region where there is less fuel. The burnout time
is therefore longer.
However, this behavior is an artifact of unity Lewis number. The simula-

tions with non–uniform Lewis number (Lek 6= 1) show that the burnout time
is faster in the higher oxidizer temperature flame compared to that at lower
oxidizer temperature (table 4.5). This behavior may be explained as follows. If

the uniform and non–uniform Lewis number flames are compared at the same
temperature, ignition time is smaller for the non–uniform Lewis number flame
since the fuel diffuses faster towards the oxidizer. Increasing oxidizer temper-
ature decreases ignition time for both cases. However, the relative reduction
in ignition time will be less for the non–uniform Lewis number flame since the
faster diffusion of the fuel contributes significantly to its lower ignition time.
A general conclusion from figure 4.12 is that each flame has a unique
burnout time, maximum temperature and flame shape which cannot be de-
termined by the homogeneous mixture problem, and is a consequence of the
coupling between the flow and chemistry.

Figures 4.13 (c) show how heat release affects the vorticity of the ring.
Note that in the absence of any reaction, the maximum vorticity increases
when fluid is being injected and then stays approximately constant until the
injection is complete and the injected shear layers roll–up into the vortex ring.
After injection, the vorticity decreases steadily due to viscous dissipation. Note
that the lean fuel/air ratio ring behaves similarly to a non–reacting vortex ring
ω̇n T
(a) time (b) time
|ω|
(c) time
Figure 4.13: Maximum value of (a) heat release, (b) Temperature in Kelvin (T )
and (c) vorticity magnitude versus time in seconds. non–uniform Lewis
number, unity Lewis number, φ = 0.25, φ = 4, oxidizer
temperature of 1800 K, L/D of 6.
4.5. Summary 65
because it has very little heat release (figure 4.13 c). The non–uniform Lewis
number simulations show the same qualitative trend as the non–reacting ring
until the instant (4 time units) when temperature (figure 4.12) reaches the core,
at which point the vorticity drops sharply due to core expansion. The uniform
Lewis number simulations show a similar trend; the main difference is that
the vorticity drops after 6 units of time reflecting the longer ignition delay. At
higher oxidizer temperature, heat is released before injection is complete. The

initial acceleration of the flow therefore increases the magnitude of vorticity
in the injected shear layer at very short times. However once the injection
is completed and the ring is formed, the vorticity drops sharply as shown in
figure 4.13 (c).
4.5 Summary
This chapter uses direct numerical simulation to study the combustion of a

vortex ring of hydrogen issuing into hot air. A detailed chemical mechanism
from Mueller et al [37] is used. The simulations use a novel, spatially non–
dissipative numerical method developed by Doom et al. [13]. The simulations
study the effect of fuel/air ratio, Lewis number, oxidizer temperature and
stroke ratio (ratio of piston stroke length to diameter). Also, the well–stirred
reactor problem is used as a baseline case to identify the effects of the flow on
the auto–ignition and evolution of the flame.
Results show that auto–ignition occurs in fuel lean regions of high temper-
ature and low scalar dissipation at the most reactive mixture fraction (Mas-
torakos et al.[32])). We also define a most reactive temperature from the

homogeneous mixture problem (TM R ) and find that while ζM R does a good
4.5. Summary 66
job of predicting the onset of ignition, it does not predict the subsequent evo-
lution. TM R is found to predict both the initial ignition and the subsequent
evolution for all cases considered.

As Gharib et al. ([17]) show, coherent vortex rings are obtained for stroke
ratios less than the formation number, while a vortex ring followed by a trailing
column of vorticity is obtained for stroke ratios greater than the formation
number. At moderately high oxidizer temperatures, coherent vortex rings first

ignite behind the vortex ring and the heat release then propagates into the core.
This is because the ring entrains surrounding oxidizer and sheds mixed fluid
in its wake. At higher oxidizer temperatures, ignition occurs instantaneously,

along the entire interface between fuel and oxidizer. At stroke ratios greater
than the formation number, ignition around the leading vortex rings is different
from that in the trailing column. Ignition initially occurs behind the leading
vortex ring, and then occurs along the length of the trailing column as more
fuel is injected.
The coupling between the flow and chemistry is quite non–trivial even for
this simple problem. The Lewis number affects both the initial ignition time
as well as the subsequent evolution of the flame. At non–uniform Lewis num-

bers, faster ignition and burnout times, but lower maximum temperatures
are obtained. The Lewis number also affects the location of the initial igni-
tion regions. The influence of higher oxidizer temperature on burnout time is
completely different when uniform Lewis numbers are considered. The effect
of fuel/air ratio may be anticipated from the homogeneous mixture problem.
Higher fuel/air ratios yield higher temperatures and shorter ignition times,
while very lean vortex rings behave essentially like non–reacting vortex rings.
Chapter 5
Turbulent H2/air diffusion

flames
This chapter uses direct numerical simulation to study ignition in a turbulent

diffusion flame and is organized as follows. Section 5.1 provides details of the
laminar diffusion flame, turbulent diffusion flame and validation. The simula-
tions results are discussed in section 5.2. In the results section, auto–ignition
(5.2.1), Lewis number (5.2.2), grid resolution (5.2.3) and flame front evolution
(5.2.4) are discussed in detail. A brief summary in section 5.3 concludes the
chapter.
5.1 Problem statement
Figure 5.1 (a) shows a laminar diffusion flame. The reference variables for
the diffusion flame are shown in table 5.1. Note that Mr , τr , Tr , and ρr are
reference Mach number, reference time, reference temperature, and reference
67
α fuel oxidizer
(a) x (b)
Figure 5.1: Schematic of initial conditions for the turbulent diffusion flame.
(a) Diffusion flame and (b) Isotropic turbulence.
density, respectively. For all simulations, Re is 1000, P r is 0.7 and µ is equal

to T 0.7 . For the unity Lewis number cases, Lek = 1 and the non–unity Lewis
numbers are ([24]):
LeH2 = 0.3, LeO2 = 1.11, LeO = 0.7, LeOH = 0.73, (5.1)
LeH2 O = 0.83, LeH = 0.18, LeHO2 = 1.10, LeH2 O2 = 1.12 (5.2)
Sck
where Lek = Pr
. The initial conditions for the diffusion flame are from
Mastorakos et al.[32]:
YH2 = YH02 ζ (x) ,
YO2 = YO02 [1 − ζ (x)] ,
T = TO + [TF − TO ] ζ (x) ,

(x − xc )
α (x) = 0.5 1 − erf . (5.3)
δ
Table 5.1: Reference variables.
Mr τr Tr ρr
0.1 1.48e-4 (sec) 300 (K) 0.8282 (kg/m3 )
ω̇n
YH2 T
(a) time (b) x
Figure 5.2: (a) Comparison of a major species & temperature between

Chemkin and present algorithm for a well–stirred reactor. ([YH2 ] Al-
gorithm, △ Chemkin & [T ] Algorithm, Chemkin). (b) Grid convergence
study showing heat release (ω̇n ).
α is the initial conditions for mixture fraction (figure 5.1 a). Note that δ equal
0.05 which is the thickness of the mixing layer and xc is the center of the
mixing layer where it is equal to π. YH02 , YO02 , TO , and TF are equal to 0.029,
0.233, 4, and 1, respectively. This will yield an equivalence ratio (φ) of one.
The non–dimensional pressure was set to zero. Density was obtained using
equation 2.14. The inlet boundary condition are set to a constant and zero
derivative boundary conditions are specified at the outflow.

5.2. Results 70
5.1.2 Turbulent diffusion flame
Three dimensional isotropic turbulence (figure 5.1 b) is superimposed on the

laminar diffusion flame. The initial isotropic turbulence field is prescribed by
the three–dimensional kinetic energy spectrum:
r 4 " #
2
2 u20 k k
E (k) = 16 exp −2 . (5.4)
π k0 k0 k0
The turbulent Reynolds number Reλ is 50 and k0 is 5. The computational
domain is 2π for x, y, and z. The computational grids for the turbulent

diffusion flame are 128 and 256 cubed. All turbulent diffusion flame figures
are 128 cubed excect figure 5.8 which is 256 cubed.
5.1.3 Validation
The homogeneous mixture problem is used to validate the complex chemistry.
Figure 5.2 (a) is a comparison to Chemkin [26]. Good agreement is observed for
the homogeneous mixture problem. A grid convergence study was performed
for the laminar diffusion flame. Figure 5.2 (b) shows results from the grid
convergence study. Uniform grids ranging from 64 to 2048 points were used.
A grid–converged solution is obtained for a grid of 128 points.
5.2 Results
5.2.1 Auto–ignition of diffusion flame
This section discusses auto–ignition of the diffusion flame and its relation to
temperature and mixture fraction. The passive scalar and mixture fractions
5.2. Results 71
ζB ζB
ζP ζP
(a) Z (b) Z
Figure 5.3: (a) scatter plot of the laminar diffusion flame with unity Lewis
number mixture fraction ( ζB and ζP ) versus passive scalar (Z) from a
non–dimensional time of zero to 10. (b) scatter plot of the laminar diffusion
flame with non–unity Lewis number mixture fraction ( ζB and ζP ) versus
passive scalar (Z) from a non–dimensional time of zero to 10.
used in the simulations are defined as:

∂ρZ ∂ρZuj 1 ∂ ∂Z
+ = µ , (5.5)
∂t ∂xj ReScZ ∂xj ∂xj

1 YF YO
ζP = φ 0 − 0 +1 , (5.6)
φ+1 YF YO

2YH2 YH 2YH2 O YOH YHO2 2YH2 O2
ZH = WH + + + + + , (5.7)
W WH WH2 O WOH WHO2 WH2 O2
H2
2YO2 YO YH2 O YOH 2YHO2 2YH2 O2
ZO = WO + + + + + ,
WO2 WO WH2 O WOH WHO2 WH2 O2
1

Z /WH + YO02 − ZO /WO
2 H
ζB = 1 0 .
Y /WH + YO02 /WH
2 H2
Here Z denotes the passive scalar variable. ζP and ζB denote mixture fractions
from Poinsot & Veynante [42] and Hilbert & Thevenin [20], respectively. Note
that ζB includes all species where ζP accounts for only fuel and oxidizer. For
the passive scalar equation, ScZ was chosen to be 0.7. Figure 5.3 is scatter
5.2. Results 72
Ignition (ζM R )
ω̇n T
ω̇n
(a) ζ (b) time
Figure 5.4: (a) evolution of mixture fraction versus heat release and the lo-
cation of ignition ζM R . (b) temporal evolution of maximum heat release and
temperature for the laminar diffusion flame. temperature of unity Lewis
number, temperature of non–unity Lewis number, heat release of
unity Lewis number, and heat release of non–unity Lewis number.
plot of the laminar diffusion flame for unity and non–unity Lewis number. If
the passive scalar and mixture fraction were equal, one would obtain a straight
line. Note that the unity Lewis number yields a straight line but the non–unity
Lewis number does not. These results show that mixture fraction (as defined
above) is a conserved scalar for the unity Lewis number but not the non–unity
Lewis number. Figure 5.3 also shows the slight differences between ζP and ζB ,
observed well after the formation of products. Therefore, ζP is adequate for

studying auto–ignition and from now on, ζ = ζP .
Figure 5.4 (a) illustrates the evolution of ignition for the laminar diffusion
flame. Note that ignition occurs in fuel lean, high temperature regions (ζ =
ζM R = 0.08) and propagates towards the stoichiometric side (ζ = 0.5). Unity

Lewis number behaves in a similar manner, but ignition occurs at a later time
(figure 5.4 a). Recall that for unity Lewis number, the mixture fraction is
5.2. Results 73
Figure 5.5: Evolution of temperature from a non–dimensional time of 0 to 4.
a passive scalar. Therefore, auto–ignition occurs in the same region (ζM R ).

On the other hand, for non–unity Lewis number, auto–ignition occur at ζP =
ζB = 0.08 for the mixture fractions and Z = 0.13 for the passive scalar. This
result shows that auto–ignition occurs in a ‘richer’ region for the passive scalar
equation than the mixture fraction for non–unity Lewis number.
For the turbulent diffusion flame, figures 5.5, 5.6 and 5.7 show the evolution
of T , YHO2 , and YOH , respectively. Each contour plot is separated by a unit
of time from zero to four. Figure 5.5 illustrates the effect of auto–ignition on
temperature. In figures 5.6 and 5.7, note the differences between YHO2 and
YOH , especially the range of length scales.

Contour plots (in the x = π plane) of ω̇n , T , ζ2 , χ2 , YHO2 , and |ω| at a non–
dimensional time of 0.9 for the turbulent diffusion flame are shown in figure 5.8.
The contour lines represent the mixture fraction over the range from 0.05 to
0.30. Note that ignition occurs in fuel lean regions (5.8 a). Echekki and Chen
5.2. Results 74
Figure 5.6: Evolution of YHO2 from a non–dimensional time of 0 to 4.
Figure 5.7: Evolution of YOH from a non–dimensional time of 0 to 4.

5.2. Results 75
ω̇n T
(a) (b)
ζP χ
(c) (d)
|ω| YHO2
(e) (f)
Figure 5.8: Contours of ω̇n , T , ζ, χ, YHO2 , and |ω| at a non–dimensional time

of 0.9 for the turbulent H2 /air diffusion flame. Contour lines are superimposed
for the mixture fraction ζ and range from 0.05 to 0.30.
5.2. Results 76
T T
(a) ζ (b) ζ
Figure 5.9: Plot of mixture fraction (ζ) versus temperature in Kelvin (T ) (a)
laminar diffusion flame and (b) is the turbulent diffusion flame.
define such ignition sites as kernels. They describe the kernels sites as regions
of high–temperature (5.8 b), fuel–lean mixtures (5.8 (c)) and low dissipation
rates (5.8 (d)) which is present in figure 5.8. Note that scalar dissipation is
defined as:

2 ∂ζ ∂ζ
χ= . (5.8)
Re ∂xj ∂xj
In the turbulent diffusion flame, there will also be regions with low vorticity
(figure 5.8 e). The ignition sites have low vorticity which allows time for fuel
and oxidizer to diffuse together to form essential products for auto–ignition to

occur e.g. HO2 . The formation of HO2 is the precursor to auto–ignition (Im
et al. [24]) and is shown in figure 5.8 (f).

Figure 5.9 shows the effect of mixing on auto–ignition at ζM R . Recall from
the laminar diffusion flame that auto–ignition occurred at ζ = ζM R = 0.08

(figure 5.4 a). The homogeneous mixture problem (chapter 4.3) predicted
5.2. Results 77
the onset of ignition (ζM R ) very well for the laminar diffusion flames. On the
other hand, ζM R does not entirely predict the onset of ignition for the turbulent
diffusion flame. This phenomenon is completely due to the turbulent mixing

which results in a range of temperatures for a given mixture fraction. For
example, if ζ = 0.2, then temperature ranges from 950 K to 1050 K (figure 5.9
b). Recall from the homogeneous mixture problem, that temperature plays
an important role in auto–ignition. Table 4.3 illustrate that a 100 Kelvin

difference changes τchem by a factor of 2.3. For the turbulence case, this can
greatly affect the onset of ignition and location. This is why auto–ignition
occurs in a very broad range for ζM R in the turbulent diffusion flame while the
laminar case is at ζM R .
These results suggest that maybe one should track temperature instead
of mixture fraction in predicting auto–ignition and the evolution of ignition.
Figure 5.10 superposes contour lines of temperature which are chosen to range
from 950 to 1150 K (TM R = 1200−ζM R [1200 − 300]). Interestingly, maximum
heat release follows TM R at every instant of time in figure 5.10. This figure
shows that the onset of ignition occurs at ζM R and TM R . As ignition continues,
ζM R is in the very fuel lean region and does not follow the path of ignition.
TM R on the other hand, follows the path of ignition at each instance in time.
This suggests that TM R instead of ζM R might be better choice in predicting

the evolution of auto–ignition.
5.2.2 The effect of Lewis number
Figure 5.4 (a) shows the profiles of maximum heat release and temperature
versus time for the laminar diffusion flame. Note that ignition occurs at a
5.2. Results 78
ζ at time = 1.8 T at time = 1.8
Figure 5.10: Contour lines of mixture fraction (0.05 to 0.3) and temperature
(950 to 1150 K) superposed on shaded contours of reaction rate at different
instants of time.
5.2. Results 79
(a) (b)
Figure 5.11: Comparison of temperature for (a) unity Lewis number and (b)
non–unity Lewis number at a non–dimensional time of 3.
non–dimensional time of around 1.6 for the one dimensional diffusion flame
with unity Lewis number while ignition occurs around 1 for the non–unity
Lewis number. The temperature difference between the unity Lewis number
and non–unity Lewis number is 450 K. The fastest ignition time was obtained
for the non–unity Lewis number turbulent diffusion flame. The first signs of
auto–ignition starts around 1.6, 1.0, 1.4 and 0.9 (non–dimensional time) for
the laminar unity Lewis number, laminar non–unity Lewis number, turbu-
lent unity Lewis number and turbulent non–unity Lewis number, respectively.
These results shows that differential diffusion plays an important in the auto–
ignition of the diffusion flame. The difference between the turbulent unity
Lewis number and non–unity Lewis number is further demonstrated in figure

5.11. The comparison between temperature in figure 5.11 shows that the two
cases are complete different, i.e. viscous diffusion is important even in the
turbulent regime.
5.2. Results 80
(a) (b)
Figure 5.12: This figure illustrates an insufficient grid resolution. Ignition

occurs in a fuel rich region shown in the orange box. The contour plot is heat
release and the contour lines are mixture fraction ranging from 0.5 to 0.9.
5.2.3 Effect of grid resolution
Grid resolution has an interesting effect on auto–ignition. If the resolution is

not adequate, one sees auto–ignition occuring in regions with fuel rich regions
at low temperatures. These regions are shown in figure 5.12 in the orange
box. The importance of this result on large eddy simulation of auto–ignition
is apparent.
5.2.4 Evolution of ignition fronts
It is generally felt (e.g. Hilbert & Thevenin [20] and Mahalingam et al.[29])
that the main difference between three dimensional and two dimensional tur-
bulent simulation is that two dimensional turbulence lacks vortex stretching
and high dissipation rates. Our results suggest that three-dimensional simu-
lations do matter in the collision of flame front and that dimensionality af-
fects the results at a more fundamental level. Mastorakos et al [32] performed

5.3. Summary 81
two-dimensional simulations using one–step chemistry. They found that fronts

propagate outward and collapse at the very lean or rich sides due to extinction,
or after collisions with separate similar fronts originating from other ignition
sites [32]. Our results show that when flame fronts collide; they can either
increase in intensity, combine without any appreciable change in intensity, or

extinguish. In most cases, the flame fronts combine together (figure 5.13).
This phenomenon is due to the extra dimension. In two–dimensional simu-

lations when flame fronts collide, there is no extra dimension for products to
travel which ‘chokes’ the flow. In the three-dimensional simulation, products
can migrate to the extra dimension which allows the flame fronts to combine
together.
5.3 Summary
This chapter uses direct numerical simulation to study auto–ignition of a hy-

drogen/air turbulent diffusion flame. A detailed chemical mechanism due to
Mueller et al [37] is used. Simulations of three dimensional turbulent diffusion

flames were performed. Isotropic turbulence is superimposed on an unstrained
diffusion flame where diluted H2 at ambient temperature interacts with hot

air. Both, unity and non–unity Lewis number are studied. The results are
contrasted to the homogeneous mixture problem (chapter 4.3) and a laminar

diffusion flames. Results show that auto–ignition occurs in fuel lean, low vor-
ticity, and high temperature regions with low scalar dissipation around ‘most
reactive’ mixture fraction, ζM R . For the laminar case, auto-ignition occurs at
ζM R while the turbulent case auto–ignites occurs in a very broad range of ζM R .

Simulation also study the effects of three-dimensionality of the turbulent diffu-
5.3. Summary 82
(a) (b)
(c)
Figure 5.13: Contour plots of ω̇n showing time evolution of flame fronts. (a)
time = 2.6, (b) time = 2.9, and (c) time = 3.2.
5.3. Summary 83
sion flame. Two–dimensional simulations show that when flame fronts collide,
extinction occurs (Mastorakos et al. [32]). Our results show that when flame
front collide; they can either increase in intensity, combine together or go ex-
tinct which is due to the extra dimension. Three dimensional simulations are
essential to turbulent diffusion flames and in particular, the collision of flame

fronts. Ignition kernels are related to the most reactive mixture fraction, but
the most reactive fraction cannot completely predict the onset of ignition be-
cause ignition occurs in a broad range of ζM R . We also show that the passive
scalar equation does not agree with non–unity Lewis number mixture fraction
and that viscous diffusion is important even in the turbulent regime.

Appendix A
Non–Dimensionalizing the
Navier–Stokes equations
This appendix illustrates how the governing equations are non–dimensionalized.

The governing equations are the compressible, reacting Navier-Stokes equation
for an ideal gas:
∂ρd ∂ρudj
+ = 0, (A.1)
∂td ∂xdj
!
∂ρd Ykd ∂ρYkd udj ∂ ∂Y d
+ d
= ρd Dkd kd + ω̇kd , (A.2)
∂td ∂xj ∂xdj ∂xj
+ = − + d (A.3)

∂ρd E d ∂ ρd E d + pd udj
+
∂td ∂xdj
! N
∂τijd udi ∂ c d
d p ∂T
d X
= d
+ d µ d
+ Qdk ω̇kd (A.4)
∂xj ∂xj P r ∂xj k=1
84
85
Ru d
pd = ρ d R d T d = ρ d T . (A.5)
Wd
where the superscript ‘d’ denotes the dimensional value. The variables ρ, p, Yk
and ui denote the density, pressure, mass fraction of species k and velocities
respectively. E = cv T d + udi udi /2 denotes the total energy per unit mass and
d ∂ud
∂u ∂ud
τij = µd ∂xdi + ∂xdj − 23 ∂xdk δij is the viscous stress tensor. Dkd , cp and P r
j i k
denote the diffusion coefficient of the k th species, specific heat at constant
pressure and the Prandtl number. For the source term, Qdk is the heat of
reaction per unit mass and Qdk ω̇kd is the heat release due to combustion for
the ‘k th ’ species. ω̇k is the mass reaction rate for the k th species. Ru is the
universal gas constant and W d is the mean molecular weight of the mixture
P
defined as: W1d = N Yk
k=1 Wk . The source term is modeled using the Arrhenius
law.
Non-dimensional variables are defined as:
ρd ud td µd pd − pr
ρ= , ui = i , t = , µ= , p= , (A.6)
ρr ur L/ur µr ρ r ur 2
Td ur ur
T = , Mr = =√ , pr = ρr Rr Tr , (A.7)
Tr ar γRr Tr
Ykd Dkd Lω̇kd Yr Qdk
Yk = , Dk = , ω̇k = , Qk = , (A.8)
Yr ur L ur ρr Yr cp,r Tr
Ru Wd Yr hds,k
Rr = , W = , and hs,k = . (A.9)
Wr Wr cp,r Tr
The corresponding non-dimensional numbers are defined as:
λd ρd ν d cdp νd ρ d ud x d
Lek = , P r = , Sc k = , Re = (A.10)
ρd cdp Dkd λd Dkd µd
Note that kinetic theory of gases (Hischelder et al. [22]) shows that λ changes
A.1. Non–dimensionalizing the governing equations 86
1
roughly like T 0.7 , ρ is T
and Dk is T 1.7 . Therefore, Lek changes only by a
few percent in the flame. Also, note that Sck = P rLek . Some basic ideal
gas assumptions are used: E = cv T + 12 ui ui , cv = R

, cp
γ−1 Rr
= γ
γ−1
, γ = cp
cv
,
cp = R+cv , a2r = γRr Tr . Note that an important assumption is that γ remains
constant. cp and cv are not constant but the ratio of the two is constant. The
following expressions are useful to the subsequent discussion.
µd cdp
λd = , (A.11)
Pr
µd
ρd Dkd = , (A.12)
Sc
Ru Ru Rr
Rd = d = = , (A.13)
W W Wr W
R Ru Ru Rr
cv = = d = = , (A.14)
γ−1 W (γ − 1) Wr W (γ − 1) W (γ − 1)
pr ρr u2r pr u2r γpr u2r γpr u2r
ρr u2r = = = = = pr γMr2 . (A.15)
pr Rr Tr γRr Tr a2r
A.1 Non–dimensionalizing the governing equa-
tions
A.1.1 Mass
∂ρd ∂ρudj
+ = 0, (A.16)
∂td ∂xdj
∂ρρr ∂ρr ρuj ur
⇒ + = 0,
∂t uLr ∂xj L
ρ u ∂ρ ∂ρu
r r j
⇒ + = 0,
L ∂t ∂xj
∂ρ ∂ρuj
⇒ + = 0.
∂t ∂xj
A.1.2 Momentum

+ = − + d, (A.17)
∂ρr ρur ui ∂ρr ρu2r ui uj ∂pρr u2r + 6 pr µr ur ∂τij
⇒ + = − + ,
∂t uLr ∂xj L ∂xi L L ∂xj L

ρr u2r ∂ρui ∂ρui uj
⇒ +
L ∂t ∂xj
2

ρ r ur ∂p µr ur ∂τij
=− + ,
L ∂xi L ∂xj L
∂ρui ∂ρui uj ∂p µr ur L ∂τij
⇒ + =− + ,
∂t ∂xj ∂xi L ρr u2r ∂xj L
⇒ + =− + .
A.1.3 Species
!
∂ρd Ykd ∂ρd Ykd udj ∂ ∂Y d
+ d
= ρd Dkd kd + ω̇kd , (A.18)
∂td ∂xj ∂xdj ∂xj
∂ρr ρYr Yk ∂ρr ρYr Yk ur uj
⇒ +
∂t uLr ∂xj L

∂ µµr ∂Yr Yk ρr ur Yr
= + ω̇k ,
∂xj L Sck ∂xj L L

ρr ur Yr ∂ρYk ∂ρYk uj
⇒ +
L ∂t ∂xj

µr Yr ∂ ∂Yk ρr ur Yr
= µ + ω̇k ,
Sck L2 ∂xj ∂xj L

∂ρYk ∂ρYk uj L µr Yr ∂ ∂Yk
⇒ + = µ + ω̇k ,
∂t ∂xj ρr ur Yr Sck L2 ∂xj ∂xj

⇒ + = µ + ω̇k .
∂t ∂xj Sck Re ∂xj ∂xj
A.1.4 Equation of state
pd = ρ d R d T d , (A.19)
Ru
⇒ pρr u2r + pr = ρr ρ Tr T,
W Wr
pρr u2r + pr ρT
⇒ = ,
ρr Rr Tr W
pγu2r ρT
⇒ +1= ,
γRr Tr W
ρT
⇒ γMr2 p + 1 = .
W
A.1.5 Energy
Term 1
∂ρd E d ∂ρr ρcv Tr T + 12 ρr ρu2r ui ui

= =
∂td ∂t uLr

pr ρT
Rr
∂ ρr ρ W (γ−1) Tr T + 12 ρr ρu2r ui ui ∂ (γ−1) W
+ 1
2
γp r M 2
r ρu i u i
L
= L
=
∂t ur ∂t ur

pr
u ∂ (γ−1) [γMr2 p+ 6 1] + 21 γpr Mr2 ui ui
r
=
L ∂t

γpr ur
∂ p + (γ−1)2
ρui ui
2
Mr . (A.20)
L (γ − 1) ∂t
Term 2

∂ ρd E d + pd udj ∂ ρr ρcv Tr T + 21 ρr ρu2r ui ui + pρr u2r + pr uj ur
= =
∂xd ∂xj L

ρr Rr Tr ρT 1 2 2
∂ (γ−1) W + 2 ρr ur ρui ui + pρr ur + pr uj ur
=
∂xj L

pr
∂ (γ−1)
(γMr2 p + 1) + 12 pr γMr2 ρui ui + ppr γMr2 + pr uj ur
=
∂xj L

∂ (γM 2 p + 1) + (γ−1)

p r ur r 2
γMr2 ρui ui + p (γ − 1) γMr2 + (γ − 1) uj
=
L (γ − 1) ∂xj

∂ γM 2 p [1 + (γ − 1)] + (γ−1)

p r ur r 2
γMr2 ρui ui + 1 + (γ − 1) uj
=
L (γ − 1) ∂x

∂ γM 2 pγ + (γ−1) γMr2

p r ur r 2
γMr2 ρui ui + Mr2
uj
=
L (γ − 1) ∂xj

(γ−1) 1
γMr2 pr ur ∂ γp + 2 ρui ui + Mr2 uj

=
L (γ − 1) ∂xj
 
(γ−1)

γpr ur
∂ γp + 2
ρu u
i i u j ∂uj 
Mr2 + . (A.21)
L (γ − 1) ∂xj ∂xj
L(γ−1)
Terms 3, 4, 5, and 6 are multiplied by γpr ur
. This factor is obtained from
terms 1 and 2.
Term 3

L (γ − 1) µr u2r ∂τij ui (γ − 1) u2r µr ∂τij ui
2
=
γpr ur L ∂xj γRr Tr ρ r ur L ∂xj
2
(γ − 1) Mr ∂τij ui
= . (A.22)
Re ∂xj
Term 4
! !
L (γ − 1) Tr µr ∂ cdp ∂T 1 Tr µr ∂ (γ − 1) cdp ∂T
µ = µ
γpr ur L2 ∂xj P r ∂xj ρr Rr Tr ur L ∂xj γ P r ∂xj
!
µr ∂ Rr cdp ∂T 1 ∂ µ ∂T
= µ = . (A.23)
ρr ur L ∂xj W cdp Rr P r ∂xj ReP r ∂xj W ∂xj
Term 5
N
L (γ − 1) cp,r Tr ρr ur Yr X
Qk ω̇k
γpr ur Yr L k=1
N
LRr cp,r Tr ρr ur Yr X
= Qk ω̇k
cp,r ρr Rr Tr ur Yr L k=1
N
X
= Qk ω̇k . (A.24)
k=1
Term 6
N
!
L (γ − 1) µr Yr cp,r Tr ∂ X µ ∂Yk
− hs,k
γpr ur L2 Yr ∂xj k=1
Sck ∂xj
N
!
Rr µr cp,r Tr ∂ X µ ∂Yk
=− hs,k
cp,r ρr Rr Tr ur L ∂xj
k=1
Sck ∂xj
N
!
1 µr ∂ X µ ∂Yk
=− hs,k
ρr ur L ∂xj k=1 Sck ∂xj
N
!
1 ∂ X µ ∂Yk
=− hs,k . (A.25)
Re ∂xj k=1 Sck ∂xj
Therefore, the non–dimensional governing equations are:
∂ρ ∂ρuj
+ = 0, (A.26)
∂t ∂xj

+ = µ + ω̇k , (A.27)
+ =− + , (A.28)

∂ γ−1 ∂ γ−1
Mr2 p+ ρui ui + γp + ρui ui uj
∂t 2 ∂xj 2

∂uj (γ − 1)Mr2 ∂τij ui 1 ∂ µ ∂T
+ = +
∂xj Re ∂xj ReP r ∂xj W ∂xj
N N
!
X 1 ∂ X µ ∂Yk
+ Qk ω̇k + hs,k (A.29)
k=1
Re ∂xj k=1 Sck ∂xj
ρT
= γMr2 p + 1. (A.30)
W
The zero Mach number reacting equations from Majda & Sethian [31] are
obtained when Mr approaches zero. Note that term 6 in the energy equation
is not implemented in this work. This term is usually negligible compared to

PN
k=1 Qk ω̇k (Poinsot & Veynante [42]).
Appendix B
Flamelet modeling
The purpose of this appendix is to develop flamelet modeling capability within

the context of the numerical methodology. The motivation is to simulate
hydrocarbons. For hydrogen/air, the number of species are generally around 9
to 13 species, but for hydrocarbons the number of species are in the hundreds.
This provides inspiration to develop and study flamelet modeling within the
numerical methodology presented in this dissertation.

The references that are used for flamelet modeling are from Poinsot & Vey-
nante [42], Claramunt et al. [9] and Pitsch [41]. The structure of non–premixed
flames can be described by means of a conserved scalar. The transport equa-
tion for the mixture fraction is defined as ([9] & [42]):

∂ρζ ∂ρζuj ∂ ∂ζ
+ = ρDζ . (B.1)
∂t ∂xj ∂xj ∂xj
The conserved scalar accounts for mixing of fuel and oxidizer where the range
of the conserved scalar is from zero to one. In reacting flows, the Lewis number
λ
is defined as: Lek = ρDk cp
, which is assumed to be constant. This parameter
92
93
compares the diffusion speeds of the heat and species.

In general, a diffusion flame can be viewed as a thin locally one–dimensional
structure where one views each structure as a small laminar flame which is
called a ‘flamelet’. The flamelet equations are written as ([9]):

∂Yk ρχ ∂ 2 Yk 1 1 cp ∂ λ ∂Yk
ρ = + −1 ρχ + ω̇k , (B.2)
∂t 2Lek ∂ζ 2 4 Lek λ ∂ζ cp ∂ζ
N
∂T ρχ ∂ 2 T ρχ ∂cp ∂T ρχ X cp,k ∂Yk ∂T
ρ = 2
+ +
∂t 2 ∂ζ 2cp ∂ζ ∂ζ 2cp k=1
Lek ∂ζ ∂ζ
N
1 X QR
− hk ω̇k + , (B.3)
cp k=1 cp
∂ζ ∂ζ
where χ = 2Dζ ∂xj ∂xj
. Temperature, species, specific heat, enthalpy, rate of
production and heat loss are T , Yk , cpk , hk , ω̇k and QR , respectively. Note
that k denotes the kth species. The simplified flamelet equations are:
∂Yk ρχ ∂ 2 Yk
ρ = + ω̇k , (B.4)
∂t 2Lek ∂ζ 2
N
∂T ρχ ∂ 2 T 1 X QR
ρ = 2
− hk ω̇k + . (B.5)
∂t 2 ∂ζ cp k=1 cp
The flamelet code currently implements the simplified equations.
The discrete flamelet equations in a ‘finite volume approach’ for the simple
equations are defined as:
1 1
t+1
Yk,i t
− Yk,i χ X ∂Yk t+ 2 t+ 2
ω̇k
= Nj Af aces + , (B.6)
∆t 2V Lek f aces ∂ζ f aces ρ
t+ 21 N
!t+ 21
T t+1 − T t χ X ∂T 1 X
= Nj Af aces + hk ω̇k .(B.7)
∆t 2V f aces ∂ζ f aces ρcp k=1
Applying central difference in space and time, the final form of the discrete
94
equations are:

χ ∆t t+1 χ ∆t t+1 χ ∆t t+1
1+ Yk,i − Yk,i+1 − Y (B.8)
2Lek ∆ζ 2 4Lek ∆ζ 2 4Lek ∆ζ 2 k,i−1
t+ 12
t χ ∆t t t t
ω̇k
= Yk,i + 2
Yk,i+1 − 2Yk,i + Yk,i−1 + ∆t
4Lek ∆ζ ρ i

χ ∆t χ ∆t t+1 χ ∆t t+1
1+ Tit+1 − T − T (B.9)
2 ∆ζ 2 4 ∆ζ 2 i+1 4 ∆ζ 2 i−1
N
!t+ 21
χ ∆t 1 X
= Tit + T t − 2Yk,it t
+ Ti−1 + ∆t hk ω̇k .
4 ∆ζ 2 i+1 ρcp k=1
i
Note that temporal and spatial derivative is the Crank–Nickolson method

which is second order in space and time ([2]). A good problem to test the
method is the following equation.
∂Y ∂2Y
− =0 (B.10)
∂t ∂ζ 2
for 0 ≤ ζ ≤ 1 and 0 ≤ t, with boundary conditions Y (0, t) = Y (1, t) = 0 and

2
the initial condition Y (ζ, 0) = sin πζ. The solution is Y (ζ, t) = e−π t sin πζ.
Figure B.1 (a) shows the spatial order of the method. Note that second–order
accuracy is obtained.
The source term in the species equation is similar to the implicit method
used by Doom et al. [13]. Recall that the source term is:
M
" N ν ′ #
X Y ρYk kj
ω̇k = Wk νkj Kf j
j=1 k=1
Wk
M
" N νkj
′′
#
X Y ρYk
−Wk νkj Krj . (B.11)
j=1 k=1
Wk
95
error YH2 T
δζ (a) time
Figure B.1: (a) flamelet code, second order. (b) Comparison of a major
species & temperature between Chemkin and present algorithm for a well–
stirred reactor. ([YH2 ] flamelet code, △ Chemkin & [T ] flamelet
code, Chemkin).
The discrete form of the source term is:

 !νkj
′

M N
t+ 12
X t+ 12
Y ρt+1 t+1
k,i Yk,i + ρtk,i Yk,i
t
ω̇k,i = Wk νkj Kf j,i 
j=1 k=1
2Wk
 !νkj
′′

M N
X t+ 12
Y ρt+1 t+1 t t
k,i Yk,i + ρk,i Yk,i
−Wk νkj Krj,i . (B.12)
j=1 k=1
2Wk
Complex chemistry is validated using the homogeneous mixture or well–

stirred reactor problem. The chemical mechanism is a 11 species, 23 reaction
mechanism from Turns [51]. The species are H2 , O2 , OH, O, H, H2 O, HO2 ,

H2 O2 , N2 , N O, and N . Figure 4.4 (b) is a comparison between Chemkin [26].
Good agreement is observed for the homogeneous mixture.
The flamelet library is illustrated in figure B.2. All ‘chemistry’ variables

are stored in this flamelet library (Yk , T , ρ, ω̇n and etc....). The flamelet library
gives a direct mapping between ζ and the chemistry variables. Therefore, the
96
T χ
(a) ζ (b) ζ
YN O
YOH
(c) ζ (d) ζ
Figure B.2: Temporal evolution of mixture fraction versus (a) T , (b) χ, (c)
YOH , and (d) YN O .
97
(a) (b)
(c) (d)
Figure B.3: Contours of (a) |ω|, (b) T , (c) YH2 O , and (d) YH .
scalar equation (ζ) in the fluid code looks up the chemistry variables based
upon the mixture fraction.
The flamelet library is applied to cold H2 /air jet at stoichiometric condi-

tions. For a cold jet, the flamelet library does not directly affect the flow. The
passive scalar or mixture fraction is used to look up the chemistry variables in
the flamelet library and yields the results in figure B.3.

Bibliography
[1] Bijl, H. & Wesseling, P., 1998, A unified method for computing incompress-
ible and compressible flows in boundary–fitted coordinates. J. Comput. Phys.
141: 153–173.
[2] Burden, R. L., & Faires, J. D., 2001, Numerical Analysis. Brooks/Cole.
[3] Chen, J., Mahalingam, S., Puri, I. K., & Vervisch, L., 1992, Effect of finite-
rate chemistry and unequal Schmidt number on turbulent non-premixed flames
modeled with single-step chemistry. CTR: Proceedings of the Summer Pro-
gram. 367-387.
[4] Chen, J., Mahalingam, S., Puri, I. K., & Vervisch, L., 1992, Structure of turbu-
lent non-premixed flames modeled with two-step chemistry. CTR: Proceedings
of the Summer Program. 389–402.
[5] Chen, S.L. & Dahm, J.A., 1998, Effect of heat release in a reacting vortex ring.
Proc. Combust. Inst. 2579–2586.
[6] Chen, S.L. & Dahm, J.A., Coupling between fluid dynamics and combustion
in a laminar vortex ring. 38th Aerospace Science Meeting and Exhibit paper.
AIAA–2000–0433.
[7] Chen, S., Dahm, W. & Tryggvason, G., 2000, Effects of heat release in a
reacting vortex ring. Proc. Combust. Inst. 28: 515–520.
98
BIBLIOGRAPHY 99
[8] Cheng, T. S., Wu, C. Y., Chen, C. P., Chao, Y. H., Yuan, T.& Leu, T. S.,
2006, Detailed measurement and assessment of laminar hydrogen jet diffusion
flames. Combust. Flame. 146: 268–282.
[9] Claramunt, K., Consol, R., Carbonell, D., & Perez–Segarra C.D., 2006, Anal-
ysis of the flamelet concept for nonpremixed laminar flames. Combust. Flame.
145: 845–862.
[10] Colonius, T., Moin, P. & Lele, S.K., Direct Computation of Aerodynamic
Sound. Report No. TF-65, Department of Mechanical Engineering, Stanford
University, Stanford, California.
[11] Cuenot, B. & Poinsot, T., 1996, Asymptotic and numerical study of diffusion
flames with variable Lewis number and finite rate chemistry. Combust. Flame.
104: 111–137.
[12] Didden, N., 1979, On the formation of vortex rings: rolling up and production
of circulation. Z. Angew. Mech. 30: 101–116.
[13] Doom, J., Hou, Y. & Mahesh, K., A numerical method for DNS/LES of tur-
bulent reacting flows. J. Comput. Phys. 226: (2007) 1136–1151.
[14] Doom, J. & Mahesh, K., DNS of reacting H2 /air laminar vortex rings. 46th
Aerospace Science Meeting and Exhibit paper. AIAA–2008–0508.
[15] Echekki, T. & Chen, J.H., 2003, Direct numerical simulation of auto–ignition
in non–homogeneous hydrogen–air mixtures. Combust. Flame. 134: 169–191.
[16] Echekki, T. & Ferziger, J., 1993, A simplified reaction rate model and its
application to the analysis of premixed flames. Combust. Sci. Tech. 89: 293–
351.
BIBLIOGRAPHY 100
[17] Gharib, M., Rambod, E. & Shariff, K., 1998, A universal time scale for vortex
ring formation. J. Fluid Mech. 360: 121–140.
[18] Glezer, A., 1988, The formation of vortex rings. Phys. Fluids. 31: 3532–3542.
[19] Hewett, J.S. & Madnia, C.K., 1998, Flame–vortex interaction in a reacting
vortex ring. Phys. Fluids. 10: 189–205.
[20] Hilbert R., & Thevenin D., 2002, Autoignition of turbulent non–premixed
flames investigated using direct numerical simulations. Combust. Flame. 128:
22–37.
[21] Hilbert R., & Thevenin D., 2004, Influence of differential diffusion on maximum
flame temperature in turbulent nonpremixed hydrogen/air flames. Combust.
Flame. 138: 175–187.
[22] Hirschfelder, J. O., Curtiss, C. F. & Byrd, R. B., 1969, Molecular theory of
gases and liquids. John Wiley & Son.
[23] Hou, Y., & Mahesh, K., 2005, A robust, colocated, implicit algorithm for
direct numerical simulation of compressible, turbulent flows. J. Comput. Phys.
205: 205–221.
[24] Im, H. G., Chen, J. H. & Law, C. K., 1998, Ignition of hydrogen mixing layer
in turbulent turbulent flows. 27th (Int.) Symp. on Comb. 1047–1056.
[25] Katta, V. R., Goss, L. P. & Roquemore, W. M., 1994, Effects of non–unity
Lewis number and finite–rate chemistry on the dynamics of hydrogen–air jet
diffusion flame. Combust. Flame. 94: 60–74.
[26] Kee, R. J., Rupley, F. M., Miller, J. A., Coltrin, M. E., Grcar, J. F., Meeks,
E., Moffat, H. K., Lutz, A. E., Dixon-Lewis, G., Smooke, M. D., Warnatz,
J., Evans, G. H., Larson, R. S., Mitchell, R. E., Petzold, L. R., Reynolds, W.
BIBLIOGRAPHY 101
C., Caracotsios, M., Stewart, W. E., Glarborg, P., Wang, C., McLellan, C. L.,
Adigun, O., Houf, W. G., Chou, C. P., Miller, S. F., Ho, P., Young, P. D.,
Young, D. J., Hodgson, D. W., Petrova, M. V., & Puduppakkam, K. V., 2006,
CHEMKIN Release 4.1, Reaction Design, San Diego, CA.
[27] Lele, S. K., 1992, Compact finite difference schemes with spectral like resolu-
tion. J. Comput. Phys. 103: 16–42.
[28] Maas, U. & Warnatz, J., 1988, Ignition processes in hydrogen/oxygen mixtures.
Combust. Flame. 74: 53–69.
[29] Mahalingam, S., Chen, J., & Vervisch, L., 1995, Finite-rate chemistry and tran-
sient effects in direct numerical simulations of turbulent non-premixed flames.
[30] Mahesh, K., Constantinescu, G. & Moin, P., 2004, A numerical method for
large-eddy simulation in complex geometries. J. Comput. Phys. 197: 215–240.
[31] Majda, A. & Sethian, J.A., 1985, The derivation and numerical solution of
the equations for zero Mach number combustion. Combust. Sci. Technol. 42:
185–205.
[32] Mastorakos, E., Baritaud, T. A. & Poinsot, T.J., 1997, Numerical simulations
of auto–ignition in turbulent mixing flows. Combust. Flame. 109: 198–223.
[33] Maxworthy, T., 1972, The structure and stability of vortex rings. J. Fluid
Mech. 51: 15–32.
[34] Maxworthy, T., 1977, Some experimental studies of vortex rings. J. Fluid Mech.
81: 465–495.
[35] Miller, J.A. & Bowman, C.T., 1989, Prog. Energy Combust. Sci. 15: 287–338.
BIBLIOGRAPHY 102
[36] Montgomery, C. J., G. Kosaly, and J. J. Riley., 1997, Direct numerical simula-
tion of turbulent non-premixed combustion with multi-step hydrogen-oxygen
kinetics, Combust. Flame. 109: 113–144.
[37] Mueller, M.A., Kim, T.J., Yetter, R.A. & Dryer, F.L., Flow Reactor Studies
and Kinetic Modeling of the H2/O2 Reaction. Int. J. Chem. Kinet. 31: 113–
125.
[38] OConaire M., Curran H.J., Simmie J.M. & Pitz W.J., Westbrook C.K., 2004,
Int. J. Chem. Kinet. 36: 603–622.
[39] Pantano, C., 2004, Direct simulation of non-premixed flame extinction in a
methane-air jet with reduced chemistry. J. Fluid Mech. 514: 231–270.
[40] Pember, R. B., Howell, L. H., Bell, J. B., Colella, P., Crutchfield, W. Y.,
Fiveland, W. A., & Jesse, J, P., 1997, An adaptive projection method for the
modeling of unsteady, low-Mach number combustion. Western States Section
of the Combustion Institute. 1–25.
[41] Pitsch, H., 2000, Unsteady flamelet modeling of differential diffusion in turbu-
lent jet diffusion flames. Combust. Flame. 123: 358–374.
[42] Poinsot, T., & Veynante, D., 2005, Theoretical and Numerical Combustion.
Edwards.
[43] Rutland, C. J. & Ferziger, J. H., 1990, Unsteady strained premixed laminar
flames. Combust. Sci. Tech. 73: 305–326.
[44] Rutland, C. J. & Ferziger, J. H., 1991, Simulations of flame-vortex interactions.

BIBLIOGRAPHY 103
[45] Safta, C. & Madnia, C. K., 2006, Auto–ignition and structure of non–premixed
CH4 /H2 flames: Detailed and reduced kinetic models. Combust. Flame. 144:
64–73.
[46] Sau, R. & Mahesh, K., 2007, Passive scalar mixing in vortex rings. J. Fluid
Mech. 582: 449–461.
[47] Shariff, K. & Leonard, A., 1992, Vortex rings. Annu. Rev. Fluid. Mech. 24:
235–279.
[48] Smith, G.P., Golden, D.M., Frenklach, M., Moriarty, N.W., Eiteneer, B., Gold-
enberg, M., Bowman, C.T., Hanson, R.K., Song, S., Gardiner, W.C., Lissianski,
V.V. & Qin, Z., GRI-Mech homepage, http://www.me.berkeley.edu, 1999.
[49] Thompson, P.A., 1988, Compressible-fluid dynamics. McGraw–Hill.
[50] Trouve, A., & Poinsot, T., 1994, The evolution equation for the flame surface
density in turbulent premixed combustion. J. Fluid Mech. 278: 1–31.
[51] Turns, S.R., 2000, An introduction to combustion. McGraw Hill.
[52] Van der Heul, D.R., Vuik, C. & Wesseling, P., 2002, A conservative pressure-
correction method for the Euler and ideal MHD equations at all speed. Intnl.
J. Num. Methods Fluids. 40: 521–529.
[53] Wall, C., Pierce, C. D., & Moin, P., 2002, A semi–implicit method for resolution
of acoustic waves in low Mach number flows. J. Comput. Phys. 181: 545–563.
[54] Williams, F.A., 1985, Combustion Theory. Westview Press.

Jeff Thesis

Diunggah oleh

Informasi Dokumen

Deskripsi Asli:

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Jeff Thesis

Diunggah oleh

Hak Cipta:

Format Tersedia

Direct numerical simulation of turbulent,

chemically reacting flows

Jeffrey Joseph Doom

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

Krishnan Mahesh, Advisor

Aerospace Engineering and Mechanics Department for providing an enjoyable

most important skills of all, good work ethic and patience.

at the University of Minnesota. Computing resources were provided by the

and the National Center for Supercomputing Applications.

hydrogen/air turbulent diffusion flame.

over a range of Mach numbers. The compressible Navier–Stokes equations are

algorithm discretely conserves kinetic energy in the incompressible, inviscid,

for stiff chemical mechanisms. The algorithm is readily applicable to complex

The algorithm is used to study auto–ignition in laminar vortex rings. A

temperature regions with low scalar dissipation at a ‘most reactive’ mixture

fraction, ζM R (Mastorakos et al. [32]). Subsequent evolution of the flame is

to predict both the initial auto–ignition as well as subsequent evolution. For

Mueller et al. [37] mechanism. Isotropic turbulence is superimposed on an

flame where auto-ignition occurs at ζM R , the turbulent flame auto–ignites over

two–dimensional simulations (Mastorakos et al. [32]) show that when flame

due to the three–dimensionality of the flow.

3.1.2 Laminar diffusion flame . . . . . . . . . . . . . . . . . 28

4 Auto–ignition of a hydrogen vortex ring reacting with hot air 42

5 Turbulent H2 /air diffusion flames 67

A Non–Dimensionalizing the Navier–Stokes equations 84

3.1 Parameters for laminar diffusion flame with one–step chemistry. 29

4.1 Inlet boundary conditions for reacting vortex ring. . . . . . . . 44

5.1 References variables for laminar H2 /air diffusion flame. . . . . 69

2.1 Storage of variables. . . . . . . . . . . . . . . . . . . . . . . . 15

3.1 Comparison of computed temperature to analytical solution for

4.1 Schematic of reacting vortex ring. . . . . . . . . . . . . . . . . 43

4.2 Comparison using one–step chemistry to Hewett and Madnia

5.1 Schematic of initial conditions for the turbulent diffusion flame. 68

5.7 Evolution of YOH from a non–dimensional time of 0 to 4. . . . 74

B.1 Comparison of a major species & temperature between Chemkin

x, y, z Coordinate axes, coordinates

Ru Universal gas constant

Dk Diffusion coefficient of the k th species

cv Specific heat at constant volume

ω̇k Mass reaction rate for the k th species

Subscripts and Superscript

RANS Reynolds–averaged Navier–Stokes

1.1 Numerical method

The direct numerical simulation (DNS) and large-eddy simulation (LES) of

turbulent reacting flows are extremely challenging. Combustion involves a

chemical reactions possess different time-scales, and different reaction zone

numerical stiffness because of a large difference between the speed of sound

their attendant problems. Desirable requirements for an algorithm to perform

waves/detonation. This dissertation proposes an approach that deals with

[44], Pember et al [40], and Mahesh et al. [30]). A second–order, staggered,

are not applicable to finite Mach number flows.

are non-dimensionalized using an incompressible scaling for pressure, and the

pressure–correction approach is used to constrain the divergence of the velocity

kinetic energy in the limit of incompressible, inviscid flow. These features

1.2 Auto–ignition of hydrogen vortex ring re-

acting with hot air

The motion of a fluid column of length L through an orifice of diameter D

a ring of gaseous fuel into hot oxidizer.

Chen et al. [5]–[7] have performed experiments on reacting vortex rings of

top–hat (U0 ), and ring circulation is estimated from hot–wire measurements

also perform simulations of methane/air vortex rings using a steady laminar

to heat release is shown to be responsible for this behavior. The experiments

used a one–step Arrhenius mechanism to model combustion of a hydrocarbon