0% found this document useful (0 votes)

10 views95 pages

Gaseous Shocks and Detonations: Nomenclature

This document provides definitions and notations for dimensional and nondimensional quantities used in the analysis of gaseous shocks and detonations. It defines over 100 variables, parameters, numbers, and abbreviations with their associated descriptions and standard SI units. The key concepts covered are variables related to mass, length, time, temperature, density, velocity, pressure, heat, viscosity, conductivity, and more. Nondimensional groups like Mach number, Reynolds number, and reduced variables used in analyses are also defined.

Uploaded by

Joaquim Valente Nogueira

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

10 views95 pages

Gaseous Shocks and Detonations: Nomenclature

Uploaded by

Joaquim Valente Nogueira

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 95

4

Gaseous Shocks and Detonations

Nomenclature
Dimensional Quantities
Description S.I. Units
a Mean molecular velocity. Sound speed m s−1
cp Specific heat at constant pressure J K−1 kg−1
cv Specific heat at constant volume J K−1 kg−1
d Thickness (of a structure) m
D Normal propagation velocity m s−1
D Molecular diffusivity m2 s−1
DT Thermal diffusivity m2 s−1
E Activation energy J molecule−1
E Energy J
h Enthalpy J kg−1
k Transverse wavenumber m−1
kB Boltzmann’s constant J K−1
l Longitudinal evolution parameter, see (12.1.11) m−1
Molecular mean free path m
L A length scale m
m Mass flux kg m−2 s−1
p Pressure Pa
qm Heat of combustion per unit mass J kg−1 ≡ (m/s)2
qv Heat of combustion per unit volume J m−3
r Radius. Radial coordinate m
R Radius of tube or combustion chamber m
s Entropy J K−1 kg−1
s Real part of growth rate s−1
S Surface area m2
t Time s
tN Reaction time at Neumann state, τr (TN ) s

210

T Temperature K
T∗ Crossover temperature K
u Mean velocity. Longitudinal velocity m s−1
UL Laminar flame speed m s−1
Ub Laminar flame speed relative to burnt gas m s−1
v Velocity in laboratory frame m s−1
ve Turnover velocity of a vortex m s−1
vp Piston velocity m s−1
v Specific volume 1/ρ m3 kg−1
w Transverse velocity m s−1
Ẇ Reaction rate kg m−3 s−1
x, y, z Coordinates m
x Self-similar variable x/t, r/t
α Local position of front m
δ A characteristic thickness m
μ Shear viscosity Pa s
ν Viscous diffusivity μ/ρ m2 s−1
λ Thermal conductivity J s−1 m−1 K−1
ρ Density kg m−3
σ (Complex) growth rate of perturbation s−1
τ A characteristic time s
τr Reaction time s
ω Angular frequency s−1
Heat release rate s−1

Nondimensional Quantities and Abbreviations

b Temperature sensitivity parameter, see (4.5.21)
cst. Constant
K see (4.3.40) and (4.3.47)
lN Length scale of heat release, see (4.5.26)
m Reduced mass flux through shock, see (4.5.3)
M Mach number u/a
n Parameter characterising strength of shock, see (4.4.4)
O(.) Of the order of
Pr Prandtl number μ/(ρDT )
P Normalised pressure, see (4.2.11)
qN Reduced heat of reaction of detonation, see (4.5.11)
Q Normalised heat of reaction, see (4.2.42)
r Parameter characterising the fluid, see (4.4.4)
Re Reynolds number u l/ν

s Degree of folding, relative increase of surface S /So

S Reduced complex growth rate, see (4.4.3)
Sa, Si Parts of reduced growth rate, see (4.5.31)
t Reduced time, see (4.5.1)
V Normalised specific volume, see (4.2.11)
ẇ Reduced reaction rate tN Ẇ
x Reduced mass-weighted coordinate, see (4.5.1)
βN Reduced activation energy of detonation E/kB TN
γ Ratio of specific heats cp /cv
A small parameter
ζ Reduced transverse coordinate, see (4.4.40) z/L
η Reduced transverse coordinate, see (4.4.40) y/L
ε A second small parameter
Reduced temperature βN (T − T N )/T N
κ Reduced wavenumber, see (4.5.29)
ν Reduced specific volume (v − vu )/vu
ξ Reduced length x/, r/(Dt)
π Reduced pressure (p − pu )/pu
σ (In Section 4.5) reduced growth rate, see (4.5.29)
φ Reduced amplitude of wrinkling, see (4.4.40) α/(L)
ψ Reaction progress variable (heat release), see (15.1.42)
Distribution of reduced reaction rate
CJ Chapman–Jouget
DDT Deflagration-to-detonation transition
ZND Zeldovich, von Neumann and Döring

Superscripts, Subscripts and Math Accents

a∗ A critical or particular value
a1 Value in upstream (unshocked) flow
a(a) Compressible (acoustic) part of flow
a(i) Incompressible part of flow
ay ∂a/∂y
ȧt ∂a/∂t
ab Burnt gas
ac Critical value
acoll Collisions
aCJ Chapman–Jouget conditions
ae Excitation or perturbation flow (vortex, etc.)
af Flame front
aind Induction (time or length)

aint Interaction (time)

aL Laminar flame
aN Neumann state (behind a shock)
ao Initial, or unperturbed or reference value
ap On the piston
ar Reaction rate
atur Turbulent
au Fresh gas
a Average value or unperturbed value
ã Fourier component of a, a(y, t) = ã(t)eiky
â Amplitude of linear harmonic perturbation, a(y, t) = âeiky+σ t
â (In Section 4.5) Dimensional quantities
ă Laplace transform of a

4.1 Introductory Remarks

Detonations are supersonic waves of combustion. They were briefly presented in Section
1.2.5. The difference of time scales between the rate of elastic collisions and that of inelastic
collisions producing the heat release makes the structure of a gaseous detonation decom-
pose into an inert shock wave (lead shock) followed by an exothermic reaction zone.[1]
Shock waves are thin regions of strong gradients propagating at supersonic speed in an inert
fluid. As recalled in Section 15.1.7, the internal structure of shock waves is controlled by
elastic collisions and their thickness is microscopic, a few tens of mean free paths in gases,
except for weak shocks (propagating at a velocity close to that of sound) that are thicker;
see Section 4.2.2. Ordinary shock waves appear as discontinuities in the solutions to Euler’s
equations. Their formation is discussed in Section 15.3. The exothermic reactions are
produced in gaseous mixtures by inelastic collisions that are much less frequent than elastic
collisions, a phenomenon that can be represented by an Arrhenius law with an activation
energy larger than the thermal energy; see Section 1.2.2. The thickness of the reaction
zone of gaseous detonations is thus macroscopic, ranging from several millimetres to one
centimetre in normal conditions, see Section 1.2.5. Unlike inert shocks, there is a regime
of autonomous propagation of detonations, without external support such as a piston or a
projectile. This self-propagating regime, with a constant velocity of propagation, is called
the Chapman–Jouguet (CJ) regime. Detonations are observed not only in gaseous mixtures,
but also in solid and liquid explosives. In contrast to gaseous detonations, detonations in
condensed phases are microscopically thin. They will not be studied in this book.
The study of gaseous detonations was initiated during the nineteenth century,[2,3,4] but
the multidimensional dynamics of the front was not understood until much later and is

[1] Vieille P., 1900, C. R. Acad. Sci. Paris, 131, 413.

[2] Rankine W., 1870, Philos. Trans. R. Soc. London, 160, 277–288.
[3] Hugoniot P., 1889, Journal de l’École Polytechnique, 58(1), 1–125.
[4] Rayleigh J., 1910, Proc. R. Soc. London, 84, 247–284.

Initial state Shocked gas

Figure 4.1 Sketch of flow through a shock wave. The flow velocities are given in the reference frame
of the shock.

still under investigation. The basic knowledge of shock waves[1] and detonations[2] is
summarised in Section 4.2, while the general theory of one-dimensional compressible flows
is recalled in Section 15.3. The results of the two last decades are presented in Sections 4.3–
4.5 where the emphasis is put on simple theoretical analyses based on physical insights.
Technical details of the more complicated analyses are presented in Chapter 12.

4.2 Planar Supersonic Waves

In this section attention is limited to planar waves propagating at constant velocity in a sim-
ple fluid. The multidimensional dynamics resulting from intrinsic instabilities is discussed
later. We start with planar shock waves supported by a piston in an inert fluid.

4.2.1 Shock Waves: Rankine–Hugoniot Relations

In the steady planar case, a shock wave separates two homogeneous and stationary flows.
Its velocity depends on the thermodynamic state of the upstream fluid and the velocity
of the driving piston, regardless of the dissipative mechanisms that control its internal
structure. For a given velocity of the piston, the propagation speed and the density and
pressure of the shocked gas are obtained using the jump relations for the conservation of
mass, longitudinal momentum and total energy in (15.1.45)–(15.1.48), where the thermal
gradients are neglected in the external regions on both sides of the shock. The propagation
velocity D of a planar shock wave supported by a piston moving at constant speed vp in
a medium initially at rest (density ρu and and pressure pu ) is obtained as follows. In the
reference frame of the shock wave, fluid of density ρu enters the wave with velocity D in
the direction normal to the plane of the shock. After crossing the shock, the fluid density
and velocity are different; see Fig. 4.1. Let uN be the speed at which the ‘shocked’ fluid
leaves the wave, denoted by UN in Section 1.2.5 and in (15.1.45)–(15.1.48). It flows in the
normal direction, and the state of shocked gas is called the ‘Neumann state’ with density
ρN and pressure pN . According to conservation of mass, the mass flow rate m through the
wave is constant,
m ≡ ρu D = ρN uN . (4.2.1)

[1] Landau L., Lifchitz E., 1986, Fluid mechanics. Pergamon, 1st ed.
[2] Zeldovich Y., Kompaneets A., 1960, Theory of detonation. Academic Press.

The wave propagates at speed D in the laboratory frame where the upstream gas is at rest,
and the shocked gas velocity is equal to the piston velocity vp in this frame, so that

D − uN = vp . (4.2.2)

The state of shocked gas (ρN , pN ) is found as a function of m from conservation of longitu-
dinal momentum (15.1.46) and energy (15.1.51). With the help of (4.2.1) it is then possible
to express D and uN as functions of m, which can finally be obtained as a function of vp
using (4.2.2).

Case of an Arbitrary Fluid

Momentum conservation (15.1.46) can be written pu + m2 /ρu = pN + m2 /ρN , or

1 1
pN − pu = −m 2
− . (4.2.3)
ρN ρu
Energy conservation (15.1.51) in the absence of heat release (qm = 0) and external fluxes,

hN − hu + (u2N − D 2 )/2 = 0, (4.2.4)

can be written, using (4.2.1), D 2 = m2 /ρu2 , u2N = m2 /ρN2 , in the form

. /
m2 1 1
hu − hN = − 2 ,
2 ρN2 ρu

where hu and hN are the enthalpies in the initial and shocked states. Eliminating m with the
help of (4.2.3) gives

1 1 1
h(ρu , pu ) − h(ρN , pN ) + + (pN − pu ) = 0. (4.2.5)
2 ρu ρN
When the enthalpy h(ρ, p) is expressed as a function of density and pressure using the
thermodynamics of the fluid, the Hugoniot relation,

1 1 1
h(ρu , pu ) − h(ρ, p) + + (p − pu ) = 0, (4.2.6)
2 ρu ρ
defines a curve in the (1/ρ, p) plane that goes through the point (1/ρu , pu ). In this same
plane, momentum conservation (4.2.3),

1 1
p − pu = −m2 − , (4.2.7)
ρ ρu
defines a straight line with negative slope that also goes through the point (1/ρu , pu );
see Fig. 4.2. The absolute slope of this line, called the ‘Michelson–Rayleigh’ line, is
proportional to the square of the mass flux. Its intersection with the Hugoniot curve defines
the point (1/ρN , pN ) and yields the shock speed D as a function of the piston velocity vp .

Michelson–
Rayleigh line

Hugoniot
curves

Figure 4.2 Hugoniot curves and the Michelson–Rayleigh line representative of a perfect gas subject
to a strong shock propagating at Mach 6. The solid curve is the Hugoniot curve corresponding to the
initial condition U. The dotted curve is the back projection of the Hugoniot curve corresponding to the
initial condition N. They are different, although both pass through the two points; see the discussion
following (4.2.23).

Case of a Polytropic Gas

Consider now the particular case of a polytropic gas, namely an ideal gas with constant
specific heats per unit mass cp and cv , γ = cp /cv > 1,
γ p
p = (cp − cv )ρT, h = cp T = . (4.2.8)
γ −1ρ
In this case Equations (4.2.2)–(4.2.4) for mass, momentum and energy conservation can be
obtained directly from the kinetic theory of gases; see Section 4.6.1. According to (4.2.8),
the Hugoniot relation (4.2.6) can then be written as

γ p pu 1 1 1
− − (p − pu ) + = 0. (4.2.9)
γ −1 ρ ρu 2 ρu ρ
Multiplying by 2(γ − 1)ρu /pu yields

p ρu p ρu
(γ + 1) −1 −1 +2 − 1 + 2γ − 1 = 0. (4.2.10)
pu ρ pu ρ
Introducing the notations,

(γ + 1) p (γ + 1) ρu
P≡ −1 , V≡ −1 , (4.2.11)
2γ pu 2 ρ
Equation (4.2.10) multiplied by (γ +1)/4γ reduces to an equilateral hyperbola in the P–V
plane with the equation

PV + P + V = 0, (P + 1)(V + 1) = 1. (4.2.12)

This curve goes through the origin P = 0, V = 0, and its asymptotes are the lines P = −1
and V = −1. Using the speed of sound in the initial gas, along with the conservation
equations for mass, (4.2.1), the Michelson–Rayleigh line (4.2.7) yields a linear relation
between P and V,

P = −Mu2 V, (4.2.13)

where Mu ≡ D/au is the Mach number of the shock. The two relations (4.2.12) and
(4.2.13) give a quadratic equation for V. One root is zero. The other, VN = −(Mu2 −1)/Mu2 ,
corresponds to the Neumann state and permits the calculation of the state of the gas just
behind the shock as a function of the thermodynamic state of the fresh gas (pu , ρu ) and the
Mach number of propagation Mu , yielding

uN ρu (γ − 1)Mu2 + 2
= = , (4.2.14)
D ρN (γ + 1)Mu2
pN 2γ Mu2 − (γ − 1)
= , (4.2.15)
pu (γ + 1)

TN 2γ Mu2 − (γ − 1) (γ − 1)Mu2 + 2
= , (4.2.16)
Tu (γ + 1)2 Mu2
(γ − 1)Mu2 + 2
MN2 = . (4.2.17)
2γ Mu2 − (γ − 1)
These are known as the Rankine–Hugoniot relations for a polytropic gas. The last expres-
sion is a symmetrical relation between Mu2 and MN2 ,

(γ − 1)MN2 + 2
2γ Mu2 MN2 − (γ − 1)(Mu2 + MN2 ) − 2 = 0, Mu2 = . (4.2.18)
2γ MN2 − (γ − 1)

General Comments (Arbitrary Fluid)

The Hugoniot curve is tangent to the isentropic at the point (1/ρu , pu ). To see this it is
convenient to introduce the specific volume v ≡ 1/ρ and use Gibbs’ relation Tds = deT +
pdv in the form dh = Tds + v dp, where s is the entropy and eT the internal energy. A
Taylor expansion of h(s, p) around (su , vu ) limited to first order in δs = s − su , but extended
to third order in δp = p − pu , yields

1 ∂v 1 ∂ 2v
h = Tu δs + vu δp + (δp) +
2
(δp)3 + · · · . (4.2.19)
2 ∂p s 6 ∂p2 s
This can be compared with an expansion of the enthalpy h along the Hugoniot relation
(4.2.6) in powers of δp,

1 ∂v ∂v 1 ∂ 2v
h = (2vu + v )δp, where v = δs + δp + (δp)2 + · · · ,
2 ∂s p ∂p s 2 ∂p2 s

which can be written to first order in δs and third order in δp:

1 ∂v 1 ∂v 1 ∂ 2v
h = vu δp + (∂s∂p) + (δp)2 + (δp)3 + · · · . (4.2.20)
2 ∂s p 2 ∂p s 4 ∂p2 s

Comparison of (4.2.19) and (4.2.20) shows that the entropy change along a Hugoniot curve
is of third order,

1 1 ∂ 2v
δs = (δp)3 . (4.2.21)
12 Tu ∂p2 s
The Hugoniot relation (4.2.6) does not correspond to an isofunction of state since it cannot
be put in the form H (1/ρ, p) = H (1/ρu , pu ). If we choose a point 1/ρu = 1/ρu , pu = pu ,
on the Hugoniot curve (4.2.6),

1 1 1
h(ρu , pu ) − h(ρu , pu ) + + (pu − pu ) = 0, (4.2.22)
2 ρu ρu
and consider the Hugoniot curve using this point as the initial condition,

1 1 1
h(ρu , pu ) − h(ρ, p) + + (p − pu ) = 0, (4.2.23)
2 ρu ρ
the new curve is not the same as the original (4.2.6). The Hugoniot curve with the initial
condition (1/ρN , pN ),

1 1 1
h(ρN , pN ) − h(ρ, p) + + (p − pN ) = 0, (4.2.24)
2 ρN ρ
still goes through the point (1/ρu , pu ), because of the symmetry of N and U, but it is
different from (4.2.6); see Fig. 4.2.
According to the second law of thermodynamics, compressibility is a positive quantity,
−(∂ v /∂p)s > 0. In addition, compressibility usually decreases

with pressure. The coef-
ficient in Equation (4.2.21) is then positive, ∂ 2 v /∂p2 s > 0. So, according to (4.2.21),
entropy changes in the same direction as pressure on the Hugoniot curve and its first and
second derivatives are zero. The Hugoniot curve (4.2.6) is tangent to the isentropic at the
point (1/ρu , pu ) where the slope of the Hugoniot curve is thus related to the sound speed in
the unshocked mixture,

(δp/δ v )u = −ρu2 (δp/δρ)u = −ρu2 a2u ,

√
where a = (∂p/∂ρ)s . This is valid only at the initial state, and, for example the speed of
sound at the point N in Fig. 4.2 is given not by the slope of the solid curve but by that of
the dotted curve.
The conservation equations do not indicate any preferential direction of propagation.
However, there are no steady rarefaction waves that propagate at constant velocity into a
compressed medium, represented by N in Fig. 4.2, transforming the matter into a less dense
state represented by U. The only steady waves are compression waves, called shock waves,
while rarefaction waves are unsteady processes, described in Section 15.3.4; see Fig. 15.11.
In gases, this is proved by the steady solution to Boltzmann’s equation showing that, in a
steady wave, the entropy of the final state should be larger than in the initial state; see
Section 4.6.1.

Because of the concavity of the Hugoniot curves, the absolute value of the slope of
the Michelson–Rayleigh line (4.2.3), m2 , is greater than the tangent to the Hugoniot curve
(4.2.6) at the initial point (1/ρu , pu ); see Fig. 4.2 (solid curve). Anticipating that the pres-
sure and density of the shocked fluid is larger than the initial fluid (a ‘rarefaction shock’
cannot exist, as we shall see Section 4.2.2), the propagation speed of the shock wave is thus
supersonic, ρu2 D 2 > ρu2 a2u . However, the absolute value of the slope of the Michelson–
Rayleigh line is smaller than the slope of the tangent to the Hugoniot curve (4.2.24) at the
final state (1/ρN , pN ) (dotted line in Fig. 4.2), ρN2 u2N < ρN2 a2N , so the flow leaving the shock
is subsonic. In summary, the propagation speed of the shock wave is supersonic and the
flow of shocked fluid is subsonic relative to the shock wave,
D > au , uN < aN ,
with aN > au since pN > pu and hence TN > Tu .

4.2.2 Inner Structure of Weak Shock

The inner structure of shock waves is controlled by dissipative transport. Under ordinary
conditions, the thickness of a gaseous shock is of the order of a few mean free paths,
d/ = O(1); see Section 15.1.7. Its inner structure is fully out of equilibrium and must
be studied using Boltzmann’s equation, recalled in Section 13.3.1. However, the thick-
ness increases and the jumps through the shock decrease as the supersonic shock velocity
decreases. In the limit of a propagation Mach number approaching unity, the ratio d/
diverges. The precise form of the divergence is an outcome of the analysis presented below;
see (4.2.40). Anticipating that, for a very weak shock propagating at a velocity slightly
larger than (but close to) the sound speed, the thickness becomes macroscopic, d/ 1,
the approximation of local equilibrium becomes valid and the fluid mechanic equations
can be used to describe the detonation structure. The problem is reduced to solving a one-
dimensional steady transonic flow between the initial state and the shocked gas; see Section
13.3.2, The problem was first solved by Rankine[1] when the viscous effects are neglected,
keeping only heat conduction. Rankine did not know the conservation of total energy that
was introduced later by Hugoniot.[2] The temperature gradient being zero on both sides of
the wave, he wrote, ‘The integral amount of heat received must be nothing’, and he used
the balance of entropy, called ‘thermodynamic function’, in the form of the first entropy
equation in (13.3.30); see also (15.1.61), for μ = 0,
s+∞
μ = 0: mTds = d(λdT/dx), x = ±∞: dT/dx = 0 ⇒ Tds = 0, (4.2.25)
s−∞
where the initial mixture is at x = −∞: s−∞ = su and the shocked gas is at x = +∞:
s+∞ = sN . These relations are valid only for weak shocks for which, as we will see below
in (4.2.37), the total jump of entropy sN − su is small. For a given mass flux m = ρu D,

[1] Rankine W., 1870, Philos. Trans. R. Soc. London, 160, 277–288.
[2] Hugoniot P., 1889, Journal de l’École Polytechnique, 58(1), 1–125.

Rankine then integrated the first equation in (4.2.25) using the thermodynamic function
(13.2.7), s(ρ, T) where ρ = nm, the ideal gas law, p/ρ = (cp − cv )T, the conservation of
mass ρu = m > 0 and momentum p + ρu2 =cst. He obtained jump relations that do not
depend on heat conductivity, the latter controlling only the thickness of the shock, d. This
result illustrates that the jump conditions result from conserved quantities, in particular
the total energy (internal + kinetic energy of the centre of mass of a ‘macroscopic’ fluid
particle), as Hugoniot realised almost 20 years later, ignoring the work of Rankine. The
advantage of Hugoniot’s method to compute the jumps through the shock is to be easily
extended to any strength of the shock in any material.

Formulation Including Viscosity

When the viscous dissipation is taken into account,[1] a quadratic term appears in the
entropy equation (15.1.61). It is then more convenient to use the conservation of energy,
instead of the entropy equation. In the reference frame moving with the planar wave,
equations (15.1.49)–(15.1.50) for ρ, p, u and T yield
du
ρu = m, p + ρu2 − μ = cst., (4.2.26)
dx

u2 dT du
m h+ −λ − μu = cst., (4.2.27)
2 dx dx
where, in an ideal gas, T and h are proportional to p/ρ; see (4.2.8). The choice m > 0 (u >
0) corresponds to an axis oriented toward the shocked gas, namely to a wave propagating
into the initial medium at x = −∞. Using the same trick as for deriving Hugoniot’s
relations (4.2.7)–(4.2.9), Equations (4.2.26)–(4.2.27) yield two coupled Equations for p
and v ≡ 1/ρ:
dv
(p − pu ) + m2 (v − vu ) = μm , (4.2.28)
dx
γ 1 γ λ d(pv ) μm dv
(pv − pu vu ) − (p − pu )(v + vu ) = + (v − vu ) ,
γ −1 2 γ − 1 mcp dx 2 dx
(4.2.29)
that are extensions of the Rankine–Hugoniot relations, including the dissipative transports.
According to the discussion in Section 13.3.2, these equations are valid for states in local
equilibrium. As we shall see at the end of the calculation, this is the case for weak shocks,
≡ Mu − 1 1, m = ρu au (1 + ). (4.2.30)
In the limit → 0, the variations of v and p across the shock are small; the nondimensional
quantities ν ≡ (v − vu )/vu , π ≡ (p − pu )/pu are both of order . Introducing the
√
nondimensional length ξ ≡ x/, where ≡ DTu /au , DTu ≡ λ/(ρu cp ) and au = γ pu /ρu ,
are the mean free path, the thermal diffusivity and the speed of sound, respectively, the

[1] Rayleigh J., 1910, Proc. R. Soc. London, 84, 247–284.

terms dν/dξ and dπ /dξ are expected to be smaller than ν and π if the thickness of the
shock wave d is anticipated to be larger than the mean free path (local equilibrium):
d dν dπ
1: = O(), ν = O(), π = O(), = O( 2 ), = O( 2 ). (4.2.31)
dξ dξ
Using an expansion in powers of , limited to the second order, (4.2.28)–(4.2.29) reduce to
1 dν
π + (1 + 2)ν = Pr , (4.2.32)
γ dξ

γ +1 1 dπ dν
πν + π + ν = + , (4.2.33)
2γ γ dξ dξ
valid up to 2 , where Pr ≡ μ/(ρu DTu ) is the Prandtl number. When the dissipative terms in
the right-hand side are neglected, the Rankine–Hugoniot jumps are recovered, π ≈ γ4γ +1 ,
ν ≈ − γ +14
, in agreement with (4.2.14)–(4.2.15) in the limit → 0, Mu2 ≈ 1 + 2.
According to (4.2.32), the quantity ν + π/γ is of second order, ν + π/γ = O( 2 ), so
that the relations dπ /dξ ≈ −γ dν/dξ and π ν ≈ −γ ν 2 can be used in (4.2.33) valid up
to order 2 . When the expression for ν + π/γ , obtained from (4.2.32), is introduced into
(4.2.33) a first-order differential equation is obtained for ν,

dν γ +1
[(γ − 1) + Pr] = ν + 2 ν, (4.2.34)
dξ 2
where the factor [(γ − 1) + Pr] appears also in the attenuation for acoustic propagation.[2]

Thickness of Weak Shock. Irreversibility

When the initial and final state, ν = ν−∞ and ν = ν+∞ at ξ = −∞ and ξ = +∞,
respectively, are introduced into (4.2.34), the first-order differential equation describing the
decrease of the reduced specific volume ν from the initial state to the final state takes the
form
2 dν
[(γ − 1) + Pr] = (ν − ν−∞ )(ν − ν+∞ ) 0 (4.2.35)
γ +1 dξ
4
ξ = −∞: ν = ν−∞ ≡ 0, ξ = +∞: ν = ν+∞ ≡ − , (4.2.36)
γ +1
where the coefficient [(γ − 1) + Pr] is positive so that dν/dξ 0. The solution of (4.2.35)–
(4.2.36) exists only for a supersonic wave (Mu > 1, > 0, ν+∞ < ν−∞ = 0). A
planar rarefaction wave, ν+∞ > ν−∞ = 0, propagating at constant subsonic velocity
(Mu − 1 < 1, < 0) with a steady inner structure, cannot exist because, according to
(4.2.35), ν should decrease, dν/dξ 0. Rarefaction waves are necessarily unsteady;
see the self-similar solution of

centred waves in Fig. 15.11. According to (4.2.21), using
π+∞ > π−∞ and ∂ 2 v /∂p2 s > 0 in an ideal gas, the total increase of entropy through
a weak shock is a small positive quantity of order 3 . However, according to (4.2.35),

[2] Morse P., Ingard K., 1986, Theoretical acoustics. Princeton University Press.

the entropy first increases inside the shock, reaches a maximum of order 2 , and then
decreases. This is easily verified using the thermodynamic expression (13.2.8) for the
entropy of an ideal gas (s − su )/cv = ln [(1 + π )(1 + ν)γ ]. Expanded up to the second
order, (s − su )/cv = π + γ ν − (π 2 + γ ν 2 )/2γ + O( 3 ), using (4.2.31) and, according to
(4.2.32) π ≈ −γ ν + O( 2 ), the variation of entropy (s−su )/(γ cv ) is given by the left-hand
side of (4.2.33), while the right-hand side yields −(γ − 1)dν/dξ ,
(s − su ) dν
= −(γ − 1) + O( 3 ), (4.2.37)
cp dξ

where ν = O() and, according to (4.2.35), dν/dξ = O( 2 ). The fact that the entropy goes
through a maximum is shown by noticing that the function ν(ξ ) has an inflexion point,
because ν−∞ and ν+∞ are bounded and dν/dξ 0. According to (4.2.37) the entropy
inside the shock cannot be smaller than in the initial state, −(γ − 1)dν/dξ 0, s s−∞ .
However, the total entropy jump is of the following order since dν/dξ = 0 on both sides of
the shock wave, (sN − su )/cp = O( 3 ). The calculation pushed up to order
3 shows

that the
jump is positive, sN > su , in agreement with (4.2.21) for pN > pu and ∂ 2 v /∂p2 s > 0 in an
ideal gas. This is also in agreement with the entropy balance in (15.1.65), obtained under the
approximation of local equilibrium, and also, as shown in Section 4.6.1, with Boltzmann’s
equation valid also for strong shocks whose structure is fully out of equilibrium.
Introducing variables of order unity y ≡ ν/ and x ≡ ξ [(γ + 1)/2] / [(γ − 1) + Pr],
where ξ = x/(/), Equation (4.2.34) takes the nondimensional form with no parameters,
γ excepted,

dy 4
= y+ y, (4.2.38)
dx γ +1
where the roots of the right-hand side correspond to the initial gas (ξ = −∞: ν = 0) and
the shocked gas (ξ = +∞: ν = − γ +1 4
). Choosing the origin to be where y = −2/(γ +1),
4
x
4 e γ +1
y=− 4 , (4.2.39)
(γ + 1) e γ +1 x + 1

the solution shows that the thickness of a weak shock is

1 d 1
(Mu − 1 1): d = [(γ − 1) + Pr] , =O (4.2.40)
2 (Mu − 1) Mu − 1
and is effectively much larger than the mean free path for Mu → 1.

4.2.3 ZND Structure of Detonations

When a shock wave propagates in a reactive flow, the increase in temperature of the shocked
gas can be sufficient to initiate an exothermic reaction behind the shock. Under certain

Reference frame of laboratory

Detonation wave Piston

Reactive mixture Burnt gas

at rest at piston speed

Reference frame of shock wave

Figure 4.3 Flow through a planar detonation driven by a piston, presented in the laboratory frame
(upper) and in the reference frame of the shock (lower).

conditions, this reaction zone may remain ‘attached’ to the shock wave to form a detona-
tion, with a Zeldovich, von Neumann and Döring (ZND) structure, sketched in Fig. 1.9 and
computed with a detailed chemical kinetic of combustion in Fig. 4.29a.

State of the Burnt Gas. Family of Solutions

Overdriven detonations are planar waves driven by a piston; see Fig. 4.3. The jump condi-
tions are obtained in a manner similar to (4.2.5) for shock waves, but including chemical
energy in the energy conservation equation; see (15.1.48), in which u is denoted U. We
will use the enthalpy of a perfect gas with constant specific heats for simplicity. The
thermodynamic conditions at the end of chemical reaction thus satisfy cp (T − Tu ) +
(u2 − D 2 )/2 = qm ,

γ p pu 1 1 1
− − (p − pu ) + = qm , (4.2.41)
γ −1 ρ ρu 2 ρu ρ
where qm is the chemical energy released per unit mass. After multiplication by [(γ 2 − 1)/
2γ ](ρu /pu ) this equation can be rewritten in the notations of (4.2.11)–(4.2.12),
γ + 1 qm
PV + P + V = Q, where by definition, Q≡ . (4.2.42)
2 cp Tu
In the P–V plane, (4.2.42) is an equilateral hyperbola,

(P + 1)(V + 1) = 1 + Q, (4.2.43)

that lies above the Hugoniot curve (4.2.12), since Q > 0, but has the same asymptotes. The
equation for momentum conservation is identical to the nonreactive case and is represented
by the Michelson–Rayleigh line (4.2.13), where Mu = D/au is the propagation Mach
number of the detonation. The final state of the burnt gas, (1/ρb , pb ), is given by the
intersection of the Michelson–Rayleigh line and the hyperbola (4.2.41), as sketched in
Fig. 4.4.

Hugoniot after
heat release

Michelson–
Rayleigh line
CJ Michelson–
Rayleigh line

Hugoniot
of shock

Figure 4.4 Sketch of a detonation in the (v , p) plane (v ≡ 1/ρ). The curve (4.2.41) is shown in bold.
The point U is the initial state, B is the state of the burnt gas and N is the Neumann state.

For a given Mach number Mu , the quadratic equation for P, obtained by substituting
(4.2.13) in (4.2.43), has two real roots if the discriminant is positive:
(Mu2 − 1)2 − 4QMu2 > 0.
Real roots exist only for a propagation Mach number greater than a threshold value MuCJ ,
called the Chapman–Jouguet Mach number in the literature, but which was first mentioned
by Michelson in 1893,[1]
√ √
Mu MuCJ ≡ Q + Q + 1, (4.2.44)
where the + sign is imposed by the condition MuCJ > 0, MuCJ = DCJ /au . There is
no solution for a steady planar detonation propagating at a constant velocity less than
DCJ (the Michelson–Rayleigh line does not intersect the Hugoniot after heat release). The
corresponding minimum velocity of the driving piston, vpCJ , is given by mass conservation
through the detonation, ρb (D − vp ) = ρu D. The tangency of the CJ Michelson–Rayleigh
line at the point BCJ indicates that the exhaust gas flow is sonic in the reference frame of
the CJ detonation but subsonic for all other regimes Mu > MuCJ (see Fig. 4.5 for more
details). As explained below, the minimum speed DCJ is also the propagation speed of
an autonomous detonation, that is, a detonation that propagates without the support of a
driving piston. In the limit Q 1 the expression for the CJ velocity in (4.2.44) reduces
to (1.2.9).

ZND Structure (1940)

For a constant Mach number of propagation Mu > MuCJ , the state of the burnt gas
(1/ρb , pb ) of a steady planar detonation corresponds to the point B in Fig. 4.4, where
the point U represents the initial state. The other root, represented by the point B , is not

[1] Shchelkin K., Troshin Y., 1965, Gasdynamics of combustion. Baltimore, Md.: Mono Book Corp.

physical, as explained few lines below; see Fig. 4.5. The transition from the initial state
to the state of the burnt gas occurs in two steps. In the first step, the leading shock wave
takes the frozen mixture from point U to point N on the Hugoniot curve plotted in Fig. 4.4,
and the exothermic reaction takes the gas from point N to point B in a second step. This
results from the difference of time scales between elastic collisions controlling the inner
structure of inert shock waves and the inelastic collisions controlling the heat-release rate
(large activation energy). Since a gas molecule undergoes only a small number of elastic
collisions as it crosses the shock wave, the exothermic reaction cannot develop until later
downstream.
To describe the internal structure of a detonation, it is useful to introduce the progress
variable ψ of the exothermic reaction, ψ = 0 in the initial state and ψ = 1 in burnt gas;
see Equation (15.1.42). For a one-step Arrhenius reaction ψ = θ = (1 − ψ), where ψ is
the reduced mass fraction of reactant. The reaction is supposed irreversible to simplify the
presentation. Extension to more general cases is possible.[2]
The speed of shocked gas with respect to the shock is subsonic u < a, except at the
CJ point where it is sonic (ubCJ = abCJ ), but it is of the same order of magnitude as the
local speed of sound, u/a = O(1), typically u/a = 0.2. The dimensional analysis in (1.2.5)
then shows that the flow speed is high compared with diffusion speeds when the activation
energy is sufficiently large,

E/kB T > 1 ⇒ u D/τr ≈ a e−E/kB T , (4.2.45)
where 1/τr is the reaction rate and D is a diffusion coefficient, given in (1.2.2) and (1.2.4),
respectively. In other words, downstream of the shock, diffusive transport (heat conduc-
tion, molecular diffusion, viscosity) is negligible compared with convective transport,
D/(uτr )2 u/(uτr ), where uτr is the length of variation associated with the reaction rate,
D ≈ a2 τcoll , with τcoll the time between molecular collisions, u ≈ a and τcoll τr . Under
these conditions, the conservation equations (15.1.33)–(15.1.34) controlling the internal
structure of the detonation, written in the coordinate system of the steady plane wave, are
d(ρu)
= 0, (4.2.46)
dx
dp du
+ ρu = 0, (4.2.47)
dx dx

γ d p du dψ
+ u − qm = 0, (4.2.48)
γ − 1 dx ρ dx dx
dψ ρ
= , ρu (4.2.49)
dx τr
where the ideal gas law, cp T = [γ /(γ − 1)]p/ρ, the progress variable ψ and the rate of heat
release 1/τr have been introduced. The latter is a chemical kinetic quantity that depends
on the composition and on the thermodynamic conditions (ρ, T). The quantity qm ψ(x) is

[2] Fickett W., Davis W., 1979, Detonation. University of California Press.

the heat released in the slab of compressed gas delimited by x = 0 (Neumann state, just
downstream of the inert shock wave, ψ = 0) and x. The first three equations are written in
a conservative form that can be integrated directly from the initial condition ψ = 0 to any
intermediate state ψ = 0, ψ 1. This allows us to express the intermediate thermodynamic
states as functions of ψ. In the notation of (4.2.42), we obtain the relation
PV + P + V = ψQ (4.2.50)
in which Equation (4.2.13) must be used. This equation is a generalisation of (4.2.42); it
describes hyperbolas that lie between those of the Hugoniot and the burnt gases. A relation
involving the speed of sound a2 = γ p/ρ, equivalent to (4.2.50), is obtained by integration
of Equation (4.2.48) from the initial state:
1 1 1 1
a2 + u2 − qm ψ = a2u + D 2 . (4.2.51)
γ −1 2 γ −1 2
For ψ = 0 this equation for the conservation of total energy across the inert shock wave
has two solutions, represented by two points in Fig. 4.4, U corresponding to the initial state
(fresh gas),
ψ = 0: u2 = D 2 , a2 = a2u ,
and N corresponding to the Neumann state just behind the inert leading shock,
ψ = 0: u2 = u2N , a2 = a2N .
For a given initial condition (1/ρu , pu ), the internal structure of the wave and the prop-
agation velocity Mu are then fully determined by the function ψ(x), which can be deter-
mined by solving the chemical kinetic problem, that is, by integration of Equation (4.2.49).
Instructive information can be obtained without considering chemical kinetics by studying
the solutions of the system (4.2.46)–(4.2.48) in the phase space u2 –ψ. Using ρu = cst.,
Equation (4.2.47) and a2 = γ p/ρ, the first term on the left-hand side of (4.2.48) can be
written as

γ p du 1 dp 1 a2 du γ du
+ = − u ,
γ − 1 ρu dx ρ dx γ − 1 u dx γ − 1 dx
and the equations describing the inner structure of a detonation take the form
du dψ du2 u2
(a2 − u2 ) = (γ − 1)qm u , = 2(γ − 1)qm 2 , (4.2.52)
dx dx dψ (a − u2 )
where the expression for a2 , as a function of u2 and ψ for a given value of D, is obtained
from (4.2.51). The solutions of Equation (4.2.52) are represented by trajectories in the
phase space u2 –ψ that pass through the two points U and N. The phase portraits of these
solutions are shown schematically in Fig. 4.5. The solid lines represent trajectories of the
physical solutions for D DCJ . They describe the passage of shocked gas from the
Neumann condition (just downstream of the inert shock wave, ψ = 0) to that of the burnt
gas (ψ = 1). The flow velocity, u, in the frame of the detonation is an increasing function

(Initial state)

Supe
rson
ic

c
soni (Burnt gas)
Sub

(Neumann state)

Figure 4.5 Phase portraits of the solutions to Equation (4.2.52).

of ψ; since the flow is subsonic, u < a. For D = DCJ , the slope is vertical at ψ = 1
(the point BCJ ); the flow velocity in the burnt gas is sonic, u = ubCJ = abCJ . The part
of the trajectories shown as dashed lines, connecting the initial state (ahead of the lead
shock wave, point U, ψ = 0) to the burnt gas (point B , ψ = 1), describes a supersonic
flow (u > a) where the flow velocity decreases as energy is released. This solution cannot
describe a self-propagating wave, because the chemical reaction is frozen at the initial
conditions (zero reaction rate, τr = ∞), and, in the absence of molecular transport (heat
diffusion and molecular diffusion) ψ cannot increase. Moreover, the wave velocity selected
by heat diffusion is very subsonic; see (1.2.5). In addition, the structure of a supersonic
wave attached to a flame holder at room temperature Tu ≈ 300 K would necessarily have
a very large thickness ≈ Dτr (Tu ), much larger than any terrestrial macroscopic length.
For example this length would be of the order of 3 × 1021 m (or 3 × 105 light-years!) for
D ≈ a ≈ 5 × 102 m/s and τr (Tu ) having the order of magnitude discussed at the end of
Section 1.2.2 for an Arrhenius law. There is therefore no physical solution going from U
to B in Figs. 4.4 and 4.5. For D < DCJ , there is no stationary solution that goes from the
state of the initial fresh gas ψ = 0 to the state of the burnt gas ψ = 1, even though the
reaction rate is nonzero.
In summary, overdriven planar detonations

can be created by a piston whose velocity
vp is greater than vpCJ = 1 − ρu /ρbCJ DCJ , vp > vpCJ . The velocity of the detonation
decreases with that of the piston and has a lower bound DCJ given by Equation (4.2.44).
For piston speeds less than vpCJ , supersonic combustion waves with a uniform flow between
the end of the exothermic reaction and the piston, and propagating at constant speed, do
not exist. Detonation waves consist of an inert shock wave followed by a reactive subsonic
flow (in the frame of the wave). Under the effect of energy release, the density and pressure
decrease from (ρN , pN ) to (ρb , pb ), while remaining always greater than their initial values,
ρb > ρu pb > pN ; see Fig. 4.4. In the reference frame of the motionless fresh gas, the

burnt gas velocity is nonzero and oriented in the direction of propagation, vp = D − ub =

D(ρb −ρu )/ρb > 0; see Fig. 4.3. The marginal case, D = DCJ , is characterised by a speed
of burnt gas at the end of heat release that is equal to the sound speed, ubCJ = abCJ .

Profiles of Temperature, Velocity and Pressure

The detonation structure is found by expressing the flow velocity and the thermodynamic
variables in terms of the progress variable ψ(x), which increases from ψ = 0 at the
Neumann state to ψ = 1 in the burnt gas. Consider a detonation propagating at a given
velocity D; Mu ≡ D/au MuCJ is fixed. Using the notations of (4.2.11), the expression
for the specific volume downstream from the shock is given in terms of ψ by the solution of
the quadratic equation for V obtained when (4.2.13) is introduced into (4.2.50). The root
that corresponds to the Neumann state VN = −(Mu2 − 1)/Mu2 at ψ = 0 is an increasing
function of ψ ∈ [0, 1],

(Mu2 − 1) 4Mu2
V − VN = 1− 1− Qψ , (4.2.53)
2Mu2 (Mu2 − 1)2

where the square root is real for Mu MuCJ , [4Mu2CJ /(Mu2CJ − 1)2 ]Q = 1. The density
and the pressure, obtained from (4.2.53) with (4.2.13), decrease continuously from the
Neumann state to the burnt gas. The temperature is then obtained using the ideal gas law in
the following form, written with the notation
2
Mu2 4Mu2 (Mu2 −1) (M −1)
Y ≡ (γ γ+1) (M 2 −1) T−TN
T u
, X ≡ (M 2 −1) 2 Qψ, A ≡ γ +1 , B ≡ (γ − 1) (γu+1) + γ1 ,
u u
√
Y + B = AX + B 1 − X, X ∈ [0, Xb 1], (4.2.54)
where for ordinary detonations B < 2A, (γ − 1)/γ < [(Mu2 − 1)/(γ + 1)](3 − γ ). Equa-
tion (4.2.54) then shows that the temperature first increases when ψ increases from 0,
√
reaches a maximum at X = Xm , 1 − Xm = B/(2A) and decreases down to the burnt
gas temperature[1] at the end of the exothermic reaction, ψ = 1. In the last part of the
reaction zone the temperature decreases because the rate of heat release is smaller than the
rate of adiabatic cooling due to gas expansion in the thermal equation (15.1.34)) where
Dp/Dt < 0. The burnt gas of a CJ wave corresponds to X = 1 in (4.2.54), VbCJ − VNCJ =
(Mu2CJ − 1)/2Mu2CJ , and the total increase of temperature in the shocked gas is
(TbCJ − TNCJ ) (2 − γ )γ 2
Mu2CJ 1: ≈ M , (4.2.55)
Tu (γ + 1)2 uCJ
yielding (TbCJ − TNCJ )/Tu ≈ 0.17 Mu2 for Mu 1 and γ = 1.3. In ordinary detonations,
the heat release is larger than the enthalpy of the initial gaseous mixture, about 10 times
larger, so according to (4.2.44), the square of the propagation Mach number is also large:
6
qm
qm cp Tu ⇒ DCJ ≈ 2(γ 2 − 1)qm , MuCJ ≈ 2(γ + 1) . (4.2.56)
cp Tu

[1] Zeldovich Y., Kompaneets A., 1960, Theory of detonation. Academic Press.

Speed of discontinuities in laboratory frame

Gas speed
in laboratory
frame

Rarefaction
wave
Detonation

Detonation
Moving
piston Static
Shock piston Shock

Figure 4.6 Instantaneous arrest of a piston supporting an overdriven detonation, ub < ab ; see
Fig. 4.3. The piston, initially at speed vp > vpCJ , is instantaneously brought to rest at a time just
after t = to . The plots show the profiles of flow velocity in the laboratory frame (where the fresh gas
is at rest) at times to and t > to . The propagation speeds of the discontinuities are shown above the
plots. The sound speed close to the piston is given by (15.3.47), ao = ab − 0.5(γ − 1)(D − ub ).

Neglecting the quantity (γ − 1) < 1 in front of 2γ Mu2 , the Neumann conditions (4.2.15)–
(4.2.17) yield
(γ − 1) 1 TN 2γ pN 2γ
γ MN2 ≈ + 2, ≈ (γ − 1)M 2
u + 2 , ≈ M2,
2 Mu Tu (γ + 1) 2 pu γ +1 u
(4.2.57)
showing that the flow Mach number at the Neumann state is relatively small but the pressure
jump across the lead shock is large, scaling as the square of the propagation Mach number.
For typical gaseous mixtures, qm /cp Tu = 8, γ = 1.3, the Mach numbers are, according
to (4.2.56)–(4.2.57), MuCJ = 6.22 and γ MN2 CJ = 0.176 and the temperature jump across
the lead shock is relatively large, TNCJ /Tu ≈ 6.66. According to (4.2.55), the increase of
temperature in the shocked gas is not large, TbCJ /TNCJ ≈ 2, so that the most important
part of the total temperature jump in a CJ detonation is across the leading shock wave.
The increase of the flow velocity across the exothermic reaction zone is of order unity,
ubCJ /uNCJ = 2.88, where ubCJ = abCJ and the pressure is divided approximately by two
across the reaction zone, pbCJ /pNCJ ≈ 0.5.

Dynamics of the Burnt Gas in a Planar Chapman–Jouguet (CJ) Wave

The selection mechanism leading to the CJ regime of an autonomous detonation (no sup-
porting piston) can be explained in planar geometry using the simple waves described
in Section 15.3.4; see Fig. 15.11. When the piston supporting an overdriven detonation,
initially at the speed vp , is suddenly stopped, a simple rarefaction wave leaves the piston,
reducing the velocity of the burnt gas flow (measured in the laboratory frame) from vp
to zero; see Fig. 4.6. The rarefaction wave propagates and widens linearly in time until it
meets and weakens the detonation. The velocity of the leading shock then decreases until it
reaches the limit DCJ . Starting at this instant, the burnt gas velocity in the frame of the CJ

Rarefaction
wave

Detonation
Static
piston Shock

Figure 4.7 Autonomous CJ detonation. The velocity profile in the detonation is identical to that of
Fig. 4.6 just after the rarefaction wave has encountered the detonation.

wave becomes sonic, ubCJ = abCJ ; the rarefaction wave can no longer penetrate the internal
structure of the detonation which then continues to propagate autonomously at the velocity
DCJ ; see Fig. 4.7.
After a sufficiently long time, the initial conditions for initiation of the detonation at the
closed end of the tube are forgotten. The whole CJ detonation wave can be considered as a
hydrodynamic discontinuity as soon as its thickness becomes negligible compared with the
characteristic length of the burnt gas flow. This flow is described by a self-similar solution,
obtained in 1941 by G. Taylor[1] but published later. The detonation front propagates at
velocity DCJ , and the velocity of the burnt gas (in the laboratory frame) at the detonation
front is also constant,
vbCJ = DCJ − abCJ , (4.2.58)
since, according to the CJ wave, ubCJ = abCJ , where abCJ is the sound speed in the burnt
gas at the detonation front. For a constant propagation velocity, the entropy is constant at
the detonation front and is thus also constant in the entire flow behind the detonation front.
This is the key simplification that is associated with a constant propagation velocity of the
detonation front. For reasons similar to those in Sections 15.3.4 and 15.3.5, the flow of
burnt gas is sought in the self-similar form of a function of space and time through the ratio
x ≡ x/t, ρ(x), v(x), where v ≡ DCJ − u denotes the flow velocity in the laboratory frame
(u is the flow velocity relative to the detonation front). The Euler equations take the form
1 dρ dv 1 dρ dv
(v − x) + = 0, a2 + (v − x) =0 (4.2.59)
ρ dx dx ρ dx dx
(see (15.3.56)), where, with the isentropic condition, the sound speed is a function of the
mass density, a(ρ). In an ideal gas with a constant ratio of specific heats (polytropic gas),

γ −1
p/ρ γ = cst., a(ρ)/abCJ = ρ/ρbCJ 2 , (4.2.60)

[1] Taylor G., 1950, Proc. R. Soc. London Ser. A, 200, 235–247.

the existence of a nontrivial solution to the system (4.2.59) requires that the determinant be
zero, a2 = (v − x)2 , so that v + a = x/t as for the simple waves corresponding to the linear
characteristics C+ introduced in Section 15.3.4. Integrating the first equation in (4.2.59)
yields
ρb
CJ a(ρ) 2

v + a = x/t, vbCJ − v = dρ = abCJ − a , (4.2.61)

ρ ρ γ −1
where the last relation is obtained using (4.2.60). This shows that the flow velocity v and
the sound speed a are linearly related. Combining the two equations in (4.2.61) leads to an
expression for the flow field in the form

γ +1 γ −1 x
v + abCJ − vbCJ = , (4.2.62)
2 2 t
obtained using the boundary conditions at the detonation front given by the burnt gas state
of the CJ wave, v = vbCJ = DCJ − abCJ and a = abCJ . The result in (4.2.62) shows
the coherence of the self-similar solution since the point where v = vbCJ , which is the
detonation front, propagates at the CJ velocity, x/t = abCJ + vbCJ ≡ DCJ . The solution
corresponds to a planar self-similar solution, introduced in general terms in Section 15.3.5,
with, in the notations of (15.3.61)–(15.3.63), ξ = x/(DCJ t), A = 1/DCJ , α = 1 and
βN = 0.
When the rear end of the tube (x = 0 where the detonation was ignited) is closed, the
velocity of the flow at that point is zero. According to (4.2.62), the flow velocity decreases
to zero, v = 0, at the point x = xo (t) such that xo /t = abCJ − vbCJ (γ − 1)/2. This
point propagates at the local sound speed of the gas at rest, x/t = ao , since, according
to the second equation in (4.2.61), ao = abCJ − vbCJ (γ − 1)/2. In the limit of a strong
CJ detonation, the sound speed ao takes a very simple form ao ≈ DCJ /2, as shown now.
In this limit the temperature and pressure of the fresh mixture are negligible in front of
those in the burnt gas, TbCJ Tu and pbCJ pu . Using the relation a2b = γ pb /ρb and
ρb ab = ρu DCJ , Equation (4.2.41) leads to a quadratic equation for X ≡ DCJ /abCJ , X 2 +
(γ 2 − 1)X/2 −(γ + 1)2 /γ = 0, yielding
DCJ ρb (γ + 1) DCJ
qm cp Tu ⇒ = CJ ≈ , ≈ (γ + 1). (4.2.63)
abCJ ρu γ vbCJ
The gas is at rest in the region between the end of tube, x = 0, and the weak discontinuity at
x = xo (t) = ao t, where, according (4.2.62)–(4.2.63), ao = DCJ /2 in the limit of strong CJ
detonation; see Fig. 4.8. When the end of the tube at x = 0 is open, the solution in (4.2.61)
extends to negative values of the flow velocity v, corresponding to burnt gas flowing out
from the tube.

4.3 Initiation of Detonation

As explained in Sections 5.2.1 and 5.2.2, chemical kinetics introduce a crossover temper-
ature T ∗ , in the range 950–1400 K for normal conditions (see Sections 5.3 and 5.4) for

0
0 1/2 1

Figure 4.8 Planar self-similar solution for the flow of burnt gas in the limit (4.2.63) for a strong CJ
detonation in a polytropic gas.

the combustion of hydrogen and methane. The rate of the exothermic reaction changes
drastically, by two to three orders of magnitude, in a small temperature range around T ∗
of a few percent T/T ∗ ∼ 10−1 to 10−2 . For example, the induction delay (the time
delay before thermal runaway) varies from 2 × 10−4 s at 1000 K to 10−1 s at 900 K for
a stoichiometric mixture of hydrogen–air at atmospheric pressure.[1] It is of the order of
102 s and 10−6 s at 750 K and 2000 K, respectively. A simple model of induction delay is
presented in Section 5.2.2.

4.3.1 Detonability Limits

A planar detonation cannot propagate unless the temperature of the shocked gas just behind
the inert lead shock, TN , is sufficient to initiate the exothermic reaction downstream of
the shock after a time (induction delay) sufficiently short for the region of heat release
to be located not too far from the lead shock. If this is not the case, the slightest heat
or radical loss in the transverse directions quenches the exothermic reaction. Roughly
speaking, the limiting composition (dilution, equivalence ratio) that defines the detonability
limits of a reactive mixture is given by the equality TNCJ = T ∗ , where TNCJ is the Neumann
temperature of the CJ detonation. No detonation can propagate autonomously if TNCJ < T ∗ .
The temperature TNCJ , like DCJ , is directly associated with the chemical energy released
per unit mass qm and is thus fully controlled by the composition of the reactive mixture,
while T ∗ depends on the detailed chemical kinetics. The detonability limits TNCJ = T ∗
concern both rich and lean mixtures.

4.3.2 Mechanisms of Detonation Initiation

In a reactive mixture whose composition lies inside the detonability limits, the ways to ini-
tiate a self-propagating detonation have been classified into three categories: spontaneous
initiation, direct initiation and the so-called deflagration to detonation transition (DDT).

[1] Sanchez A., Williams F., 2014, Prog. Energy Combust. Sci., 41, 1–55.

The two last mechanisms rely on the formation of an inert shock wave that becomes
sufficiently strong to generate a pressure larger than, or at least equal to, pNCJ . Due to the
large value of the ratio pNCJ /pu > 30 for ordinary combustible mixtures (see (4.2.57)),
this condition makes DDT difficult to be observed in open space. In a channel DDT arises
through flame acceleration and/or DDT is coupled to spontaneous ignition in hot spots
ahead of the flame. Direct initiation refers to the formation of a detonation in open space
in the decay of a strong blast wave produced by a powerful concentrated energy source.
If sufficient energy is deposited (quasi-)instantaneously, E Ec , a CJ wave is formed at a
certain distance from the source. The challenge is to predict the critical initiation energy Ec .
Spontaneous initiation can occur in a preconditioned medium through a gradient of induc-
tion time. Roughly speaking, a detonation is initiated when the velocity of the induction
front is decreased to the velocity of pressure wave (sound speed) so that a runaway of the
pressure occurs, leading quasi-spontaneously to a CJ wave. This mechanism is involved
in some dangerous explosions and in the transition from deflagration to detonation. The
intense pressure peak, observed in bad ignition of liquid rocket engines, could be related to
the spontaneous initiation of a detonation, followed rapidly by detonation quenching. The
fundamental studies of these initiation mechanisms are presented in the next sections.

4.3.3 Direct Initiation of Detonation

After a successful initiation in open space, E Ec , at large distance from the ignition
region, much larger than the detonation thickness, a spherical detonation front is locally
planar and propagates at the CJ velocity, DCJ . The self-similar solution[2] for a spherical
detonation propagating at constant velocity DCJ is first presented below. The solution is
obtained in a way similar to, but more complicated than, the planar solution in Fig. 4.8. In a
second step we consider the solution for spherical detonations including the modifications
to the inner structure due to the front curvature. Finally, an expression for the critical energy
for direct initiation, Ec , is derived.

Self-Similar Flow of Burnt Gas in a Spherical CJ Wave (1950)

When the intrinsic instabilities of a detonation wave are disregarded, curvature of the front
and internal unsteady phenomena can be neglected when the detonation radius is much
larger than the thickness so that the internal structure of the wave is that of the planar
wave. The detonation can then be considered as a spherical discontinuity of radius rf (t) =
DCJ t, across which the Rankine–Hugoniot condition (4.2.43) is satisfied with, according
to (4.2.13), P = −Mu2CJ V, where MuCJ is a constant quantity given by (4.2.44). The radial
flow field is sought in the form of a self-similar form, v(x), ρ(x), where x ≡ r/t, r being
the radial coordinate. The Euler equations become
1 dρ dv 2v 1 dρ dv
(v − x) + + = 0, a2 + (v − x) = 0, (4.3.1)
ρ dx dx x ρ dx dx

[2] Taylor G., 1950, Proc. R. Soc. London Ser. A, 200, 235–247.

so that after elimination of (1/ρ)dρ/dx,

x dv 2
= − v−x
2 > 0. (4.3.2)
v dx 1− a

The first equation in (4.3.1) differs from the planar case (4.2.59) by the additional geomet-
rical term 2v/x, and numerical solutions are required. Since the CJ velocity, DCJ , and the
thermodynamic conditions in the burnt gas at the detonation front, pbCJ , ρbCJ , are constant
and fully determined by the conditions of the initial mixture, pu , ρu , the entropy is constant
in the burnt gas, ∂p/∂r = a2 (ρ)∂ρ/∂r, where a2 (ρ) is given in (4.2.60). Following the
method in Section 15.3.5 for self-similar solutions of first kind, the solution is sought in the
form

ξ ≡ r/(DCJ t), v = vbCJ U(ξ ), ρ = ρbCJ R(ξ ),

with a/abCJ = R (γ −1)/2 . Using the strong detonation limit (4.2.63) for simplicity,
DCJ /vbCJ = γ + 1, a/vbCJ = γ R (γ −1)/2 , Equations 4.3.1 take the form

1 dR dU 2U
[U − (γ + 1)ξ ] + + = 0, (4.3.3)
R dξ dξ ξ
1 dR dU
γ 2 R γ −1 + [U − (γ + 1)ξ ] = 0, (4.3.4)
R dξ dξ
where the only parameter is γ . The solution is obtained by integrating this system of
equations numerically, with the boundary conditions

ξ = 1: U = 1, R = 1; ξ = 0: U = 0,

where the last relation comes from the spherical geometry. The numerical calculation starts
from the front, ξ = 0, and is stopped at the point ξ = ξo , where U = 0, 0 < ξo < 1.
According to (4.3.2), the derivative of the velocity with respect to the radius diverges at the
detonation front since, according to (4.2.58), a2bCJ − (vbCJ − DCJ )2 = 0, and, according to
(4.3.3)–(4.3.4), limξ →0 (1 − U)2 = [2γ /(γ + 2)] (1 − ξ ). The numerical solutions[1] are
shown in Fig. 4.9 for γ = 1.25.
The flow of burnt gas is bounded by two spheres, the detonation front rf (t) = DCJ t
and an inner sphere of radius ro (t) = ao t, where the velocity is zero and ao denotes the
sound speed in the burnt gas at rest inside this sphere. This can be seen as follows. Let
xo ≡ ro /t = 0 be the point at which v becomes zero; one has limx→xo d ln v/dx = +∞
since limv→0 ln v = −∞, so that, according to (4.3.2), xo = ao . This point is a weak
discontinuity where the velocity and its first derivative with respect to the radius are both
zero, but the second derivative is discontinuous.[2]

[1] Liñan A., et al., 2012, C. R. Mécanique, 340, 829–844.

[2] Landau L., Lifchitz E., 1986, Fluid mechanics. Pergamon, 1st ed.

Figure 4.9 Self-similar solutions for the normalised temperature, density, pressure and gas velocity
inside a spherical CJ detonation. From Liñan A. et al., 2012, C. R. Mécanique, 340(11–12), 829–844
with permission. Copyright 2012, published by Elsevier Masson SAS. All rights reserved.

Zeldovich Criterion (1956)

The direct initiation of gaseous detonation by an energy source, and the determination of
the critical energy Ec are old problems that have been extensively studied and reviewed[3,4]
since the pioneering work of Zeldovich et al..[5] Much experimental data for Ec is available
since the 1980s, but the origin of the critical nature of the phenomenon was not understood
and described analytically until a decade later.[6] Let us first summarise the pioneering
analyses.
Consider an igniter of negligible size that deposits an energy E in a negligibly short time.
The 1941 self-similar solution for a point blast explosion,[7] recalled in Section 15.3.5, may
be considered as the initial condition, since, at early times, the chemical heat release is
negligible compared with E. Since the propagation of a spherical CJ wave is also described
by a self-similar solution, the successful direct initiation of detonation may be then con-
sidered as the transition between the two self-similar solutions, at least when multidimen-
sional and/or unsteady effects (pulsating or cellular detonations) are disregarded. The first
numerical solutions[8] were obtained by considering the detonation as a discontinuity across
which the planar jump conditions are satisfied. This approximation, often called ‘infinitely
fast chemistry’, does not have a critical energy: the strongly overdriven detonation that
is initially generated by the sudden deposition of heat at a point relaxes systematically
to a CJ wave no matter how small the value of E, and the onset of the spherical CJ
detonation occurs at a well-defined radius of the order of (E/ρu qm )1/3 . A more detailed
analysis of this problem has been recently performed.[9] These studies show that the critical

[3] Lee J., 1977, Ann. Rev. Phys. Chem., 28, 75–104.
[4] Lee J., Higgins A., 1999, Proc. R. Soc. London Ser. A, 357(3503-3521).
[5] Zeldovich Y., et al., 1956, Sov. Phys.–Tech. Phys., 1, 1689–1713.
[6] He L., Clavin P., 1994, J. Fluid Mech., 277, 227–248.
[7] Taylor G., 1950, Proc. R. Soc. London Ser. A, 201(1065), 159–174.
[8] Korobeinikov P., 1971, Ann. Rev. Fluid Mech., 3, 317–346.
[9] Liñan A., et al., 2012, In Vazquez-Cendon E., et al., eds., Numerical methods of hyperbolic equations, vol. 61-74, Taylor
and Francis.

energy Ec is associated with the finite thickness of the inner structure of the detonation
wave.
A first criterion was proposed by Zeldovich et al.[1] in the following way. The time taken
∗ ≈ (E/ρ )1/3 (1/D )5/3 (see
for the blast wave velocity to decrease to the CJ velocity, τCJ u CJ
(15.3.67)), must not be shorter than the induction delay for ignition of the mixture at
the Neumann state of the CJ wave. The induction delay is of same order of magnitude
as the transit time of a fluid particle across the wave, τCJ ≈ dCJ /uNCJ , where dCJ
∗ ≈ τ
is the thickness of the planar CJ detonation. The condition τCJ CJ yields Ec ≈
ρu DCJ τCJ ≈ ρu (DCJ /uNCJ ) DCJ dCJ for the critical energy. Equations (4.2.44) and
5 3 3 2 3

(4.2.14) in the limit of a strong CJ detonation, DCJ2 ≈ 2(γ 2 − 1)q , DCJ ≈ (γ +1) , then
m uN (γ −1)
CJ
give
(γ + 1)4 3
Ec ≈ 2ρu qm d (4.3.5)
(γ − 1)2 CJ
for the critical energy obtained by Zeldovich et al..[1] The radius at which the CJ wave is
formed is of the order of the thickness of the CJ wave, dCJ . The expression for Ec in (4.3.5)
is related to the chemical energy in a sphere of fresh mixture of radius of same order as
dCJ . This critical value of energy is smaller than the experimental data[2] for Ec by a factor
10−5 to 10−6 .
This suggests that the modifications to the planar structure of detonations play an essen-
tial role. There are two possibilities: either unsteady effects inside the detonation structure
or curvature effects. What is surprising at first sight is that such small effects may have
such a large influence. As we shall see, this is due to the high sensitivity of the reac-
tion rate to temperature. A correct order of magnitude for Ec was obtained[3] in the limit
of a high sensitivity of the reaction rate to temperature by taking account of nonlinear
curvature effects in the inner structure of the detonation wave. This analysis is presented
next.

Inner Structure of Detonations in Spherical Geometry (1994)

Neglecting viscosity and heat conduction, Euler’s equations (conservation of mass, momen-
tum and entropy) in spherical geometry take the form (15.3.58)–(15.3.59). To study the
inner structure of the detonation wave it is more convenient to use the reference frame of
the lead shock. Denoting by rf (t) and D ≡ drf /dt the radius and the velocity of the lead
shock, we introduce the flow velocity in the reference frame of the shock, u = D(t) − v,
and the distance from the shock, x = rf (t) − r, where v and r denote the flow velocity and
the radius in the laboratory frame, x > 0 and u > 0 in the shocked gas. Using the change of
variables, r → x, ∂/∂r → −∂/∂x, ∂/∂t → ∂/∂t + D∂/∂x, Equations (15.3.58)–(15.3.59)
become

[1] Zeldovich Y., et al., 1956, Sov. Phys.–Tech. Phys., 1, 1689–1713.

[2] Lee J., 1984, Ann. Rev. Fluid Mech., 16, 311–336.
[3] He L., Clavin P., 1994, J. Fluid Mech., 277, 227–248.

which can be rewritten

∂(ρu) ∂(ρu2 + p) 2 dD
+ + ρ(D − u)u = , (4.3.7)
∂t ∂x rf − x dt

where the last term in the left-hand side is introduced by the flow divergence in spheri-
cal geometry (in cylindrical geometry, the 2 is replaced by 1). When heat conduction is
neglected, Equation (15.1.34) for conservation of energy, written in the moving frame of
the lead shock, D/Dt → ∂/∂t + u∂/∂x, takes the form

∂ ∂ 1 ∂ ∂
+u (cp T − qm ψ) − +u p = 0,
∂t ∂x ρ ∂t ∂x

where the notations below (15.1.42) have been used. Using the expression for (1/ρ)(∂p/∂x)
from the second equation in (4.3.6), one gets

∂ ∂ u2 1 ∂p dD
+u cp T + − qm ψ − −u = 0. (4.3.8)
∂t ∂x 2 ρ ∂t dt

Equations (4.3.6)–(4.3.8) are valid both in the burnt gas (ψ = 1), where they reduce to
(15.3.58)–(15.3.60), and inside the inner structure of the detonation front, where ψ(x, t)
is given by the solution of the chemical equation in (12.2.2). The geometrical term that
introduces the curvature effect is the divergence of the flow in the mass conservation in
(4.3.6).
Without the unsteady and curvature terms, Equations (4.3.6)–(4.3.8) reduce to (4.2.46)–
(4.2.48) for the structure of planar waves propagating at constant speed, leading to the
jump conditions (15.1.45)–(15.1.48) (where u is denoted U) with no external heat flux.
When these jumps are incorporated into Euler’s equations, the numerical results[4] show
systematic initiation of a detonation, for any value of the deposited energy, when the self-
similar solution of point blast explosion is used as an initial condition.
The perturbation analysis of the inner structure of detonations, presented next, is per-
formed in the limit of a large activation energy for a weakly curved front and in a quasi-
steady-state approximation. The latter is valid if the characteristic time for evolution of the
radius rf (t) and of the propagation velocity D(t) is much longer than the transit time of a
fluid particle across the inner structure of the detonation wave. In this case the equations
reduce to the system

[4] Korobeinikov P., 1971, Ann. Rev. Fluid Mech., 3, 317–346.

∂(ρu) 2 ∂(ρu2 + p) 2
≈− ρ(D − u), ≈− ρ(D − u)u, (4.3.9)
∂x rf − x ∂x rf − x
2

∂ γ γ−1 ρp + u2 − qm ψ
≈ 0. (4.3.10)
∂x
The validity of the approximation is limited not only by the conditions in the external
flows, outside the inner structure of a detonation, but also by the stability properties of the
detonation structure itself since instabilities may lead to pulsating phenomena and cellular
fronts, both evolving on a time scale as short as the transit time; see Sections 4.5.1 and
4.5.2.
Integrating (4.3.9)–(4.3.10) across the inner structure of the detonation gives the curva-
ture induced modifications to the jump conditions,

(ρb ub − ρu D) (ρb u2b + pb ) − (ρu D 2 + pu )

≈ −I1 , ≈ −I2 , (4.3.11)
ρu D ρu D 2
. /
γ pb u2b γ pu D2
+ ≈ + + qm (4.3.12)
γ − 1 ρb 2 γ − 1 ρu 2
d d
2 ρ(D − u) 2 ρu(D − u)
where I1 ≡ dx, I2 ≡ dx,
ρu D 0 (rf − x) ρu D 0 (rf − x)
2

and d is the detonation thickness. The integrals in I1 and I2 are well defined when there
is a clear separation of length scales between the inner structure of the detonation and
the external flow in the inert burnt gas. This is possible for weakly curved detonations,
≡ d/rf 1. Since I1 and I2 are of order , the inner structure of a quasi-steady and
weakly curved detonation can be obtained by a perturbation analysis in the limit 1. To
first order in this limit the perturbation terms in (4.3.11) take the form
d d
ρ(x) dx ρu dx
I1 ≈ 2 −1 , I2 ≈ 2 1− , (4.3.13)
0 ρ u d 0 ρ(x) d
where ρ(x) denotes the density distribution in the planar detonation at velocity DCJ ,
ρ(x)uN (x) = ρu DCJ .
Consider an expanding spherical detonation and look for the propagation velocity as
a function of its radius, D(rf ). Because of the high sensitivity of the induction delay to
temperature variations, a nonlinear analysis can be performed in the framework of the
weakly curved approximation. The solution for D(rf ) has a turning point that plays a key
role for the critical energy; see Fig. 4.10. This is shown as follows.
According to the chemical kinetics of gaseous combustion, studied in Chapter 5, the
internal structure of a detonation can be decomposed into two parts: an induction zone of
thickness dind (TN ) (induction length) where the release of chemical energy is negligible,
followed by an exothermic reaction zone of thickness dexo ; see Fig. 4.29. The thickness of
these two zones is typically of same order of magnitude in ordinary gaseous detonations, so
that the detonation thickness d is effectively of order of the induction length d/dind = O(1).

The expression for dind in terms of the Neumann temperature TN (t) is easily computed for
the one-step model (2.1.3) with a large activation energy βN ≡ E/(kb TN ), using a quasi-
steady-state approximation and limiting attention to small variations, δTN /TN = O(1/βN ),
sufficient to produce a strong variation of the reaction rate. For simplicity, consider first a
quasi-isobaric approximation in which the compressible effects are neglected; the exten-
sion to the more general case is straightforward. The modifications to the distribution of
temperature T and reduced species concentration Y/Yu = 1 − ψ are of order unity. They
are solutions to (8.1.1)–(8.1.2) (without diffusion terms) in which the small modifications
to the convection terms, of relative order 1/βN , are neglected:

d(cp T/qm ) dψ 1−ψ E 1 1
uN ≈ uN ≈ exp − −
dx dx τr (TN ) kB T TN

x = 0: T = TN (t), ψ = 0.
Here, the varying reaction time at the Neumann state, τr (TN ), has been used as the reference
time and the overbar denotes the unperturbed solution for TN = T N . The thermal runaway,
which marks the end of the induction zone, occurs when (T − TN )/TN becomes of order
1/βN , so that ψ ≈ 0 throughout the induction zone. Introducing the notation ≡ βN (T −
TN )/TN , the solution takes the form

cp T N uN τr (TN ) d cp T N uN τr (TN ) 1
≈ exp , dind ≈ de− .
qm βN dx qm βN 0
When compressible effects are taken into account, namely when the term udu/dx is retained
in (4.2.48), the result takes a similar form[1]
. 2
/
dind 1 1 − MN cp T N δdind δTN
βN 1: = 2
, = −βN , (4.3.14)
uN τr (TN ) βN 1 − γ M qm dind TN
N
where the second relation is valid to leading order in the limit βN → ∞ and comes from
the fact that the variation of dind with TN is given by the exponential term in the reaction
−1 −E/kB TN
rate τr−1 (TN ) = τcoll e .
The solution for D(rf ) is obtained easily using the so-called square-wave model. This
model is the large activation energy limit of the one-step ZFK model (2.1.3). The ratio of
the length of the exothermic zone to the induction length vanishes as 1/βN in the limit
βN ≡ E/kB TN → ∞, dexo /dind → 0, so that the detonation thickness becomes equal to
the induction length, d = dind . The exothermic reaction then proceeds inside an infinitely
thin zone, considered as a discontinuity and located at a distance dind from the shock front.
The variables, u(x), ρ(x) and p(x), which are discontinuous across the shock front, are also
discontinuous across the thin reaction layer and are constant inside the induction layer
separating the two discontinuities. The quantities I1 and I2 in (4.3.13) are then simple
functions of the Neumann state,

[1] He L., Clavin P., 1994, J. Fluid Mech., 277, 227–248.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 19:26:43, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.006
240 Gaseous Shocks and Detonations

dind ρN dind ρu ρu
I1 ≈ 2 −1 , I2 ≈ 2 1− = I1 . (4.3.15)
rf ρu rf ρN ρN
The two quantities I1 and I2 are proportional to the curvature of the front with a coeffi-
cient involving the induction length, dind (TN ), which is, according to (4.3.14), a nonlinear
function of TN and thus of the detonation velocity D. For a given initial mixture (qm , ρu ,
pu ), the three quantities defining the burnt state (ub , ρb , pb ) are solutions to three equations,
(4.3.11)–(4.3.12), involving rf and D through I1 and I2 . The condition for self-sustained
propagation introduces an additional constraint, the sonic condition in the burnt gas,

u2b = γ pb /ρb , (4.3.16)

which, in principle, determines the function D(rf ).

The analysis thus starts by computing I1 and I1 in terms of D and rf . The sensitivity
of the reaction rate 1/τr (T) to temperature implies that the quantities I1 and I2 are very
sensitive to variations of the detonation velocity, δD/D 1 ⇒ δI1,2 /I1,2 = O(1).
According to the Arrhenius law (1.2.2) with a large reduced activation energy,

E τr (TN ) TN − TNCJ
βN = 1: ≈ exp −βN , (4.3.17)
kB TN τr (TNCJ ) TNCJ

where TNCJ is the Neumann temperature of the planar CJ wave and TN is a function of
D given by (4.2.16). In the limit βN 1, it is sufficient to consider small variations
of TN around TNCJ , (TN − TNCJ )/TNCJ = O(1/βN ), and thus also small departures of
the detonation velocity from its CJ value: (D − DCJ ) /DCJ = O(1/βN ) with βN ≈
E/kB TNCJ 1. Neglecting small terms of order 1/βN coming from uN and ρN , and using
(4.2.14) and (4.2.16) in the strong shock limit for simplicity,

ρu γ −1 TN (γ − 1) TN D 2
(γ − 1)Mu2 1 ⇒ ≈ , ≈ 2γ Mu2 , ≈ ,
ρN γ +1 Tu (γ + 1)2 TNCJ DCJ
(4.3.18)
where Mu = D/au ; Equations (4.3.14) and (4.3.17) yield

(D − DCJ )
βN 1, dind = dCJ exp −2βN , (4.3.19)
DCJ
where dCJ is the thickness of the planar CJ detonation, equal to the induction length in the
square-wave model. Therefore, according to (4.3.15),

4 dCJ (D − DCJ ) γ −1
I1 (D) ≈ exp −2βN , I2 ≈ I1 . (4.3.20)
γ − 1 rf DCJ γ +1
Considering small curvature, dCJ /rf , and small changes in the detonation velocity,
δD/DCJ ≡ (D − DCJ )/DCJ , both of order 1/βN , dCJ /rf = O(1/βN ) and δD/DCJ =
O(1/βN ), the two quantities I1 (D) and I2 (D) are nonlinear functions of βN δD/DCJ =
O(1). However, the values I1 and I2 are still small, of order 1/βN . According to
(4.3.11)–(4.3.12), the same is true for the quantities δub (D)/abCJ , δρb (D)/ρbCJ and

δpb (D)/pbCJ , where the notations δub ≡ ub − abCJ , δρb ≡ ρb − ρbCJ and δpb ≡ pb − pbCJ
have been introduced, and where the subscript CJ refers to the planar CJ wave. These
quantities can all be computed by using a linear approximation in the left-hand sides of
(4.3.11) and (4.3.12),
δρb δub δD
+ = − I1 , (4.3.21)
ρbCJ abCJ DCJ

1 δpb δρb δub DCJ δD
+ +2 = 2 − I2 , (4.3.22)
γ pbCJ ρbCJ abCJ abCJ DCJ
2
1 δpb 1 δρb δub DCJ δD
− + = , (4.3.23)
γ − 1 pbCJ γ − 1 ρρCJ abCJ abCJ DCJ
where the relations γ pbCJ /ρbCJ = a2bCJ and ubCJ = abCJ have been used. The sonic
condition (4.3.16) yields the additional relation
δpb δρb δub
− −2 = 0. (4.3.24)
pbCJ ρρCJ abCJ
The four quantities δpb /pbCJ , δρb /ρbCJ , δub /abCJ and δD/DCJ , solutions to the four linear
equations (4.3.21)–(4.3.24), are thus expressed in terms of I1 and I2 that are, according to
(4.3.20), nonlinear functions of δD/DCJ . The expression δD(I1 , I2 )/DCJ then yields a
nonlinear equation for the detonation velocity,
- 0
γ +1 γ − 1 DCJ 2 DCJ δD γ +1 DCJ
1+ −2 = I1 (D) − I2 (D),
γ γ + 1 abCJ abCJ DCJ γ abCJ

which takes a simple form in the limit of strong detonation (4.2.63):

DCJ − D −2βN DCJ −D
16γ 2 dCJ
2βN e DCJ
= 2 βN . (4.3.25)
DCJ γ −1 rf
This equation, of the type Xe−X = h, gives a ‘C’-shaped curve for the detonation velocity
D versus the radius rf of the lead shock, with a turning point located at rf = r∗ , D = D ∗ ,
where

∗ 16eγ 2 ∗ 1
r = 2 βN dCJ , D = 1− DCJ (4.3.26)
γ −1 2βN
(see Fig. 4.10), where the curve was obtained numerically without introducing the approxi-
mation of the square-wave model. The latter introduces only a small quantitative difference.
This shows that the critical radius is much larger than dCJ when βN is large, typically by
a factor greater than 102 ! There are two branches of solutions to (4.3.25) for rf > r∗
with D + (rf ) > D − (rf ). Both are close to DCJ when βN 1 and the upper branch
corresponds to a propagation velocity slightly smaller than the CJ velocity, D + DCJ
with limrf →∞ D + = DCJ . The two branches merge at rf = r∗ , D + = D − = D ∗ , and there
is no solution to (4.3.25) at a smaller radius, r < r∗ . There is no spherical detonation, with
a radius smaller than r∗ , whose the inner structure is quasi-steady and having a velocity of

(a) (b)

Figure 4.10 D + and D + branches of the reduced detonation velocity versus reduced radius, obtained
numerically for γ = 1.4 and βN = 5.33. (a) The dashed line shows the trajectory of a CJ blast
wave with an initial energy just sufficient to reach r∗ with the velocity D ∗ . (b) The dotted curves
show the numerical solutions of the full unsteady equations (15.1.33)–(15.1.34), with zero thermal
conductivity, for the initiation of spherical detonations sustained by an Arrhenius law in (15.1.40) with
qm = 12.5. Curves 1–4 correspond to four different values of the energy source, E/(ρu DCJ dCJ 3 )=

3.3 × 107 , 5.69 × 107 , 1.34 × 108 and 2.64 × 108 , respectively. Reproduced from He L., Clavin P.,
1994, J. Fluid Mech., 277, 227–248, with permission

burnt gas at the end of the reaction zone ub equal to the local sound velocity, ub = ab . All such
solutions have a burnt gas velocity smaller than the sound velocity, ub < ab , so that, in the
presence of an expanding wave in the burnt gas, these detonations are systematically slowed
down. For r > r∗ , spherical detonations having a quasi-steady internal structure do exist
for D > D + and D < D − but with ub < ab so that the latter solutions are slowed down.
No solution exists for an intermediate velocity D − < D < D + and a radius larger than r∗ ,
rf > r∗ . The lower branch of solutions to (4.3.25) is thus not physical since these spherical
detonations slow down and will extinguish as soon as TN decreases below the crossover
temperature mentioned at the beginning of this section, Section 4.3. To summarise, only
the upper branch of solutions may attract quasi-steady detonations in spherical geometry.

Critical Ignition Energy (1994)

In view of the above results, it is reasonable to assume that a detonation can be initiated by
a blast wave only if the radius of the shock rf is equal to or larger than r∗ in (4.3.26) when
the shock velocity reaches DCJ , so that the radius at which the CJ wave is formed cannot be
smaller than r∗ , which is typically 102 larger than the detonation thickness, r∗ /dCJ ≈ 102 .
The marginal blast wave defined by this criterion is shown in Fig. 4.10a. Neglecting the
chemical heat release before ignition, and using the relation between the radius rf , the
velocity D and the deposited energy E in the self-similar solution for inert blast waves
(15.3.67), E ≈ ρu rf3 D 2 , the critical energy using (4.3.26) is[1]
2 ∗3
Ec ≈ ρu DCJ r ≈ 2ρu qm (γ 2 − 1)r∗3 . (4.3.27)

[1] He L., Clavin P., 1994, J. Fluid Mech., 277, 227–248.

The ratio between the two expressions (4.3.27) and (4.3.5) for the critical energy is
extremely large, of order 106 for βN = 8 and γ = 1.3. This is a consequence of the large
ratio of lengths, r∗ /dCJ ≈ 103 . The orders of magnitude of r∗ and Ec given by (4.3.26) and
(4.3.27) are in agreement with the experimental data. This estimation is also confirmed by
direct numerical simulations,[1] shown by curves 1–4 in Fig. 4.10b. The energy source of
curve 1 has the value given by (4.3.27). The detonation is effectively ignited by curve 3,
which has an energy source four times greater. These curves also indicate that one-
dimensional detonation fronts are easily destabilised to have an oscillating propagation
velocity.
Three assumptions limit the quantitative applicability of this result, namely that the blast
wave is inert and described by the Taylor self-similar solution (15.3.67), an oversimplified
chemical kinetic model and a quasi-steady-state structure of the inner detonation wave.
The first approximation is valid for a radius smaller than r∗ but larger than the size of
the igniter, and for times larger than the energy deposition time. A promising approach to
take into account a finite size of the igniter and a finite deposition time is presented in the
2013 analysis of Liñan et al..[2] A complex chemical kinetic scheme will not change the
results drastically, since the thermal sensitivity of the induction length is the key mechanism
of the phenomenon. Concerning the quasi-steady-state approximation, direct numerical
calculations[1] in spherical geometry shows that unsteadiness of the inner structure of the
detonation may increase the critical energy but does not modify the order of magnitude
predicted by (4.3.27); see Fig. 4.10 and the references in a 2012 review paper.[3] The above
analysis points out the essential role of nonlinear curvature effects that can be described in
an approximation of weak curvature, thanks to the strong thermal sensitivity.
However, unsteadiness is unavoidable since detonation waves are unstable and exhibit
both one-dimensional oscillations, called galloping detonations, and transverse structures
on the detonation front, called cellular detonations. Addressing intrinsic unsteadiness
requires considering times as short as the transit time. Such effects therefore cannot be
correctly taken into account by assuming slow temporal evolution, and their analytical
study has been performed only in some limiting cases; see Section 4.5.

4.3.4 Spontaneous Initiation and Quenching of Detonations

Ignition of a detonation in the absence of a strong blast wave is not an easy task since a
pressure rise higher than pNCJ is required in order to generate a self-propagating CJ wave.
For example, such a high pressure cannot be obtained by the combustion of a pocket of fresh
mixture of fixed volume, since, typically, pb /pu = Tb /Tu 10 for adiabatic combustion
while pNCJ /pu > 30. However, a detonation can be initiated inside a nonhomogeneous hot
spot, without need for the deposition of a high energy density in a tiny kernel. Such a spon-
taneous initiation was predicted by Zeldovich et al.[4,5] in a preconditioned medium with a

[2] Liñan A., et al., 2012, C. R. Mécanique, 340, 829–844.

[3] Clavin P., Williams F., 2012, Philos. Trans. R. Soc. London Ser. A, 370, 597–624.
[4] Zeldovich Y., et al., 1970, Acta Astronaut., 15, 313.
[5] Zeldovich Y., 1980, Combust. Flame, 39, 211–214.

gradient of induction time (e.g. produced by a temperature gradient). This mechanism was
also identified by Lee et al.[1] The research on this topic was still active more than 20 years
later.[2,3]

Criterion for Spontaneous Ignition (1970)

Suppose that the chemical energy is liberated by combustion after an induction delay τind .
Suppose also that a gradient of induction delays, dτind /dx > 0, has been created in a
reactive medium during a time scale shorter than the shortest induction time. This can be
done in the laboratory by heating the reactive mixture with a high-power laser pulse. Other
less intuitive possibilities mentioned in the literature are heating in the boundary layer
developing behind a strong shock propagating through a reactive mixture in a tube, and also
the fast turbulent mixing of strong hot jets injected into a reactive mixture. A propagating
induction front is generated by a gradient of induction delay; the slices of gas having a
longer induction delay react after those having a shorter induction delay. Roughly speaking,
a detonation is initiated when the velocity of the induction front, namely the inverse of the
gradient of induction delay, is equal to the propagation velocity of pressure disturbances
(sound speed au ) so that a runaway of the pressure occurs, leading quasi-spontaneously
to a CJ wave. For small induction gradients, the induction front propagates faster than
acoustic waves and combustion proceeds approximately at constant volume. For very steep
induction gradients the combustion proceeds at constant pressure. In both cases the pressure
increase remains small, well below pNCJ , and the reaction wave is systematically damped
out by the expansion waves developing behind the reaction region.
Quantitative insight is provided by a study of the coupling between acoustic waves and
heat release. Suppose that after the induction delay, τind (x), the release of chemical energy
takes place on a time scale shorter than that of acoustics, and also that the heat conductivity
can be neglected on this short time scale. Each slice of gas reacts, one after the other, at
constant volume without affecting the neighbouring slices. In planar geometry the gradient
of induction delays produces a distribution of energy release rate (per unit volume)

q̇v = qv (t − τind (x)) ,

where, by definition, the heat release rate (t) is positive for t > 0, is nullfor t < 0, has
the dimension of the inverse of the reaction time and is normed to unity, (t)dt = 1.
The time scale of the rate of heat release is usually of same order as the induction time
τind . To simplify the presentation we assume that it is shorter, τind (t) > 1. Consider
a constant gradient in planar geometry, dτind /dx = cst., τind = τind o + x(dτ /dx). The
ind
instantaneous distribution of q̇v results from an induction front propagating at constant a
speed (dτind /dx)−1 , giving rise to a source term in d’Alembert’s equation for the acoustic

[1] Lee J., et al., 1978, Acta Astronaut., 5, 971–982.

[2] Bartenev A., Gelfand B., 2000, Prog. Energy Combust. Sci., 26, 29–55.
[3] Kapila A., et al., 2002, Combust. Theor. Model., 6, 553–594.

pressure (2.5.1) or (15.2.21) of the form

q̇v = qv t − τind
o
− x(dτind /dx) , (4.3.28)
o is a constant. Suppose now that the propagation speed of the induction
where the delay τind
front is equal to the sound speed, assumed to be constant for simplicity,
(dτind /dx)−1 = a. (4.3.29)
The simple acoustic wave propagating in the same direction as the induction front, solution
to Equation (2.5.1),
∂ ∂

δp + a δp = (γ − 1)qv t − τind o
− x/a , (4.3.30)
∂t ∂x
has a secular term whose amplitude increases linearly in time:

δp = t(γ − 1)qv t − τind

o
− x/a . (4.3.31)
The synchronisation of a simple acoustic wave with the induction front thus leads to a rapid
increase of pressure (local runaway), on the time scale of energy release, which is short
compared with the acoustic time. A strong shock wave with a pressure peak sufficiently
high to ignite a CJ detonation quasi-instantaneously is created at the point where relation
(4.3.29) is verified. The increase of pressure is saturated by the damping due to the expan-
sion wave which develops behind the shock to match the conditions at the origin. For the
Arrhenius law (1.2.2) and a gradient of initial temperature characterised by a length scale
L ≡ Tu /|dTu /dx|, the criterion (4.3.29) yields the following condition for the temperature
in the fresh mixture, Tu∗ , at which spontaneous ignition occurs:
E τind (Tu∗ )a(Tu∗ )
≈ 1, (4.3.32)
kB Tu∗ L(Tu∗ )
−1
since τind dτind /dx = (E/kB T)T −1 dT/dx, valid for a large activation energy E/kB Tu∗ 1.
Therefore the condition in (4.3.32) is consistent with a characteristic time of heat release
( τind ) smaller than the acoustic time L/a as required for spontaneous initiation. Such
a separation of time scales is possible at a sufficiently high temperature Tu∗ . For a bell-
shaped distribution of initial temperature, the acoustic time L/a is too large near the centre
where the temperature gradient is small, limx→0 L = ∞. The pressure runaway can occur
at a distance from the centre where the temperature gradient is still sufficiently high, more
precisely at the location where L(Tu∗ ) becomes sufficiently small for the relation (4.3.32) to
be satisfied.
However some 20 years after the discovery of spontaneous ignition, an antagonistic
phenomenon was identified;[4] the temperature gradient that causes spontaneous ignition
of a detonation at high temperature can also lead to spontaneous quenching of the CJ
detonation at a lower temperature, showing that the conditions for ignition of a detonation
are a delicate compromise.

[4] He L., Clavin P., 1992, Proc. Comb. Inst., 24, 1861–1867.

Spontaneous Quenching by Thermal Gradient (1992)

The quenching mechanism is explained in simple terms by the rate of increase of the
induction length dind as TN decreases. In the quasi-steady-state approximation, when the
−1
rate of variation of dind is negligible compared with the inverse of the transit time, τind ≡
uN /dind , the induction length responds quasi-instantaneously to changes in the Neumann
temperature, and according to (4.3.14)

E 1 d E 1 d
1: dind ≈ − TN . (4.3.33)
kB TN dind dt kb TN TN dt
In the strong shock limit (γ − 1)Mu2 1, used for simplicity (see the relations (4.3.18)),
the variations of TN and D are simply related,
TN (γ − 1) δTN δD
≈ 2γ Mu2 ⇒ ≈2 , (4.3.34)
Tu (γ + 1)2 TN D
where the relation Mu2 ∝ D 2 /(cv Tu ) has been used. According to (4.2.44), the variation
of the CJ velocity DCJ with the temperature of the fresh mixture Tu is small for a fast
detonation,
δDCJ 1 1 δTu 1 δTu
= ≈ . (4.3.35)
DCJ 2Mu2CJ 1 + cp Tu Tu 2Mu2CJ Tu
(γ +1)qm /2

Then, for a large temperature sensitivity, βN ≡ E/kB T N 1, where the overbar denotes a
steady reference detonation, and for small variations of the detonation velocity, (D−D)/D
of order 1/βN , the induction length is a nonlinear function of the detonation velocity D that
takes a simple exponential form

dind β (D−D)
−2 N 1 d 1 dD −2 βN (D−D)
≈e D , dind ≈ −2βN e D , (4.3.36)
dind dind dt D dt
in agreement with (4.3.19). For small departures from the planar CJ wave in a temperature
gradient characterised by the length L, using the CJ wave at Tu as the reference and the
relation d/dt ≈ DCJ d/dx, Equations (4.3.35)–(4.3.36) yield,
. /
τ ind d βN DCJ dind −2 βN (D−D CJ )
dind ≈ 2
e DCJ
, (4.3.37)
dind dt MuCJ uNCJ L

where the relation dind = uNCJ τ ind has been used. According to (4.2.14), the quantity
(1/Mu2CJ )(DCJ /uNCJ ) is of order unity for (γ − 1)Mu2CJ = O(1). Therefore, for small depar-
tures from CJ waves, (D − DCJ )/DCJ of order 1/βN , the quasi-steady-state assumption
(τind /dind )(ddind /dt) 1 is satisfied when the temperature gradient is not too large on the
scale of the detonation thickness,

βN (dind /L) 1, (4.3.38)

a condition which is typically satisfied in experiments.

Hugoniot after
heat release

Hugoniot
of shock

Figure 4.11 Hugoniot diagram for a detonation propagating in a negative temperature gradient. The
line NB has a smaller absolute slope than the Rayleigh line NU showing that the mass flux through the
reaction zone is smaller than the mass flux through the lead shock. Reproduced from He L., Clavin
P., Proc. Comb. Inst., 24, 1861–1867. Copyright 1992, with permission from Elsevier.

The small rate of increase of the induction length (4.3.36), which is a consequence
of the changing conditions in the fresh mixture, induces a small difference of mass flux
across the lead shock and across the reaction layer, ρNCJ (d/dt)dind = 0. This is illustrated
by the numerical results.[1] They show that the inner structure of the detonation that is
formed in a temperature gradient is effectively in a quasi-steady-state, close to that of a
unperturbed CJ wave, but there are small differences that lead to the drastic quenching
effect in (4.3.40). Plotting the pressure p as a function of the specific volume 1/ρ through
the reaction zone, a straight line NB is obtained, linking the Neumann state (labelled N)
to the burnt gas (labelled B). As in CJ waves, this line is tangent to the equilibrium curve
for burnt gas corresponding to the local temperature Tu of the initial state of the fresh gas
in the temperature gradient; see Fig. 4.11. This shows that the exothermic reaction zone
(where the chemical energy is released) is in a steady state. However, the slope of the line
NU, linking the Neumann state to the initial state, pu , 1/ρu (labelled U) across the lead
shock wave, is slightly larger than the slope of the line NB, whereas the two lines NB
and NU are aligned in a CJ wave; see Fig. 4.4. This shows that the mass flux across the
inert lead shock is slightly larger than the mass flux across the exothermic zone, so that
the latter separates slowly from the lead shock on a time scale longer than the transit time
τind (TN ) = dind (TN )/uN . The fact that the exothermic zone is in full equilibrium while
the induction length is slowly increasing is due to the fact that the transit time across the
exothermic zone is substantially shorter than across the induction zone, essentially because
the flow velocity increases with the temperature.
The difference of mass flux corresponding to the difference of slopes of the two lines
NI and NU may be computed by a geometrical construction around the CJ wave as a linear
function of the departure of the shock velocity D from the CJ velocity DCJ , so that the
result is proportional to ρu (DCJ − D). The presentation simplifies in the limit of a strong

[1] He L., Clavin P., 1992, Proc. Comb. Inst., 24, 1861–1867.

shock, (γ − 1)Mu2 1: the coefficient of proportionality goes to unity in this limit.[1]

Thanks to this geometrical construction, which is a consequence of the particular quasi-
steady state of the inner structure of the detonation, the time derivative (d/dt)dind can be
computed in a way totally different from (4.3.36), without the need to consider the chemical
kinetics,
d (DCJ − D)
(γ − 1)Mu2 1: dind ≈ uNCJ . (4.3.39)
dt DCJ
Combining the two expressions (4.3.36) and (4.3.39) gives a nonlinear equation for the
local velocity D of the lead shock of the detonation propagating in the temperature gradient
Tu−1 |dTu /dx| ≡ L−1 (Tu ). According to (4.3.36) in which dind = τind (TNCJ )uNCJ , D = DCJ ,
and using again the relation dD/dt ≈ DCJ dDCJ /dx, since (D − DCJ )/DCJ is small, of
order 1/βN , the equation for X ≡ 2βN (DCJ − D)/DCJ takes the form

DCJ − D −2βN DCJ −D
dDCJ
2βN e DCJ
= K with K ≡ 4βN τind (TNCJ ) −
2
,
DCJ dx
(4.3.40)
where βN ≡ E/kB T N 1. This equation has a turning point for a critical value K = 1/e
corresponding to detonation quenching; there is no solution for D when K > 1/e. There
are two branches of solution for K < 1/e, but only the one going to DCJ in the limit K → 0
is physical. The detonation velocity at quenching is close to that of the CJ wave in the same
conditions, (DCJ − D)/DCJ = 1/(2βN ).
According to (4.3.35), the velocity gradient and thus the coefficient K in (4.3.40) may
be expressed in terms of the temperature gradient,
dDCJ 1 au βN2 au τind (TNCJ )
=− , ⇒ K=2 . (4.3.41)
dx 2MuCJ L(Tu ) MuCJ L(Tu )
A comparison with numerical simulations[2,3] (see Fig. 4.12) shows a good agreement
with the spontaneous quenching predicted by (4.3.40)–(4.3.41). The coefficient K may be
expressed in terms of the conditions for spontaneous ignition (labelled by ∗ ) by introducing
(4.3.32) into (4.3.41),
2 βN2 au τind (TNCJ ) L(Tu∗ )
K=
MuCJ (E/kB Tu∗ ) a∗u τind (Tu∗ ) L(Tu )
(4.3.42)
2 E Tu τind (TNCJ ) L(Tu∗ )
≈ ∗ ,
MuCJ kB TNCJ TNCJ τind (Tu∗ ) L(Tu )

where the relations MuCJ /Mu∗CJ ≈ Tu∗ /Tu and au /a∗u = Tu /Tu∗ have been used and
where Mu∗CJ is the Mach number of propagation of the CJ wave for a temperature of fresh
mixture equal to that at spontaneous initiation, Tu∗ .

[1] Clavin P., Williams F., 2012, Philos. Trans. R. Soc. London Ser. A, 370, 597–624.
[2] He L., Clavin P., 1992, Proc. Comb. Inst., 24, 1861–1867.
[3] He L., Clavin P., 1994, Proc. Comb. Inst., 25, 45–51.

Numerical
simulation

Theory

Initiation Propagation Quenching

(cm)

Figure 4.12 Maximum overpressure as a function of distance in a detonation initiated in a negative

temperature gradient of 600 K/cm satisfying conditions (4.3.29)–(4.3.32) for spontaneous initiation
of a detonation in a gradient. The dashed grey curve is the local CJ value. The black curve is the
prediction of (4.3.40)–(4.3.41) and shows the turning point. The solid grey curve is the result of direct
numerical simulation, showing initiation followed by quenching. Reproduced from He L., Clavin P.,
Proc. Comb. Inst., 24, 1861–1867. Copyright 1992, with permission from Elsevier.

Conclusion
A detonation which is initially spontaneously initiated by a hot pocket of fresh mixture can
be transmitted to the surrounding uniform medium only if the temperature profile of the
pocket is such that K is nowhere larger than 1/e. If this is not the case, the detonation
is quenched inside the hot pocket before reaching the uniform medium. For L(Tu ) =
cst., the condition for quenching by the same temperature gradient that is responsible for
spontaneous ignition at high temperature then takes the form
2 E Tu τind (TNCJ )
K≈ = 1/e. (4.3.43)
MuCJ kB TNCJ TNCJ τind (Tu∗ )
∗

This expression gives a relation between the Neumann temperature of the CJ wave at
quenching, TNCJ , and the temperature of fresh gas at spontaneous ignition, Tu∗ ; the tem-
perature gradient is involved only through τind (Tu∗ ) (see (4.3.32)).
For a detonation propagating into a region of decreasing temperature Tu near the tem-
perature of spontaneous initiation Tu∗ , defined in (4.3.32), with Tu Tu∗ , the transit time
τind (TNCJ ) across the induction zone of the CJ detonation increases quickly as Tu decreases,
since, according to (4.3.14), a small variation of TNCJ , of order 1/βN , produces a variation of
order unity of τind and thus also of K. For a constant temperature gradient, a rough estimate
of the critical condition for quenching is obtained from (4.3.43) as follows. In contrast with
τind , the variation of the coefficient (2/Mu∗CJ )(E/kB TNCJ )(Tu /TNCJ ) is not large and its value
is of order unity, so that quenching occurs when the transit time across the induction zone,
τind (TNCJ ), increases to reach a value comparable to the induction delay at spontaneous
ignition τind (Tu∗ ). This corresponds roughly to TNCJ close to Tu∗ , relating quenching to
spontaneous ignition. It is worthwhile stressing that the quenching mechanism described

here is not associated with the crossover temperature T ∗ studied in Sections 5.3 and 5.4,
TNCJ = T ∗ .
The quasi-steady-state approximation, which is the basic assumption in the above anal-
ysis, is not valid if the CJ wave is unstable. Therefore, the validity of the analysis is ensured
only if the thermal sensitivity is smaller than that of the instability threshold of galloping
detonations (pulsating instability) studied in Section 4.5.1. This is the case at sufficiently
high temperature Tu . Another limitation is the planar geometry. In spherical and cylindrical
geometry a critical radius, such as the one in Section 4.3.3, must also appear in the criterion
for spontaneous initiation.[1]

4.3.5 Deflagration-to-Detonation Transition (DDT)

In contrast to Section 4.3.3, we consider here mild energy discharges that initiate a sub-
sonic flame, also called a deflagration (see Section 2.4.2 for the condition of initiation),
and we discuss the deflagration-to-detonation transition (DDT). The basic DDT mech-
anisms are extremely complex. Despite more than a century of research, identification
of these mechanisms is not yet achieved, and, according to various reviews of the last
40 years,[2,3,4,5,6,7] full understanding of DDT remains one of the major challenges of
combustion theory; see also the special issue of the Philosophical Transactions of the
Royal Society A, (2012) 370, no. 1960, pp. 531–799. However, understanding of DDT
has improved recently thanks to theoretical analyses and accurate numerical solutions of
simplified models. We limit our presentation to a summary of the state of the art of the
knowledge, without entering into the detailed analyses, the simplest ones being the most
instructive.

Phenomenology of DDT
To the best of our knowledge, in the modern literature, DDT is not reported for spher-
ical flames propagating in free space filled with a quiescent and uniform gaseous com-
bustible mixture. Such a transition was reported in early Soviet experiments[8] around
1950, concerning very energetic mixtures such as acetylene–oxygen mixtures (UL ≈ 5–
10 m/s) enclosed in a soap bubble or a rubber sphere. The phenomenon was attributed to
flame acceleration by front wrinkling, due either to intrinsic instabilities (self-turbulising
flames) or to the interaction between the flame and shock waves reflected from the boundary
delimited by the bubble. It is not clear whether or not it was a DDT or a direct initiation, as
discussed in Section 4.3.3. However, severe explosion events are reported in large vapour

[1] He L., Law C., 1996, Phys. Fluids, 8(1), 248–257.

[2] Oppenheim A., Soloukhin R., 1973, Ann. Rev. Fluid Mech., 5, 31–58.
[3] Lee J., 1977, Ann. Rev. Phys. Chem., 28, 75–104.
[4] Lee J., Moen I., 1980, Prog. Energy Combust. Sci., 6, 359–389.
[5] Lee J., Berman M., 1997, Advances in Heat Transfer, 29, 59–126.
[6] Lee J., 2008, The detonation phenomenon. Cambridge University Press.
[7] Ciccarelli G., Dorofeev S., 2008, Prog. Energy Combust. Sci., 34, 499–550.
[8] Shchelkin K., Troshin Y., 1965, Gasdynamics of combustion. Baltimore, Md.: Mono Book Corp.

clouds.[9] They are not fully explained, even though obstacle-generated turbulence (e.g. the
trees) is thought to be the principal ingredient.
Since the discovery of detonation waves in gases by Berthelot and Vieille (1881) and
Mallard and Le Chatelier (1883), it is known that, for many combustible mixtures ignited
at the closed end of a tube, a flame ignited by a weak spark can accelerate to a detonation.[2]
For sensitive fuel oxygen mixtures involving acetylene, ethylene or hydrogen in tubes
with cross sections of a few tens of square centimetres, the onset of detonation occurs
abruptly after a flame travel typically of the order of a metre and acceleration to a velocity
above 300 m/s (in the laboratory frame). For most hydrocarbon–air mixtures the transition
distance (if it exists in smooth tubes) is much larger, exceeding the length of laboratory-
scale experiments.[3]
Flame acceleration and preconditioning of the fresh mixture ahead of the flame due to
compression waves are key features of DDT in an open-end tube ignited at the closed end.
The early experimental detonation studies were performed in smooth tubes. However, as
early as 1926, obstacles (orifice plates) were placed in tubes to promote flame acceleration
up to few hundred metres per second.[5,7]
Different mechanisms of DDT have been proposed since 1940. The mechanism of
flame acceleration still gives rise to some controversy. More likely, several mechanisms
promoting DDT are involved in real experiments, not a single one. Before discussing the
orders of magnitude and the experimental results, it is convenient for pedagogical reasons
to present a qualitative overview of the general ideas on DDT that have emerged from
experiments, analyses and, more recently, from numerical studies.

Turbulence-Induced DDT
Between 1940 and 1945 Shchelkin proposed that flame acceleration in tubes is governed
by the turbulent fluctuations of the flow in the fresh mixture ahead of the flame, leading
to an increase of the flame surface; see Section 3.2. The general idea is as follows. For
flame ignition at the closed end of a tube, a flow of fresh mixture is generated upstream
from the flame front, due to expansion of the burnt gas; the flame acts as a semi-permeable
piston (see Section 2.2.1). For a sufficiently large Reynolds number, the expansion-induced
flow becomes turbulent and a coherent feedback mechanism can develop: the turbulence-
induced increase of propagation speed of the wrinkled flame brush (3.1.11) leads to an
increase of the flow velocity and then of the turbulent intensity that in turn increases the
turbulent flame speed. The nonslip condition at the wall creates transverse gradients of
the longitudinal flow velocity. In tubes without obstacles the turbulence develops in the
boundary layers at the wall and is strongly affected by wall roughness. Obstacles facilitate
the development of turbulence and DDT.

[9] Bradley D., et al., 2012, Philos. Trans. R. Soc. London Ser. A, 370, 544–566.

Localised Thermal Explosion (1973)

Another feedback mechanism, of a quite different nature, involving gas compressibility,
also develops. As recalled in Section 15.3 for the case of a piston, the fresh mixture is
put into motion by a quasi-planar compression wave propagating upstream from the flame.
After a sufficiently long time in sufficiently long tubes a shock wave formed, somewhere
upstream, depending on the details of flame acceleration. Due to this compression wave,
the temperature on the cold side of the flame front increases with time and a longitudinal
temperature gradient develops to match the initial conditions far upstream. This increase
of temperature in the fresh mixture has two effects: it increases the laminar flame speed,
thus producing a feedback mechanism for flame acceleration, and it shortens the induction
delay. If the temperature increase is sufficient, typically 1000 K, self-ignition can occur
ahead the flame.
There is now a general consensus that such temperature gradients can lead locally
to spontaneous formation of detonation by a gradient of induction delay (Zeldovich’s
mechanism, described in Section 4.3.4). This phenomenon was called ‘localised thermal
explosion’ or ‘explosion in the explosion’.[1,2] As discussed later in more quantitative
terms, such a mechanism, based on compressibility-induced preconditioning of the fresh
mixture, is possible only for a very fast flame brush, (deflagration) propagating at a velocity
above many hundreds of metres per second.
In insulated tubes, heating is reinforced in the boundary layers at the wall of the tube
where the transverse velocity gradient produces a temperature gradient by viscous dissipa-
tion of mechanical energy. The faster the longitudinal flow, the stronger is such a dissipation
mechanism. However, heat loss at the wall is expected to weaken the dissipation-induced
DDT mechanism.

Friction Effect (2000–2015)

In a more recent series of accurate numerical simulations in simple configurations, still in
open-end tubes, it has been shown by Sivashinsky and co-workers[3,4] that DDT can be
produced by the increase of temperature due to the friction-induced adiabatic compression.
To be more specific, in the presence of sufficiently strong friction, the dominant effect is due
not necessarily to viscous dissipation, described by the last term in the right-hand side of
(15.1.37) or (15.1.61), but to the first term in the right-hand side of (15.1.27) or (15.1.34).
This mechanism has been first schematised by a planar model for shockless initiation of
detonation, called hydraulic resistance.[3,4] In this model, a volumetric friction force is
introduced in the momentum equation. As a result, a shockless temperature gradient devel-
ops in the fresh mixture and DDT can be observed even when the dissipative counterpart
of friction in the energy equation is artificially turned off.

[1] Oppenheim A., Soloukhin R., 1973, Ann. Rev. Fluid Mech., 5, 31–58.
[2] Ciccarelli G., Dorofeev S., 2008, Prog. Energy Combust. Sci., 34, 499–550.
[3] Brailovsky I., Sivashinsky G., 2000, Combust. Flame, 122, 492–499.
[4] Brailovsky I., et al., 2012, Philos. Trans. R. Soc. London Ser. A, 370, 625–646.

Hot gas Cold gas

at rest at rest

Flame Shock
brush wave

Figure 4.13 Sketch of the one-dimensional runaway configuration showing the folded turbulent flame
brush and associated preceding shock wave.

More surprising, when a cut-off temperature T ∗ is introduced in the reaction rate so that
the chemical heat cannot be released for T < T ∗ , DDT can be still observed in numerical
simulations of simplified flame models even when the increase of temperature in the fresh
mixture just ahead of the the flame is smaller than T ∗ , ruling out the Zeldovich mechanism
of a localised thermal explosion in the fresh mixture.[5,6] A few years previously, DDT
with a chemically frozen flow adjacent to the flame below T ∗ ≈ 550 K was observed
experimentally for flames propagating at a high velocity, typically 500 m/s or even more,[7]
in highly sensitive stoichiometric hydrogen–oxygen and ethylene–oxygen mixtures at ordi-
nary initial temperature Tu = 293 K and for different initial pressures pu = 0.5–0.75 bar.
Even though a Zeldovich mechanism cannot be fully excluded inside the flame structure,
a different DDT mechanism seems to be involved. It could be related to the 26-year-old
analytical work of Deshaies and Joulin,[8] presented below, showing that there is an upper
bound for the propagation velocity of a flame brush, a phenomenon observed later in direct
numerical simulations.[9]

Runaway Mechanism (1989)

A DDT mechanism, different from the localised thermal explosions induced by a ther-
mal gradient, was identified by considering a simplified configuration,[8] namely the one-
dimensional self-similar solution of a weak shock at constant velocity D > au , generated
by a turbulent flame brush propagating at constant velocity, Utur , from the closed end of
a tube; see Fig. 4.13. Due to thermal sensitivity, the small increase of flame temperature
by the compressible effects, namely here by the weak shock, produces a velocity increase
of the flame brush leading to a constructive feedback. The burnt gas being at rest, the
velocity of the flame brush is given by (3.1.11) when UL is replaced by Ub = (ρu /ρb )UL ,
Utur = Ub s, where s ≡ S /So is the degree of folding and Ub is subsonic. As shown below,
there is a critical degree of folding s∗ (critical velocity of the flame brush) above which there
is no self-similar solution[8] (a shock wave followed by a flame, both at constant velocity;
see Fig. 4.14). If s increases for any reason, for example because of flame instabilities or

[5] Kagan L., Sivashinsky G., 2014, In G.S. Roy, S. Frolov, eds., Transient Combustion and Detonation Phenomena, Torus
Press, Moscow.
[6] Kagan L., et al., 2015, Proc. Comb. Inst., 35, 913–920.
[7] Kuznetsov M., et al., 2010, Combust. Sci. Technol., 182(1628-1644).
[8] Deshaies B., Joulin G., 1989, Combust. Flame, 77, 201–212.
[9] Gamezo V., et al., 2011, In U. Irvine, ed., Proceedings of 23rd ICDERS, 24–29.

some positive feedback with the upstream turbulent flow, a runaway of the velocity of the
flame brush with a sudden increase in the flame temperature is expected to occur when s
reaches s∗ . This could correspond to an abrupt transition to detonation. Further analytical
studies of the runaway will improve the understanding of DDT, for example by considering
a slow increase of s, meaning slow on the time scale of the acoustic waves.
Assuming adiabatic conditions (insulated walls) the temperature of the gas just ahead
of the flame brush is the temperature TN behind the shock. According to the Rankine–
Hugoniot conditions (4.2.14) and (4.2.16) for weak shock
D − UN 2 TN 2(γ − 1) 2
(Mu2 −1) 1: ≈ (M 2 −1), −1 ≈ (Mu −1), (4.3.44)
au γ +1 u Tu γ +1
the temperature of the gas ahead of the flame, TN , can be expressed in terms of the velocity
of the flame brush Utur as follows. The flow velocity ahead of the flame brush, namely
downstream of the weak shock, is v = D − UN , where UN denotes the flow velocity
relative to the shock at the Neumann state. Neglecting the small difference between ρN
and ρu , mass conservation across the flame brush gives also v = (1 − ρb /ρu )Utur . Using
the relation D − UN = (1 − ρb /ρu )Utur , Equations (4.3.44) lead to an expression for the
increase of temperature in the fresh mixture in terms of the velocity of the flame brush Utur
and the speed of sound au :

TN ρb Utur
− 1 = (γ − 1) 1 − .
Tu ρu au
Using (8.2.3), Tb = Tu + qm /cp , δTb /Tb = (ρb /ρu )δTu /Tu , the small increase of flame
temperature due to the presence of the shock wave takes the form

Tb ρb TN ρb ρb Utur
−1= − 1 = (γ − 1) 1 − , (4.3.45)
Tbo ρu Tu ρu ρu au
where Tbo is the flame temperature for the initial temperature Tu ahead of the shock. The
thermal sensitivity of the laminar flame speed Ub (2.1.9) yields

E Tb −Tbo
E/2kB To 1: Ub /Ubo ≈ e 2kB Tbo Tbo
, (4.3.46)
and the relation Utur = sUb then leads to a nonlinear equation for X ≡ K(Ub /Ubo ) when
(4.3.45) is introduced into (4.3.46),
E (Tbo − Tu ) ULo
Xe−X = K, K ≡ s(γ − 1) , (4.3.47)
2kB Tbo Tbo au
where ULo = (ρb /ρu )Ubo is the laminar flame speed for the initial flame temperature Tu
and K is a nondimensional parameter proportional to the folding s.
The solution Ub /Ubo to (4.3.47), plotted versus K, that is also the velocity of the flame
brush Utur versus the folding s, has two branches of solution with a turning point for the
critical value K = 1/e. There is no solution for K > 1/e and there are two solutions for
K < 1/e; see Fig. 4.14. For ordinary hydrocarbon–air mixtures with a flame velocity UL
about 0.35 m/s, and for E(Tb − Tu )/kB Tb2 ≈ 8, γ = 1.4, the critical value for the degree

Figure 4.14 Plot of the solution to (4.3.47) for the folded flame brush velocity. There is no solution
beyond a critical degree of folding, proportional to K.

of folding is about 2 × 102 . For very energetic mixtures such as a stoichiometric ethylene–
oxygen mixture, UL ≈ 10 m/s, the critical degree of folding is about 8. This corresponds to
a velocity of the flame brush in the laboratory frame, Utur , about 500 m/s. A preliminary
dynamical study of the problem was performed by solving the acoustic problem in the
burnt gas.[1] The result shows that the lower branch Ub− of solution is stable and the upper
branch Ub+ unstable.

Flame Acceleration
Numerical studies have shown that one of the key ingredients for DDT, namely flame
acceleration, can be produced by mechanisms other than flow turbulence:

• A first mechanism is the flame front folding by the different instabilities described in
Sections 2.2 to 2.5, especially if the initial flame is initiated in a quasi-planar geometry.
• Another possibility is the Richtmyer–Meshkov type of instability in Section 2.8.3 if shock
waves are reflected by the walls and impinge the flame.
• One may also invoke the initial flame elongation just after ignition by a spark at the centre
of the closed end of the tube, described in (2.8.8). This leads to an exponential increase of
the flame surface area at a rate of the order 2Ub /R, where R is the radius of the tube. This
acceleration lasts only during a short lapse of time after ignition, ≈ R/Ub . However, for
a very energetic reactive mixture, such as a stoichiometric mixture of hydrogen–oxygen,
UL ≈ 10 m/s, Ub ≈ 85 m/s, (Ub − UL ) ≈ 75 m/s, the flow velocity can reach a velocity
greater than 550 m/s (e2 ≈ 7.39) at the end of this short delay.
• A subsequent flame acceleration has also been considered. It is associated with the two-
dimensional character of a parallel flow that is generated ahead of the flame by expansion
of the burnt gas. Due to viscous effects (nonslip boundary condition at the wall), this
parallel flow is nonuniform in the transverse direction, zero at the wall and maximum at
the centre of the tube. The flame then takes an elongated form with a surface increasing
with time. Assuming a quasi-isobaric approximation and a laminar flow, an approximate
analysis[2] predicts an exponential increase in time of the propagation velocity of the

[1] Deshaies B., Joulin G., 1989, Combust. Flame, 77, 201–212.
[2] Bychkov V., et al., 2005, Phys. Rev. E, 72(4), 046307.

flame tip with a growth rate ≈ UL /R, smaller than (2.8.8) but lasting in principle indefi-
nitely. Further analyses predict that compressible effects in the burnt gas can saturate the
growth rate.[1]
• An antagonistic effect can be produced if the boundary layer at the wall becomes tur-
bulent and accelerates the flame propagation near the wall, leading to a tulip-shaped
flame.
• As already mentioned, a constructive feedback can also be produced by an increase of
temperature ahead of the flame due to compression waves in the presence of friction
without shock formation.[2] In addition the role of viscous dissipation cannot always be
neglected, especially in narrow channels; see the end of this section.
During the last decade a huge number of numerical studies have investigated these
different possibilities.[3,4,5,6,7,8,9,10,11,12] Each of these numerical simulations succeeded in
producing a DDT, but it is hard to say to what extent the real experimental phenomena
are represented. It is probable that a runaway phenomenon similar to the one described by
Deshaies-Joulin[13] is involved in most of these simulations, whatever be the mechanism of
flame acceleration.

DDT Experiments in Ordinary Tubes

A review[14] of the DDT experiments in tubes of typically 5 cm transverse size was pre-
sented in 2008, prior the 2010 experiment already mentioned[15] ; see also the book of
Lee.[16]
Much information was already obtained from the pioneering Soviet experiments during
the first half of the last century.[17] About 20 years later, the sequence of high-resolution
stroboscopic Schlieren photographs by Oppenheim and coworkers[18,19,20] represents
a milestone in the study of DDT. These experiments concern mainly a stoichiometric
hydrogen–oxygen mixture at atmospheric pressure (UL ≈ 10 m/s) in a rectangular tube

[1] Valiev D., et al., 2009, Phys. Rev. E, 80, 036317.

[2] Kagan L., et al., 2015, Proc. Comb. Inst., 35, 913–920.
[3] Kagan L., Sivashinsky G., 2003, Combust. Flame, 134, 389–397.
[4] Liberman M., et al., 2006, Int. J. Transp. Phenomena, 8, 253–277.
[5] Oran E., Gamezo V.N., 2007, Combust. Flame, 148, 4–47.
[6] Kagan L., Sivashinsky G., 2008, Combust. Flame, 154, 186–190.
[7] Valiev D., et al., 2008, Phys. Lett. A, 372, 4850–4857.
[8] Kessler D., et al., 2010, Combust. Flame, 157, 2063–2077.
[9] Akkerman V., et al., 2010, Phys. Fluids, 22, 053606.
[10] Ivanov M., et al., 2011, Phys. Rev. E, 83, 056313.
[11] Dzieminska E., et al., 2012, Combust. Sci. Technol., 184, 1608–1615.
[12] Ivanov M., et al., 2013, J. Hydrogen Energy, 38, 16427–16440.
[13] Deshaies B., Joulin G., 1989, Combust. Flame, 77, 201–212.
[14] Ciccarelli G., Dorofeev S., 2008, Prog. Energy Combust. Sci., 34, 499–550.
[15] Kuznetsov M., et al., 2010, Combust. Sci. Technol., 182(1628-1644).
[16] Lee J., 2008, The detonation phenomenon. Cambridge University Press.
[17] Shchelkin K., Troshin Y., 1965, Gasdynamics of combustion. Baltimore, Md.: Mono Book Corp.
[18] Urtiew P., Oppenheim A., 1966, Proc. R. Soc. London Ser. A, 295(13-28).
[19] Meyer J., et al., 1970, Combust. Flame, 14(1), 13–20.
[20] Oppenheim A., Soloukhin R., 1973, Ann. Rev. Fluid Mech., 5, 31–58.

Flame

Shock wave

Flame Boundary layer

Shock wave

Flame

Figure 4.15 Formation of a tulip flame by acceleration in the flame-induced turbulent boundary layers
along the walls. The full curve u(x) represents the velocity profile of the induced flow ahead of the
flame, and the dashed line schematises the width of the boundary layer (not to scale).

of cross section 9.7 cm2 with a test section located about 3 m away from the igniter at the
closed end. The sequence of events has been established as follows:

1. Initial acceleration of a wrinkled flame generating shock waves

2. A long period of turbulent flame propagation following the formation of a tulip shape
3. Transition to detonation triggered by a local explosion within the shock–flame complex,
most of the time either near the wall or at the flame tip.

Following the early work of Shchelkin, it has been established experimentally by dif-
ferent authors that flame acceleration is strongly affected by wall roughness: the larger
the roughness, the faster the growth of the turbulent boundary layers, leading to flame
acceleration close to the walls, as sketched in Fig. 4.15. For a wall roughness of 1 mm
in a square channel of 5 cm height, the formation of a tulip-shaped flame, propagating at
more than 300 m/s, has been clearly observed before onset of detonation.[21] Propagation
is no longer possible below a critical radius of the tube. Just above this radius, spinning
detonations are observed. They were first observed in cylindrical tubes as early as 1926.[17]
In this marginal regime, burning is concentrated around a triple point rotating around the
tube axis, of the same type as that described later in Section 4.4.2. This phenomenon
appears when the tube diameter is of the order of the typical size of the cellular structure of
the detonation front; see Section 4.5 for a discussion of the cellular detonations. Spinning
detonations are more easily observed when approaching the detonation limit, namely when
the Neumann temperature approaches the crossover temperature; see the beginning of
Section 4.3. This phenomenon can be easily understood by noticing that the length of
the induction zone increases near the detonability limit so that transverse propagation of
detonation waves is possible inside the induction zone, leading to a spinning detonation
front even in conditions for which the planar detonation cannot exist.

[21] Kuznetsov M., et al., 2005, Shock Waves, 14(3), 205–215.

Lead
shock
Hugoniot
of shock

Michelson–
Rayleigh lines
Hugoniot after
heat release

Figure 4.16 Sketch in the (v , p) plane (v ≡ 1/ρ) showing the CJ detonation point, BCJ , and the CJ
deflagration point, FCJ , characterised by the lower tangency point of a Michelson–Rayleigh line with
the Hugoniot of the reacted gas.

Choked Regime of Turbulent Flame

In long tubes of large diameter, equipped with regularly spaced orifice rings, turbulent
flames propagating at a steady velocity of the order of the speed of sound in the combustion
products, 600–1000 m/s, have been observed.[1] This fast propagation regime of turbulent
flames, called the ‘choked regime’, is sometimes considered as the so-called CJ deflagration
speed, determined by the lower tangency point of the Michelson–Rayleigh line with the
Hugoniot curve and characterised by a decrease of pressure in the combustion products;
see Fig. 4.16. For a convenient blockage ratio in an obstacle-laden tube, turbulence can
sustain flames in the ‘choked regime’ without transition to detonation, while such fast
flames normally transit systematically to a detonation in smooth tubes. In the absence of a
lead shock wave, the fresh mixture is not warmed up and DDT cannot be produced. Another
interpretation of the choked regime is based on the multiplicity of the friction-controlled
propagation regimes.[2,3]

DDT Experiments in Capillary Tubes (2007)

For ordinary fuel–air mixtures (UL ≈ 0.5 m/s), flames cannot propagate in capillarity
tubes because they are quenched by thermal loss at the wall; see the detailed study in
Section 8.5.1. According to the rough estimates in (8.5.1) and (8.5.4), the nondimensional
parameter characterising the critical heat loss for flame quenching is D2T /(βN R2 UL2 ), where
βN is the reduced activation energy (8.2.8), so that more energetic mixtures (faster flames)
have a smaller quenching radius R. Motivated by gaseous combustion in porous media, by
possible applications of microscale combustion to energy production and by numerical and

[1] Lee J., Berman M., 1997, Advances in Heat Transfer, 29, 59–126.
[2] Brailovsky I., Sivashinsky G., 2000, Combust. Flame, 122, 492–499.
[3] Sivashinsky G., 2002, Proc. Comb. Inst., 29, 1737–1761.

theoretical studies of the last decade,[2,4] recent experiments[5] have shown that stoichio-
metric ethylene–oxygen flames (UL ≈ 10 m/s) can also accelerate to DDT in smooth-
walled capillary tubes with diameters as small as 2 mm or even 0.5 mm. The onset of
detonation occurs at a distance about 100 times the diameter. This mixture is very energetic
and sensitive. The typical size of the cellular structure is about 0.7 mm and the critical
diameter is below 0.3 mm. Three typical scenarios were observed:

• Abrupt transition to a CJ detonation (DCJ = 2373.5 m/s), typically 100 μs after ignition
during turbulent flame propagation at 300 m/s (Reynolds number Re ≈ 4 × 104 )
• CJ detonation initiation followed by extinction
• Flame acceleration to a constant speed about 1600 m/s (choked regime?).
Other behaviours have been observed[6] for different equivalence ratios, for example an
oscillating flame mode with a propagation velocity between 5 and 10 m/s in a 2 mm tube
filled with lean mixtures (conditions for which DDT does not occur), galloping detonations
(spinning?) travelling at 2000 m/s, and also flame or detonation quenching in rich mixtures.

Orders of Magnitude. Discussion

Whatever be the basic mechanism responsible of DDT, runaway or localised explosion,
the fact that the onset of detonation in tubes is observed systematically for a propagation
velocity (in the laboratory frame) of the flame brush always greater than 300 m/s can be
understood as follows. First, as already mentioned, the critical velocity of the flame brush
involved in the runaway mechanism (4.3.47) of Deshaies and Joulin[7] is about 500 m/s in
the laboratory frame. Second, a localised explosion through Zeldovich’s mechanism also
requires such a large flame velocity as shown now. Roughly speaking, according to this
scenario, the temperature of the fresh mixture ahead the flame must be sufficiently high
for the induction (ignition) time to be sufficiently small. Because of the drastic increase
of the induction time below the crossover temperature T ∗ , an ignition time, sufficiently
short for the onset of detonation, requires a relatively high temperature and therefore a
large flow velocity. For example, the ignition time of a stoichiometric hydrogen–oxygen
mixture[8] at 1 atm is 10 s at 800 K, about 2×10−1 s at 900 K, 10−4 s at 1000 K and 10−5 s at
1250 K. Heating the fresh mixture to a temperature above 1000 K by compression through
a shock wave and/or an isobaric compression wave is possible only for a sufficiently high
propagation velocity of the flame brush. The temperature in an adiabatic compression wave
may be evaluated using the relation a/ao = (T/To )1/2 and the Riemann invariant (15.3.47)
of a simple wave in planar geometry, (T/To )1/2 = 1 + [(γ − 1)/2] (u/ao ), where the
subscript o denotes the initial state of the fresh mixture To ≈ 273 K, p = 1 atm and γ ≈ 1.4.

[4] Bychkov V., et al., 2005, Phys. Rev. E, 72(4), 046307.

[5] Wu M., et al., 2007, Proc. Comb. Inst., 31, 2429–2436.
[6] Wu M., Wang C., 2011, Proc. Comb. Inst., 33, 2287–2293.
[7] Deshaies B., Joulin G., 1989, Combust. Flame, 77, 201–212.
[8] Sanchez A., Williams F., 2014, Prog. Energy Combust. Sci., 41, 1–55.

For a flow velocity of the order of the sound speed, u ≈ ao , this leads to a gas temperature
below 500 K, not high enough for ignition. Also, the temperature increase behind the shock
wave generated by such a flow is too low. According to the Rankine–Hugoniot relations
(4.2.14)–(4.2.16), a shock wave with Mu = 2 in a quiescent gaseous mixture at ordinary
conditions requires a gas velocity in the laboratory frame, (D − UN ), larger than the initial
sound speed and corresponds to a temperature of the compressed gas also less than 500 K.
Therefore, according to the typical laws controlling ignition delay, the compressibility-
induced increase of temperature for a flow velocity of the order of the sound speed is not
high enough to produce a detonation.
So, for a flow velocity of few hundred metres per second, an additional heating mech-
anism must be involved. Denoting the velocity of the fresh mixture by u and the local
sound speed by a, an evaluation of the heating rate by viscous dissipation using entropy
production, recalled in (15.1.60), yields (ν/δ 2 )(u2 /a2 ), where ν = μ/ρ is the viscous
diffusion coefficient and δ is the thickness of the laminar boundary layer that develops
downstream from the shock. The latter grows with time as δ 2 ≈ νt. Discarding heat loss
at the wall, the relative temperature increase of a fluid particle inside the boundary layer
after the transit time t from the leading edge is thus predicted to be of order (u/a)2 . Thus,
for a flow velocity of the order of the sound speed, the cumulative effect of viscous dissi-
pation and compressibility could create sufficient heating to produce a local explosion and
transition to detonation. DDT by a localised explosion through the Zeldovich mechanism
cannot occur ahead of the flame for smaller flow velocities. Moreover, for flow velocity
of the order of a, the Reynolds number is large, for example already greater than 104 in a
1 mm capillarity tube, and so turbulent boundary layers develop in larger tubes, as observed
in experiments.[1] A friction-driven increase in temperature, resulting from the existence of
the boundary layer in a compressible flow, could also play a role.
The remaining problem to be understood is flame acceleration to a velocity higher than
300 m/s during the period after flame ignition when the induced flow is still laminar and
the lead shock too weak to produce a significant increase in temperature. For very energetic
mixtures (UL ≈ 10 m/s) this is not difficult to understand since, as already explained, the
initial quasi-isobaric acceleration (2.8.8), at a rate 2Ub /R, following a point ignition at the
closed end of the tube, is sufficient to rapidly produce an upstream flow velocity exceeding
500 m/s.
For hydrocarbon–air mixtures (UL < 0.5 m/s) in smooth tubes, DDT is generally not
observed for a tube length less than 30 m. Flame acceleration leading to detonation in
such conditions (if any) is much more difficult to decipher. The quasi-isobaric elongation
mechanism of the flame shape, investigated by Bychkov et al.[2] leading to a parallel flow
velocity increasing exponentially in time with a growth rate of the order UL /R, could be

[1] Kuznetsov M., et al., 2005, Shock Waves, 14(3), 205–215.

[2] Bychkov V., et al., 2005, Phys. Rev. E, 72(4), 046307.

relevant in capillary tubes (R ≈ 1 mm), as shown in numerical simulations.[3] However, the

experiments[4] were not able to confirm the corresponding tubular flame structure.
To summarise, DDT in smooth tubes is understood for energetic mixtures such as stoi-
chiometric acetylene–oxygen or hydrogen–oxygen (UL ≈ 10 m/s); however, the situation
is not so clear for much less energetic mixtures such as ordinary fuel–air mixtures (UL <
0.5 m/s).
To conclude this section on DDT, recent large-scale experiments in methane–air mix-
tures (UL ≈ 0.38 m/s), performed in the context of explosion in coal mines,[5] are worth
mentioning. Detonations in such conditions have been ignited by a very strong explosion
at the closed end of a tube filled with very lean mixtures near to the limit where a planar
detonation cannot be initiated in ordinary tubes. The wave takes the form of a spinning
detonation. The following explanation could be relevant: just downstream of a planar shock
near to (or even just below) the detonability limit, the Neumann temperature is too low to
sustain an ordinary planar detonation, but a detonation wave can propagate in the transverse
direction into the shocked gas, forming triple point structures whose trajectory is a spiral.

4.4 Dynamics of Shock Fronts

The multidimensional dynamics of ordinary detonation waves is strongly associated with
that of the inert lead shock. It is thus useful to begin with a study of the latter.

4.4.1 Linear Stability of Shock Waves

It has been known for a long time that gaseous planar shock waves are stable. Experimental
and theoretical pioneering studies[6,7,8] reported power law relaxations of initial distur-
bances. Typically, the initial disturbances were produced either when the shock wave meets
wedges on the wall of the tube or when the shock is reflected normally from a perturbed
flat wall. Attention was focused on acoustic disturbances propagating in the shocked gas.
Moreover, stroboscopic Schlieren photographs, taken to study the relaxation, also showed
the formation of Mach stems (triple points) propagating in the transverse direction on the
shock front. Power laws cannot be obtained by a normal-mode analysis; it is necessary
to solve the initial-value problem using the Laplace transform method.[9,10] Nevertheless,
it is instructive to start with a normal-mode analysis because it illustrates the physical
mechanisms in the simplest terms. This topic was first investigated by D’yakov[11] and

[3] Valiev D., et al., 2013, Phys. Fluids, 25, 096101–16.

[4] Wu M., et al., 2007, Proc. Comb. Inst., 31, 2429–2436.
[5] Oran E., et al., 2015, Combust. Sci. Technol., 187, 324–341.
[6] Lapworth K., 1959, J. Fluid Mech., 6, 469–480.
[7] Briscoe M., Kovitz A., 1968, J. Fluid Mech., 31(3), 529–546.
[8] Van-Mooren K., George A., 1975, J. Fluid Mech., 68(1), 97–108.
[9] Erpenbeck J., 1962, Phys. Fluids, 5(10), 1181–1187.
[10] Bates J., 2004, Phys. Rev. E, 69, 056313.
[11] D’yakov S., 1954, Zh. Eksp. Teor. Fiz., 27, 288.

Kontorovich[1] for arbitrary media. Earlier references, some of them published in the form
of reports, as well as recent works, may be found in the modern literature.[2,3] Here we
focus attention on polytropic gases, including both inert and reactive shock waves. The
normal-mode analysis is straightforward but its interpretation is laborious. The details of
the calculation are presented in the second part of the book (Section 12.1), in a simpler way
than in the original works and in textbooks.[4] In this section we limit the presentation to
the method of solution and to a discussion of the main results.

Formulation and Method. Acoustic Wave and Entropy–Vorticity Wave

The shock wave is considered as a hydrodynamic discontinuity of zero thickness. To sim-
plify the notation we consider a two-dimensional geometry. The extension to three dimen-
sions is straightforward. Consider a planar shock wave propagating in the negative x direc-
tion at a constant (supersonic) velocity D in a uniform medium. This requires an external
device to trigger the shock, for example a piston at constant velocity in the shocked gas.
We assume that such a piston is at infinity and does not generate disturbances. This is
called an ‘isolated shock’ in the literature. In the reference frame of the unperturbed planar
solution, the unperturbed front stands perpendicular to the x-axis at x = 0 and the shocked
material flows at a constant (subsonic) velocity uN > 0 in the positive x direction. Let
x = α(y, t) represent the perturbed position of the shock front at the transverse position y
and at time t. For any physical quantity f we introduce the decomposition f = f +δf , where
f represents the unperturbed solution. The wave being supersonic, the upstream medium is
unperturbed and is assumed to be uniform. The x and y components of the flow velocity
in the shocked gas, written in the reference frame of the unperturbed shock, are denoted
by u = uN + δu and w = δw. The flow is considered as inviscid and heat conduction is
neglected. The perturbed flow δu, δw, δρ and δp is the solution of the four linearised Euler
equations (12.1.1)–(12.1.3) in which the entropy equation (12.1.3) is introduced since the
entropy is modified at the Neumann state of the wrinkled shock front. The material can be
characterised by an equation of state written in the entropy form. Expressing the entropy
in terms of the density and pressure, s(p, ρ), the sound speed is a = ∂p/∂ρ|s=cst. . The
boundary conditions are given at x = 0 by the linearised Hugoniot relations, (12.1.25)–
(12.1.26), and a boundedness condition is enforced downstream in the shocked material for
x → ∞.
Generally speaking, the disturbance of the flow, the solution to the linearised Euler
equations around a uniform state, is decomposed into two different linear modes: acous-
tic waves, solutions to d’Alembert’s equation (12.1.5), and an isobaric incompressible
rotational flow, called a vorticity wave, which is simply convected downstream at the
unperturbed flow velocity, along with the entropy disturbance; see (12.1.7)–(12.1.8). The
method of solution is as follows. The acoustic waves in the shocked gas are solved in terms

[1] Kontorovich V., 1957, Zh. Eksp. Teor. Fiz., 33, 1525.
[2] Bates J., 2007, Phys. Fluids, 19, 094 102–1–6.
[3] Bates J., 2012, J. Fluid Mech., 691, 146–164.
[4] Landau L., Lifchitz E., 1986, Fluid mechanics. Pergamon, 1st ed.

of the dynamics of the shock front using the pressure disturbances at x = 0 given by the
Rankine–Hugoniot conditions. The usual acoustic relations then yield the flow velocity in
the acoustic waves, and in particular the value at the Neumann state, x = 0. The vorticity
wave is then computed using its value at x = 0, which is given by the difference between
the total flow velocity and the flow velocity in the acoustic wave; see (12.1.16)–(12.1.17).
An equation for the evolution of the wrinkles on the front is then obtained by imposing
that the vorticity wave is an incompressible flow and that the acoustic flow is bounded at
infinity.

Dispersion Relation from a Normal-Mode Analysis

For a harmonic perturbation of the front position, a normal-mode decomposition is used,
as in (2.2.1)–(2.2.2),

α(y, t) = α̃(t)eiky , α̃(t) = α̂eσ t , δf (x, y, t) = f̃ (x)eiky+σ t , (4.4.1)

where k is the transverse wave vector (a real quantity), σ = s + iω is a complex whose real
part s is the linear growth rate (or damping rate if s < 0) and the imaginary part ω is the
frequency of oscillation in time. Power laws α̃(t) ∝ tν can be obtained for s = 0 by using
the Laplace transform, as discussed later.
Following the method outlined above, the solution takes the form of an equation for σ in
terms of k, called the dispersion relation. In the absence of other reference length and time
scales (the shock being considered as a hydrodynamic discontinuity), the complex linear
rate σ must be proportional to the inverse of the wavelength times a velocity, namely the
unperturbed shock velocity, σ ∝ D|k|. Considering the shocked material in the reference
frame of the unperturbed wave, the natural velocities are the sound speed aN and the
flow velocity uN = M N aN , both of which are proportional to D with a coefficient of
proportionality involving the Mach number M N ; see for example the Rankine–Hugoniot
relations (4.2.14)–(4.2.17) for a polytropic gas. The dispersion relation then takes the form
of an equation for σ/(aN |k|), or more precisely for S,

±2M N S 1 + S2 = (1 + r)S2 + (1 − r)n, (4.4.2)
2
where (1 − M N )1/2 S ≡ σ/(aN |k|) (4.4.3)

(see Section 12.1.2), and where r and n are nondimensional parameters characterising the
material and the strength of the shock wave,
2
(ρu D)2 ρN MN
r≡− > 0, n≡ . (4.4.4)
dpN /dρ −1
N
ρu 1 − M 2
N

The parameter r involves the mass flux across the planar wave ρu D and the slope of the
Hugoniot curve (4.2.24), at the Neumann state, in the plane (p, 1/ρ). For usual gases 0 <
r < 1; see Fig. 4.2. The parameter n > 0 involves the density ratio ρ N /ρu and the Mach
number of the flow of shocked gas at the Neumann state of the unperturbed planar wave,

Unstable Neutral Stable

Non-
Radiating radiating

Figure 4.17 General stability limits for shock waves; σ is the complex linear growth rate. The critical
value r∗ is given in (12.1.28).

M N < 1. The square-root term on the left-hand side of (4.4.2) is a pure pressure term,
whereas the right-hand side involves the vorticity wave. The sign must be chosen such that
the downstream acoustic waves are bounded; see the text below (12.1.12). The solutions
to (4.4.2) are also roots of the quadratic equation for S2 , obtained by taking the square of
(4.4.2). Only the roots that satisfy both Equation (4.4.2) and the boundedness condition at
infinity should be retained to represent the physical solution.

Relation with the Initial-Value Problem. Power Laws

In parameter space, the regions of instability, Re(σ ) > 0, and of linear stability, Re(σ ) < 0,
are separated by a wide region (of finite size) where the normal modes are neutral, Re(σ ) =
0; see Fig. 4.17. This region corresponds to intermediate values of r, −(1+2M N ) r r∗ ,
where r∗ is defined in (12.1.28). All the normal modes of gaseous inert shock waves and
of reactive gaseous shocks (detonations without modifications to the inner structure), are
neutral; see Sections 12.1.3–12.1.5.
Neutrality of modes does not necessarily mean that the amplitude of initial disturbances
does not evolve in time in the linear approximation. It only excludes exponential growth
or damping. Neutral modes may well evolve with power laws in time or with even more
complicated laws. This can be better understood by considering the linear solution to the
initial-value problem obtained by Laplace transform (with respect to time) ᾰ(z),
∞ s+i∞
1
ᾰ(z) ≡ e−zt α̃(t)dt, α̃(t) = ezt ᾰ(z)dz,
0 2iπ s−i∞
where s is taken to be the largest imaginary part of all the singularities in ᾰ(z). A neutral
mode represents a pure oscillation of an initial harmonic disturbance of the shock front
(neither damped and nor amplified) only if it corresponds to a simple pole z = iω of ᾰ(z)
on the imaginary axis of z. Denoting by F(σ ) = 0 the dispersion relation obtained by the
normal-mode analysis, the Laplace transform of the amplitude of the wrinkles takes the
form ᾰ(σ )/α̃(0) = N(σ )/F(σ ), where N(σ ) is a function of the complex variable that has
no other singularity than a discontinuity through a branch cut.[1,2] Due to the √presence of
acoustics, according to (4.4.2), the function F(σ ) involves a square-root term, S2 + 1, that
introduces a branch cut in the complex plane for σ . Therefore, purely imaginary roots of

[1] Bates J., 2007, Phys. Fluids, 19, 094 102–1–6.

[2] Bates J., 2012, J. Fluid Mech., 691, 146–164.

the dispersion relation (neutral normal modes) do not necessarily correspond to undamped
oscillations of harmonic disturbances of the shock front. This may be illustrated by√simple
mathematical
√ examples; the inverse Laplace transforms of 1/zν , n! /(z−a)1+n , 1/ z2 + 1
and 1/[ z2 + 1 + z] are, respectively, t(ν−1) / (ν), tn eat , J0 (t) and J1 (t)/t, where J0 (t)
and J1 (t) are the zero-order and first-order Bessel functions of the first kind.[3] The third
example shows that the roots σ = ±i do not yield a simple harmonic oscillation. It turns out
that the initial disturbances of fronts in the domain (n−1)/(n+1) < r < r∗ are damped,[1,4]
with an asymptotic dependence for t → ∞ as t−3/2 times an oscillatory function.

Spontaneous Emission of Sound. Acoustic Wave Reflexion

The stability of shock waves is characterised by neutral normal modes; see Section 12.1.3.
The stability analysis was discussed initially in a different way by considering the reflexion
of planar acoustic waves impinging on the front from the shocked gases.[5,6] This analysis
is presented in Section 12.1.4. The front is considered to be unstable if there exist particular
incident waves for which the reflexion coefficient diverges. This is the case when the
acoustic wave of a neutral normal-mode is ‘radiating’, that is, when it propagates from
the shock front towards infinity in the shocked gases. This situation is called ‘spontaneous
emission of sound’. More precisely, the divergence of the reflexion coefficient occurs when
the reflected acoustic wave matches the radiating normal mode. Such a divergence never
occurs if all acoustic modes are nonradiating, that is, when they propagate from infinity
towards the shock front. Therefore the instability threshold is defined by the critical con-
dition separating the cases for which all acoustic modes are nonradiating from those for
which one, at least, of the acoustic modes is radiating (spontaneous sound emission). The
corresponding critical value for r, (n − 1)/(n + 1), in the middle of the range of neutral
modes in Fig. 4.17 is considered to be the stability limit in parameter space. The fronts
characterised by neutral modes are unstable for r < (n − 1)/(n + 1) and stable in the
opposite situation. Inert gaseous shock waves are stable since they have only neutral modes
with nonradiating acoustic waves, (n − 1)/(n − 1) < r < r∗ ; see Section 12.1.4.

Concluding Remarks
The nonradiating acoustic waves of the normal modes of an ‘isolated’ shock could lead one
to question the physical relevance of such modes, since no energy in the compressed gas
can be sent to the front from infinity (causality condition). The shock–vortex interaction
helps to clarify the role of normal modes with nonradiating acoustic waves. If the turnover
velocity of a subsonic vortex of size L in the initial gas is small compared with the shock
velocity D, the lapse of time, τint = L/D, during which the shock crosses the vortex is
shorter than the turnover time. After τint , the shock front is weakly wrinkled and propagates
into a quiescent medium. The transmitted vortex emits acoustic waves in every direction

[3] McQuarrie D., 2003, Mathematical methods for scientists and engineers. University Science Books.
[4] Majda A., Rosales R., 1983, SIAM J. Appl. Math., 43(6), 1310–1334.
[5] Kontorovich V., 1957, Zh. Eksp. Teor. Fiz., 33, 1525.
[6] Fowles G., 1981, Phys. Fluids, 24(2), 220–227.

Shock Slip
wave line
Secondary
shock
Secondary
shock
Slip Shock
line wave

Figure 4.18 Structure of triple points propagating in the positive and negative y-directions.

(backward and forward) in the compressed gas. This is also the case when a shock front is
perturbed by crossing a bump on the wall of the tube. The normal modes with nonradiating
acoustic waves may then be useful to describe the evolution of the wrinkles of the shock
front for t > τint .
After a sufficiently long time, nonlinear effects dominate the linear dynamics of the
wrinkled shock front. The linear analysis is a preliminary step before performing a nonlin-
ear study of the multidimensional dynamics of shock and detonation fronts.

4.4.2 Formation of Mach Stems

The formation of singularities could be anticipated by Huygens’ construction shown in Fig.
2.10 for flames. However, this is not exactly the mechanism at work in shock waves. The
Mach stems (triple points), which are systematically observed on weakly perturbed shock
fronts, are formed by a similar but different mechanism. In agreement with the oscillation
frequency of the normal modes, these triple points propagate in the transverse direction
at a phase velocity close to the speed of sound in the shocked gas. The triple points are
constituted by a cusp of the shock front, a slip line and a secondary shock in the shocked
gas; see Fig. 4.18. The formation of singularities of the slope of the front is important for
understanding the cellular patterns of detonations, which are also characterised by triple
points; see Section 4.5.2. Patterns similar to those of cellular detonations are also observed
on inert shock waves reflected normally from an undulated wall;[1] see the Schlieren pic-
tures in Fig. 4.19, extracted from a recent experiment.[2] Triple points are also formed
during the shock–vortex interaction, reproduced in two-dimensional geometry by direct
numerical simulations for about 20 years.[3,4] A recent simulation[5] is presented in Fig.
4.20. From the theoretical point of view the first analyses focused attention on the sound
wave generated by shock–vortex interaction.[6]

[1] Briscoe M., Kovitz A., 1968, J. Fluid Mech., 31(3), 529–546.
[2] Biamino L., et al., 2011, Pattern of triple points on a shock wave reflected from an undulated wall, private communication.
[3] Guichard L., et al., 1995, AIAA J., 33(10), 1797–1802.
[4] Ellzey J., et al., 1995, Phys. Fluids, 7(1), 172–184.
[5] Lodato G., Vervisch L., 2014, DNS of shock-vortex interaction using spectral difference high-order methods, private
communication.
[6] Ribner S., 1985, AIAA J., 23(11), 1708–1715.

(a) (b) (c)

Figure 4.19 Schlieren images showing a regular pattern of Mach stems propagating in the transverse
direction on a shock front after reflexion on a undulated wall. The reflected shock propagates from
right to left. Shock Mach number: 1.1. Time between pictures: 10−4 s, extracted from a film at a
frame rate of 106 image/s. Courtesy of L. Biamo, G. Jourdan and L. Houas, IUSTI Marseilles.

(a) (b)

Figure 4.20 Direct numerical simulation of shock–vortex interaction in two-dimensional geometry.

(a) Pressure field; the initial gas is on the left and the shocked gas on the right, pressure ratio 4.5. (b)
Density gradient obtained by a simulated Schlieren method. Shock Mach number: 2. Rotation Mach
number of the initial cylindrical vortex: 0.8. Courtesy of G. Lodato and L. Vervisch 2014, CORIA
Rouen.

Triple Points
The formation of cusps of the shock front is more easily analysed in two-dimensional
geometry. The linear flow of the compressed gas helps us to understand both the secondary
shock and the slip line of the Mach stem. According to the linear analysis, the flow takes
the form (12.1.7)–(12.1.8)

δu = δu(i) (y, t − x/uN ) + δu(a) (x, y, t),

δw = δw(i) (y, t − x/uN ) + δw(a) (x, y, t),

where the superscripts (i) and (a) denote the incompressible vorticity wave and the acoustic
wave, respectively; see Fig. 4.21. Using the normal mode decomposition in (4.4.1) and the
notations of (12.1.10) and (12.1.37), the flow in the progressive waves of a normal mode
may be written

δu(i) = ũ(i) eik[y−(ω/uN k)x+(ω/k)t] , δw(i) = w̃(i) eik[y−(ω/uN k)x+(ω/k)t] ,

δu(a) = ũ(a) eik[y+(l/k)x+(ω/k)t] , δw(a) = ũ(i) eik[y+(l/k)x+(ω/k)t] ,

Normal mode

dary
Nonradiative Secon
acoustic wave s ck
h o

Slip
line

Vorticity wave Mach stem

(shear flow) (triple point)

Figure 4.21 Flow of shocked gas associated with a progressive normal mode of a wrinkled shock
front propagating into a quiescent gas (two-diimensional geometry). The picture shows the upward
propagation of the progressive wave associated with a harmonic disturbance of the front. The
symmetric picture is obtained for the progressive disturbance propagating downwards. The shear
flow in the vorticity wave is indicated by the dashed lines; the location of the velocity extrema of
the nonradiative acoustic wave is indicated by the dotted lines. The sketch on the right shows Mach
stems formed by the nonlinear evolution of the progressive disturbance propagating upwards. To
leading order in the limit (4.4.9) the phase velocity, ω/k, of the wrinkles on the front is equal to the
sound speed in the shocked gas aN .

where ω and l are positive; see (12.1.36)–(12.1.38). The waves travelling in the opposite
y-direction are obtained by changing the sign of y. The vorticity wave being incompressible,

∂δu(i) /∂x + ∂δw(i) /∂y = 0, ∂δu(i) /∂x = −(1/uN )∂δu(i) /∂t, (4.4.5)

it is a shear flow since the velocity vector is aligned with the iso-velocity line, w̃(i) /ũ(i) =
ω/(uN k); see Fig. 4.21.
If, for any reason, a cusp is formed on a wrinkle of the shock, the corresponding part
of the shear flow degenerates into a slip line with a slope ω/uN k, and a secondary shock
with a slope close to l/k is created on the acoustic wave. The triple point associated with
these progressive waves then propagates in the transverse direction on the front with a
phase velocity close to ±ω/k; see Fig. 4.18. The difficult problem is to decipher the basic
mechanism of cusp formation: is it created by wave breaking of the acoustics? Or, inversely,
is the secondary shock a consequence of cusp formation? The intriguing relation between
the formation of cusps on the shock front and that of the secondary shock is described
analytically below in a limiting case.

Generally speaking, the pressure and density jumps at the front x = α(y, t) are obtained
when the expression for the local Mach number of the propagation in the normal direction

D + δu1f − α̇t − δw1f αy
Mu = 1/2 , α̇t ≡ ∂α/∂t, αy ≡ ∂α/∂y, (4.4.6)
2
au 1 + αy
is introduced into (4.2.14)–(4.2.15). Here the subscript f denotes the value at the front,
and a disturbance of the upstream flow (δu1 , δw1 ) has been retained in view of the study
of the shock–vortex interaction. Conservation of mass (15.1.45) and transverse momentum
(15.1.47) yield (see also (12.1.21)),

ρuf D + δu1f − α̇t − δw1f αy = ρN uN + δuN − α̇t − δwN αy (4.4.7)

δw1f + D + δu1f αy = δwN + (uN + δuN ) αy , (4.4.8)
where ρu ≡ ρ u + δρ1 , ρN ≡ ρ N + δρ and (δuN , δwN ) are the modifications to the
components (in the frame attached to the planar and unperturbed front) of the flow at the
Neumann state (shocked gas at the shock front). Equations (4.4.6)–(4.4.8) are general; they
are limited not by the amplitude of wrinkling compared with the wavelength, but only by
the approximation of a shock wave considered as a hydrodynamic discontinuity separating
two inviscid gas flows.

Strong Shock in the Newtonian Approximation

Except for a semi-phenomenological geometrical approach[1] and a weakly nonlinear anal-
ysis for radiating acoustic waves,[2] there was no analysis of the Mach stem formation
on wrinkled shock fronts until recently when a nonlinear analysis was performed[3] for a
strong shock M u 1 in the Newtonian limit (γ − 1) 1,
2
M u 1, M u (γ − 1) = O(1). (4.4.9)
According to the Rankine–Hugoniot relations (4.2.14)–(4.2.17), the flow in the shocked
gas is strongly subsonic in this limit,
2 2
M N ≈ (γ − 1)/2 + 1/M u , (4.4.10)
and a perturbation analysis can be performed using the Mach number in the shocked gas as
the small parameter, 2 ≡ γ MN2 1,
−1 2
≈ (aN /au )M u = aN /D, (aN /au )2 ≈ [2 + (γ − 1)M u ]/2 = O(1), (4.4.11)
2
uN /D = ρ u /ρ N ≈ , 2
pN /pu ≈ Mu = O(1/ ).2
(4.4.12)
−2
Both quantities M u > 0 and (γ − 1) > 0 are of order 2 , (γ − 1)/(2 2 ) < 1. The
perturbation analysis in the limit (4.4.9) provides a qualitatively good physical insight into

[1] Whitham G., 1957, J. Fluid Mech., 2(02), 145–171.

[2] Majda A., Rosales R., 1983, SIAM J. Appl. Math., 43(6), 1310–1334.
[3] Clavin P., 2013, J. Fluid Mech., 721, 324–339.

the formation of Mach stems. In the linear approximation, the jump relations (4.2.14)–
(4.2.15) with (4.4.6) take the form

2

δpN δp1f δu1f − α̇t δρN δρ1f au δu1f − α̇t

− ≈2 , − =2 , (4.4.13)
pN pu D ρN ρu aN D
2
[(γ − 1)M u − 2]

(δuN − α̇t ) = 2
δu1f − α̇t , δwN ≈ Dαy + δw1f , (4.4.14)
2M u
where, for simplicity, some unimportant terms of order 2 have been omitted in the first
equation (4.4.13) and in the second equation (4.4.14).
To leading order in the limit (4.4.9), the dispersion relation (4.4.2), in the form (12.1.29)
for a polytropic gas, reduces to S2 ≈ −1, S ≈ ±i, σ 2 + a2N k2 = 0, ω ≈ aN k, corresponding
to a wave equation for the wrinkles on the front

ω ≈ aN k, ∂ 2 α/∂t2 − a2N ∂ 2 α/∂y2 = 0. (4.4.15)

This is because the flow in the acoustic wave is negligible compared with the incom-
pressible shear flow in this limit, as shown now. According to (12.1.15) and (12.1.37),
l/|k| = O(), the order of magnitude of the acoustic wave is

δu(a) = O 2 δpN /ρ N uN , δw(a) = O δpN /ρ N uN . (4.4.16)

According to (4.4.13) and (4.4.14),

(δuN − α̇t ) = O( 2 α̇t ), δwN = O(α̇t /), δpN /pN = O(α̇t /D). (4.4.17)

The order of magnitude of the acoustic wave is then obtained by introducing (4.4.17) into
(4.4.16) using pN ≈ ρ N uN aN /M N and aN /M N = (aN /au )D/(M u M N ) = O(D):

δpN /(ρ N uN ) = O(α̇t ) ⇒ δu(a) = O( 2 α̇t ), δw(a) = O( α̇t ).

The comparison with (4.4.14), δuN ≈ α̇t , δwN ≈ Dαy = O(α̇t /), shows that the ratio
of the acoustic flow to the flow in the vorticity wave is of order 2 , the latter being a shear
flow quasi-parallel to the unperturbed shock front,

|δu(a) /δu(i) | = O( 2 ), |δw(a) /δw(i) | = O( 2 ), |δu(i) /δw(i) | = O().

Therefore, to leading order, the linear dynamics of a wrinkled shock wave propagating
in a quiescent gaseous medium is controlled by the vorticity wave. The vorticity wave is
obtained from the leading order of the linearised Rankine–Hugoniot relations, (4.4.13)–
(4.4.14),

δu(i) |x=0 = δuN ≈ α̇t , δw(i) |x=0 = δwN ≈ Dαy ,

to give δu(i) = α̇t (y, t − x/uN ), δw(i) = Dαy (y, t − x/uN ). (4.4.18)

The wave equation (4.4.15) is then obtained directly from the incompressible condition
(4.4.5) using (4.4.11)–(4.4.12) in the form DuN ≈ a2N . The transverse propagation, at the

sound velocity a2N of disturbances on the front, is produced by the tilted shear flow. Com-
pressible effects (acoustic waves) in the shocked gas are not involved in this phenomenon.

Weakly Nonlinear Analysis for the Formation of Folds

In the same conditions and for sufficiently small amplitude of the wrinkles, acoustic waves
are also negligible in the nonlinear dynamics of a weakly wrinkled shock propagating in
a quiescent medium. Denoting the quadratic terms U and W, the Euler equations in the
compressed gas can be written in the form
∂u ∂u 1 ∂p ∂w ∂w 1 ∂p
+ uN =U− , + uN =W− ,
∂t ∂x ρ ∂x ∂t ∂x ρ ∂y
∂u ∂u ∂w ∂w
−U ≡ δu +w , − W ≡ δu +w ,
∂x ∂y ∂x ∂y
where δu ≡ u − uN and U = W = 0 in the unperturbed flow. These equations are
completed by the equation for mass conservation. A weakly nonlinear approximation is
valid when the quadratic terms U and W are small compared with the unsteady terms.
When expressed in terms of the vorticity wave (4.4.18) and using the relation DuN ≈ a2N ,
the quantities U and W take the form
1 ∂H 1 D ∂H
U≈ , W≈− , (4.4.19)
2 ∂x 2 uN ∂y

where H ≡ [−α̇t2 (y, t − x/uN ) + a2N αy2 (y, t − x/uN )].
The relative order of magnitude of these source terms compared with the linear terms is ε ≡
|αy |/ ≈ |α̇t |/uN . A perturbation analysis using the small parameter ε 1 is thus valid
for sufficiently small amplitudes of the wrinkles |αy | . The acoustic waves introduce
quadratic terms that are smaller by at least a factor . The weakly nonlinear analysis is thus
performed in the limits
1 and ε ≡ |αy |/ ≈ |α̇t |/uN 1. (4.4.20)
The source terms U and W in (4.4.19) satisfy the same equations as the linear vorticity
wave, ∂U/∂t +uN ∂U/∂x = 0 and ∂W/∂t +uN ∂W/∂x = 0, so that they would introduce
a secular contribution (a term growing linearly in time) to u and w if they were not balanced
by a pressure term. However, it is not necessary to carry out this calculation here, since the
source terms U and W are zero for simple progressive waves, α̇t = ±aN αy ⇒ H = 0. This
is because the shear flow associated with the progressive waves of normal modes is an exact
solution of the incompressible Euler equations. More generally, for an initial disturbance of
the front of finite size L, the source terms U and W cannot influence the front geometry,
since, after a short finite time, the term H vanishes for t > L/aN , namely when the two
progressive waves (propagating up and down) no longer overlap. These terms cannot play a
significant role in the ultimate formation of cusps on wrinkles of small amplitude. However,
this is not true for initial wrinkles whose amplitude is of the same order as the wavelength,

which is a case beyond the scope of the present analysis. To leading order of the weakly
nonlinear analysis, the Euler equations then reduce to their linear form and the flow takes
the same form as in the linear analysis, (12.1.7)–(12.1.8),

δu ≡ u − uN = u(i) (y, t − x/uN ) + u(a) (x, y, t), (4.4.21)

w = w(i) (y, t − x/uN ) + w(a) (x, y, t), (4.4.22)

where the pressure fluctuations are fully taken into account by the acoustic flow (u(a) , w(a) ),
and the flow (u(i) , w(i) ) is incompressible, ∂u(i) /∂x + ∂w(i) /∂y = 0.
In the laboratory frame of the unperturbed front, the quadratic terms that introduce
corrections of order ε to the dynamics come from the boundary conditions at the front.
The latter are obtained by introducing (4.4.6)–(4.4.8) into (4.2.14)–(4.2.15)
pN α̇t
= 1 − 2 − αy ,
2
(4.4.23)
pN D

a2u α̇t
+ αy = wN αy , wN = Dαy − δuN αy ,
2
(δuN − α̇t ) 1 − 2 (4.4.24)
a2N D

where the notation δuN ≡ uN − uN has been used and where, for simplicity, some unim-
portant terms of order uN /D ≈ 2 in the linear approximation (4.4.13)–(4.4.14) have been
neglected in (4.4.24). They do not change the nonlinear dynamics but introduce corrections
to the linear dynamics that are not useful to describe cusp formation in a weakly nonlinear
analysis. Anyway they are negligible in the intermediate regime 3 < |αy | < . Limiting
attention to nonlinear corrections of order ε, the boundary conditions in (4.4.24) for the
flow velocity in the shocked gas reduce to

x = α: δu ≡ δuN (y, t) ≈ α̇t + Dαy2 , w ≡ wN (y, t) ≈ Dαy , (4.4.25)

where the terms of order 2 and the quadratic term in the expression (4.4.24) for wN ,

−δuN αy , which is of order |α̇t |αy = aN αy2 , have been neglected. The latter corresponds to
a correction of order |αy | = 2 ε relative to the linear approximation. The only correction

term of order ε ≡ |αy |/ is Dαy2 , D|αy2 /α̇t | = O(|αy |/) since |α̇t | = O(aN αy ) and
D/aN = O(1/). The shift of the front position, induced by the wrinkling, also introduces
quadratic terms into the boundary values at the origin

x = 0: δu ≡ uf (y, t) ≈ δuN − αux , w ≡ wf (y, t) ≈ wN − αwx , (4.4.26)

where ux (y, t) ≡ ∂u/∂x|x=0 , and wx (y, t) ≡ ∂w/∂x|x=0 . The coupling between the acoustic
wave and the vorticity wave is negligible since it introduces nonlinear contributions that are
smaller than those in (4.4.25) by a factor . The correction to the pressure in (4.4.23) being
even smaller, the flow (u(a) , w(a) ) is negligible in front of (u(i) , w(i) ), to leading order in the
weakly nonlinear analysis. Consequently the boundary conditions at x = 0 for (u(i) , w(i) )
are given by (4.4.26). The incompressible condition ∂u(i) /∂x + ∂w(i) /∂y = 0 then leads to
a nonlinear equation for the wrinkles in the form −u−1 N ∂uf /∂t+ ∂wf /∂y = 0. As shown

just below, the nonlinear terms coming from the shift of origin in (4.4.26) do not contribute
to this equation, so that the nonlinear equation for the front wrinkles reduce to
1 ∂uN ∂wN
− + = 0, (4.4.27)
uN ∂t ∂y
where uN and wN are given in (4.4.25). This is shown by computing

−uN ∂(αux )/∂t = −uN α̇t ∂u(i) /∂x|x=0 + α∂ 2 u(i) /∂x2 |x=0 ,
∂(αwx )/∂y = αy ∂w(i) /∂x|x=0 + α∂ 2 w(i) /∂x∂y|x=0 .

According to the incompressible condition, the sum of the last terms in the right-hand side

of these two equations is zero. It is also the case for the first terms for H ≡ a2N αy2 − α̇t2 = 0
when u(i) and w(i) in the quadratic terms are replaced by the linear approximation (4.4.18),
∂δu(i) ∂δw(i) D 1 ∂H
−uN α̇t + αy = α̇t α̈t − α α̇ ≈ = 0.
∂x ∂x uN y ty 2 ∂t
According to (4.4.25) and (4.4.27), the weakly nonlinear equation for the evolution of the
shock is then[1]

∂ 2α 2
2 ∂ α ∂ ∂α 2
− a N + D = 0, (4.4.28)
∂t2 ∂y2 ∂t ∂y

where the relation a2N ≈ DuN , valid to leading order in the limit (4.4.9), has been used.

The ratio of the nonlinear term to the linear terms is effectively of order ε, Dαy2 /|α̇t | ≈
(D/aN )|αy | ≈ ε since |α̇t | ≈ aN |αy |. This result can also be obtained when working in the
frame attached to the front; see Section 4.6.2. Equation (4.4.28) has two time scales: a short
time τs ≡ L/aN , where L is the wavelength of the wrinkles, and a longer time τl ≡ τs /ε
for the effect of the nonlinear term. This is more easily seen by introducing the reduced
time scale and length scale τ ≡ t/τs , η ≡ y/L, and the reduced amplitude of order unity,
A ≡ α/(εL) (see (4.4.20)),

∂2A ∂2A ∂ ∂A 2
− +ε = 0. (4.4.29)
∂τ 2 ∂η2 ∂τ ∂η
This equation can be written in a parameter-free form by introducing εA, but the form
(4.4.29) is convenient to point out the two-time-scale nature of the problem. For any smooth
initial wrinkle of amplitude of order unity, A = O(1), the slope ∂A/∂η develops a
singularity (a corner of the front) after a finite time on the long time scale, τ = τ/ε.
Considering a simple progressive wave η = η ± τ , A = A(η , τ ), Equation (4.4.29) takes
the form of Burger’s equation (without dissipation) for the slope, A ≡ ∂A/∂η ,

∂A /∂τ + A ∂A /∂η = 0, (4.4.30)

[1] Clavin P., 2013, J. Fluid Mech., 721, 324–339.

known to produce a discontinuity after a finite time; see Section 15.3.1. This describes in
simple terms the basic mechanism of Mach stem formation on gaseous shock waves in the
limit (4.4.9).

Concluding Remarks
To summarise, Mach stem formation on a strong shock in the Newtonian approximation is
associated with the formation of folds (lines of singularity of slopes) on the shock front. The
nonlinear mechanisms responsible for their formation come from the boundary condition
at the front (4.4.7) for the vorticity wave. They are different from Huygens’ construction
even though the nonlinear term in the equation controlling the weakly nonlinear dynamics
of front has the same form. More precisely, the dominant nonlinear effect comes from
the term wN αy in the normal component of the flow velocity at the Neumann state. This

introduces a corrective term Dαy2 to α̇t ; see (4.4.25). By comparison, a term Dαy2 /2 would
have been introduced by Huygens’ construction for a front propagating at a constant normal

velocity Un = D in a quiescent medium, Un ≡ (D − α̇t )/(1 + αy2 )1/2 ≈ D − α̇t − Dαy2 /2;
see (10.1.6).
In the limit (4.4.9) the acoustic waves introduce smaller corrections that could be
obtained by pushing the weakly nonlinear analysis to the next order. Such an extension of
the perturbation analysis is necessary to describe the secondary shock wave. This has not
yet been done.
Numerical simulations[1] of (4.4.29) or, more precisely, its parameter-free form,
using an initial harmonic disturbance of the shape of the front show qualitatively good
agreement with the experiments reported in Fig. 4.19. Moreover, an interesting coalescence
property is also observed when the simulation is started using a white noise disturbance
of the shape of the front: the formation of a large-scale structure delimited by folds
is observed on the front.[1] This is similar to the large-scale structure of the universe
obtained, following the Zeldovich model[2] with a multidimensional form of Burgers’
equation.[3]

4.4.3 Shock–Vortex and Shock–Turbulence Interaction

As already mentioned, the shock–vortex interaction has been extensively studied by
numerical simulation in two dimensions. From the theoretical point of view the analyses
focused attention on the sound wave generated by shock–vortex interaction.[4] Impressive
direct numerical simulations for the shock–turbulence interaction have been performed
by the Standford group.[5] In their work the attention is focused on the transmitted
turbulence in the shocked gas. On the theoretical side, extensive analytical studies have

[1] Denet B., et al., 2015, Combust. Sci. Technol., 187(1-2), 296–323.
[2] Shandarin S., Zeldovich Y., 1989, Rev. Mod. Phys., 61(2), 185–222.
[3] Gurbatov S., et al., 2012, Sov. Phys.–Uspeki, 55(3), 223–249.
[4] Ribner S., 1985, AIAA J., 23(11), 1708–1715.
[5] Larsson J., Lele S., 2009, Phys. Fluids, 21, 126101.

Weak acoustic
pulse

Vo ave
r tic
w
ity
Transmitted
Initial vortex
vortex Strong
acoustic
burst

Figure 4.22 Sketch of the shock–vortex interaction. Reproduced from Clavin P., 2013, J. Fluid Mech.,
721, 324–339, with permission.

been carried out in the linear approximation.[6] A nonlinear analytical study has been
recently performed in the limiting case (4.4.9) and for very subsonic velocity fluctuations
of the upstream flow[7] , ve /au . The objective of this study, summarised below, is
to set up a simple model equation for the dynamics and geometry of the shock front;
see (4.4.41).

Interaction of a Strong Shock and a Weak Vortex

Consider a cylindrical and very subsonic vortex of diameter L and turnover velocity
ve /au (vortex strength: = π Lve ). The axis of the vortex is parallel to the planar
shock front and perpendicular to the plan Oxy. The interaction is sketched in Fig. 4.22.
The velocity of the vortex centre, relative to the upstream gas, will be neglected compared
with D. The analysis is performed below for a vortex of finite extension, L. It can be
extended to take account of the long tail of real vortices.[7] The method of solution for a
strong shock in the Newtonian approximation (4.4.9) can be summarised as follows. In the
limiting case under investigation, a small distortion of the shock front is generated during
the short lapse of time τint = L/D, taken by the vortex to cross the shock (interaction
time). The amplitude of the wrinkle is much smaller than its transverse extension L, the
2
order of magnitude of the slope of the wrinkle being of order ve aN /D ; see (4.4.38). After
the interaction time, t > τint , the wrinkled shock front propagates in a quiescent medium.
According to the previous linear analysis, the wrinkles on the front are then expected to
propagate in the transverse direction at a velocity aN . This introduces a time scale of order
L/aN that is larger than τint by a factor 1/ in the limit (4.4.9). Because of this difference
of time scales, the initial condition for the subsequent nonlinear evolution of the front
wrinkling t > τint is, roughly speaking, provided by a linear analysis of the vortex–shock
crossover, 0 < t < τint .

[6] Wouchuk J., et al., 2009, Phys. Rev. E, 79(066315).

[7] Clavin P., 2013, J. Fluid Mech., 721, 324–339.

Solution during the Short Period of Crossover

The small short-lived perturbations of the upstream flow, δu1f , δw1f , δp1f , to be introduced
into the linear jump conditions (4.4.13)–(4.4.14), are

δu1f (y, t) = ue |x=−Dt , δw1f (y, t) = we |x=−Dt , δp1f (y, t) = pe |x=−Dt , (4.4.31)

where ue (r), we (r), pe (r) denote the turnover flow and pressure of the vortex and where the
origin of time t = 0 is the beginning of the shock–vortex crossover.
Consider first the pressure disturbances that are generated in the compressed gas during
the vortex–shock crossover. To leading order in the limit (4.4.9), the linear equations of
acoustics in the compressed gas yield
2
∂u(a) 1 ∂p ∂w(a) 1 ∂p ∂ 2p ∂ p ∂ 2p
≈− , ≈− , ≈ a 2
N + , (4.4.32)
∂t ρ N ∂x ∂t ρ N ∂y ∂t2 ∂x2 ∂y2
where the Doppler shift has been neglected for 0 < t < τint . The time scale, τint , and
transverse length scale, L, are imposed by the quasi-impulsive source terms at the front
x = 0,

x = 0, 0 < t < τint : ∂/∂t = O(D/L), ∂/∂y = O(1/L). (4.4.33)

According to the last equation in (4.4.32), the pressure burst generated in the compressed
gas during this short lapse of time is then quasi-planar,

∂p/∂x ≈ (1/aN )∂p/∂t ≈ (D/aN )δp/L ≈ −1 (∂p/∂y);

see (4.4.11). A simplification occurs at low turnover Mach number, ve /D 1. In this

limit the disturbances of pressure and density, δp1 /pu and δρ1 /ρ u , are of order (ve /au )2 .
The upstream pressure fluctuation δp1f becomes negligible in the jump conditions for the
pressure (4.4.13) for a sufficiently small turnover velocity

ve /au : δpN /pN ≈ 2 δu1f − α̇t /D, (4.4.34)

since δp1f /pu is smaller than the first term in the right-hand side of the first equation
in (4.4.13), (δp1f /pu )/(ve /D) ≈ (ve /au )2 /(ve /D) ≈ (ve /au )/ 1. In this limit the
pressure pulse in the compressed gas is generated by an impulsive source, of lifetime τint
and transverse extension L, constituted mainly by a fluctuation of the longitudinal flow. The
longitudinal component of flow velocity associated with the quasi-planar pressure pulse
radiated in the compressed gas during this short lapse of time, δu(a) ≈ δp/(ρ N aN ), takes
the form

ve /au , 0 < t < τint : δu(a) |x=0 ≡ δu(a)

N ≈ 2(aN /D)(δu1f − α̇t ), (4.4.35)

and, according to the first and the second equations in (4.4.32), |δw(a) /δu(a) | = O().
According to the jump conditions (4.4.14), δuN ≈ α̇t , δwN ≈ Dαy + δw1f , and to the

flow splitting in (4.4.21)–(4.4.22), the flow velocity of the vorticity wave at the front,
(i) (a) (i) (a)
δu(i) |x=0 ≡ δuN = δuN − δuN , δw(i) |x=0 ≡ δwN = δwN − δwN ,
takes the form
(i)
δuN (y, t) ≈ α̇t − 2(aN /D)δu1f , (i)
δwN (y, t) ≈ Dαy + δw1f − δw(a)
N , (4.4.36)

where a correction of order 2 has been neglected in the first equation.

The equation for the disturbances of the front position during the short interaction time
0 < t < τint is obtained by the incompressible condition of the vorticity wave, ∂δu(i) /δx +
∂δw(i) /δy = 0, in the form

ve /au , 0 < t < τint : α̇t ≈ 2(aN /D)δu1f , (4.4.37)

as shown now. According to the expressions for the vorticity wave, δu(i) (x, y, t) =
(i) (i)
δuN (y, t − x/uN ), δw(i) (x, y, t) = δwN (y, t − x/uN ), the relation between the derivatives
with respect to space and time, ∂/∂x = −(1/uN )∂/∂t, with, according to (4.4.33),
∂/∂t = O(D/L), gives the order of magnitude for the derivative with respect to x of
the longitudinal component, ∂δu(i) /∂x = O(|δu(i) |/ 2 L), where the order of magnitude,
uN /D ≈ 2 in (4.4.12), has been used. Using ∂δw(i) /δy = O(δw(i) /L), it is then found
that the vorticity wave must be quasi-transverse, |δu(i) /δw(i) | = O( 2 ). The second term
in the right-hand side of the first equation in (4.4.36) leads to a contribution to δu(i)
that is of order δu1f . This term is too large, by a factor 1/, to satisfy the order of
magnitude imposed by the incompressible condition, |δu(i) | = O( 2 |δw(i) |), as can be
seen by anticipating that, to leading order, the second equation in (4.4.36) reduces to
(i)
δwN ≈ δw1f , where |δw1f /δu1f | = O(1). Therefore the second term, 2(aN /D)δu1f , in
(4.4.36) must be balanced by the first term, α̇t . This leads to (4.4.37) with, according to
(4.4.31), δu1f (y, t) = ue (−Dt, y), where ue (x, y) is the longitudinal velocity of the vortex
in the upstream gas. Then the amplitude of wrinkling is

aN 0 ue (x, y)
ve /au , 0 < t < τint : α(y, t) = 2 dx , (4.4.38)
D −Dt D
2
corresponding to a very small slope of the front |αy | of order ve aN /D = O( 2 ve /au ). The
symmetry property of the wrinkle α(y, t) = −α(−y, t) results from the limiting case of a
weak symmetrical vortex. However, because of the nonlinear character of the engulfment
mechanism, the wrinkling is no longer symmetric for a large turnover velocity, ve of the
order of au ; see Fig. 4.20. This effect is not taken into account in the present analysis.
The shock–vortex interaction during the crossover, 0 < t < L/D, for ve /au in the
limit (4.4.9), can be summarised as follows:

(i) A quasi-planar acoustic pulse is generated by the longitudinal component of turnover

velocity of the vortex,
(a)
δpN /pN ≈ 2δu1f /D, δuN ≈ 2δu1f , |δw(a) /δu(a) | = O().

(ii) The transmitted vortex is a vorticity wave quasi-parallel to the front generated by the
(i)
transverse component of turnover velocity of the vortex, δw(i) = δwN (y, t − x/uN ),
(i)
δwN ≈ δw1f , |δu(i) | = O( 2 |δw(i) |).

(iii) Wrinkles of very small amplitude are generated on the shock front during the
crossover,

α̇t ≈ 2δu1f , |αy | = O(ve /D).

Evolution after Crossover

After the crossover, t > τint , the shock wave propagates in the quiescent medium. The pres-
sure pulse in the compressed gas becomes multidimensional and takes a quasi-cylindrical
shape centred on the transmitted vortex core, as sketched in Fig. 4.22. However, in the
limiting case considered here, the sound intensity varies with the azimuthal angle and
decreases to negligible values near the shock front. This is because the sound is initially
radiated from the shock front at x = 0 by a pulse of longitudinal flow of transverse
extension L, while the transmitted vortex is an isobaric vorticity wave quasi-parallel to the
front. The conditions for the subsequent evolution of the wrinkled shock front are then the
same as in Section 4.4.2; Equation (4.4.28) holds for t > τint and describes the subsequent
Mach stem formation. As mentioned at the beginning of Section 4.4.3, the key point for
the validity of this simple description lies in the difference of time scales describing the
crossover and the transverse propagation of the wrinkles, aN τint /L being a small quantity
of order .

Model Equation for Shock–Turbulence Interaction (2013)

A composite solution for the wrinkled shock front produced by the shock–vortex interaction
is given by a nonlinear equation that interpolates the linear equation (4.4.37) for 0 < t <
τint and the nonlinear equation (4.4.28) describing the wrinkle dynamics for t > τint and
cusp formation in the long time limit,
2
|δu1f | ∂ 2α 2
2 ∂ α ∂ ∂α aN ∂δu1f
: − aN +D =2 , (4.4.39)
au ∂t 2 ∂y 2 ∂t ∂y D ∂t
where the forcing term is given by the vortex blob in (4.4.31) and varies on the short time
scale of crossover, τint = L/D.
The extension to a two-dimensional surface α(y, z, t) is straightforward. The correspond-
ing equation can also be used as a model for shock–turbulence interaction. In this case the
forcing term δu1f is the fluctuation of the longitudinal component of the velocity of the
turbulent flow at the front. In contrast to flames, due to the supersonic propagation of the
shock wave, the upstream turbulence is not modified by the wrinkling when approaching
the front. However, the turbulence in the compressed gas is strongly different from that
upstream.

Introducing the reduced variables

η ≡ y/L, ζ ≡ z/L, τ ≡ aN t/L, φ ≡ α/(L), (4.4.40)
Equation (4.4.39) multiplied by (L)−1 (L/aN )2 yields the nondimensional equation for the
surface of the shock front φ(η, ζ , τ ) in the form

∂ 2φ ∂|∇φ|2 ∂ψ δu1f
− φ + = , where ψ ≡ 2 (4.4.41)
∂τ 2 ∂τ ∂τ aN
and where, according to (4.4.11), the relation ≈ aN /D has been used, aN /au = O(1).
For ue /au the forcing term ∂ψ/∂τ is a small function of order ue /(au ) varying in
space with the transverse variables (η, ζ ) and in time with the short reduced time scale τ/.
Focusing attention on the geometry and the dynamics of folds on the wrinkled shock
front, representative of lines of Mach stems, instructive results are obtained from the numer-
ical study[1] of (4.4.41) in a two-dimensional periodic box. The forcing term ψ(η, ζ , τ/) is
extracted from a turbulence generator characterised by a k−5/3 Kolmogorov cascade. The
main physical outcome is related to the coalescence property mentioned earlier; in the long
time limit, the characteristic size of the cells delimited by the folds is not that of the length
scales of the turbulent flow. The average cell size increases continuously with time and
soon becomes much larger than the integral scale of the turbulent flow upstream from the
shock front; see Fig. 4.23. The saturation mechanism of the cell size and the role of the size
of the box (periodicity of η and ζ ) are not yet clearly identified.

4.5 Instabilities of Detonation Fronts

It has been known for a long time that gaseous detonations are unstable and exhibit trans-
verse structures. The experimental studies started as early as 1926 (spinning detonations).
A detailed history of the topic can be found in the literature.[2,3] The cellular structure
of detonation fronts was first observed in the 1960s by the markings left on soot-coated
foils on the walls,[2] showing more or less regular diamond-shaped patterns. These mark-
ings correspond to the trajectories of triple points (Mach stems) on the lead shock, cou-
pled to a pulsation of the internal structure of the detonation; see Fig. 4.30. More recent
experiments[4] have shown more complex patterns in H2 -(NO2 /N2 O4 ) mixtures, such as
double cellular structures; see Fig. 4.24. Modern optical methods,[5] including laser diag-
nostic techniques for imaging of chemical species within the detonation,[6] have improved
the observation of cellular structures. The research in this field is still active;[7] see also

[1] Denet B., et al., 2015, Combust. Sci. Technol., 187(1-2), 296–323.
[2] Shchelkin K., Troshin Y., 1965, Gasdynamics of combustion. Baltimore, Md.: Mono Book Corp.
[3] Strehlow R., 1979, Fundamentals of combustion. New York: Kreiger.
[4] Joubert F., et al., 2008, Combust. Flame, 152, 482–495.
[5] Presles H., et al., 1987, Combust. Flame, 70, 207–213.
[6] Mevel R., et al., 2014, J. Hydrogen Energy, 39, 6044–6060.
[7] Taylor B., et al., 2013, Proc. Comb. Inst., 34, 2009–16.

(a) (b)

Figure 4.23 Numerical simulation of (4.4.41) modelling the propagation of a shock wave in a
turbulent flow. The simulations are performed in a square box of reduced size 5, with a reduced
turbulent kinetic energy 0.078 and an integral scale of 0.15, corresponding to an amplitude of the
forcing term ψ = 0.396 and a time scale of fluctuations of 0.1. The small integral scale is selected
in order to highlight the ‘coalescence property’; the length scale of the structures on the wrinkled
front increases with time and soon becomes notably larger than the integral scale of the vorticity in
the flow, which is also the length scale of variation of ψ. The initial wrinkled front shown in (a) is
characterised by integral scale of the upstream turbulence. The scale of the wrinkles in (b) at reduced
time τ = 4 is typically 10 times larger. Courtesy of B. Denet, IRPHE Marseilles.

(a) (b)

Figure 4.24 (a) Visualisation of a cellular detonation obtained by an optical method recording the
deformation of an aluminised Mylar sheet hit by the detonation front. Reproduced from Presles
H., et al., Combustion and Flame, 70, 207–213, Copyright 1987 with permission from Elsevier. (b)
Markings of double cellular structure left on the wall by a detonation in a H2 -(NO2 /N2 O4 ) mixture.
Reproduced from Joubert F., et al., Combustion and Flame, 152, 482–495, Copyright 2008 with
permission from Elsevier.

the special issue Philos. Trans. R. Soc. A 370 (2012). The explanation of detonation insta-
bilities has been elusive for a long time, even though the diamond pattern of the cel-
lular structure of an unstable detonation front has been reproduced since the 1990s in
two-dimensional geometry by direct numerical simulations,[1,2,3] including more recently
double cellular structures.[4,5] The problem is too complicated to be studied analytically
in the general case. The purpose of this section is to provide understanding of the basic
mechanisms through analytical solutions of simplified models. The nonlinear phenomena
that control the spatio-temporal patterns of cellular detonation fronts have been analysed
in a systematic way by weakly nonlinear analyses that are valid near to the instability
threshold. These analyses are expected to point out the dominant physical mechanisms at
work in real detonations.
Two opposite limiting cases have been investigated using the Newtonian approxima-
tion: strongly overdriven detonations[6] and weakly overdriven detonations,[7] including
the CJ wave. Unfortunately, in both cases, the multidimensional stability limits concern a
small heat release, in contrast to ordinary detonations that are strongly unstable. However,
instructive insights are provided in the first case, the analysis of which is summarised in
Section 4.5.2 and developed in Section 12.2.4. The cellular instability of the second case,
near the instability threshold of CJ waves, is presented in Section 12.2.3. Such waves are
not physically relevant because the flow is transonic throughout the detonation structure,
including the lead inert shock, so that the ZND structure is possible only for reaction rates
that are quite artificial. However, the analysis is worth performing because it shows that the
oscillatory instability is similar to that at large overdrive.
The underlying physical mechanisms are better understood by considering first the one-
dimensional instability, presented now.

4.5.1 One-Dimensional Pulsations. Galloping Detonations

One-dimensional oscillations, called galloping detonations, were first observed in the 1960s
in direct numerical simulations.[8] They were visualised in the 1970s in the Schlieren
photograph of a blunt projectile traversing a combustible gaseous mixture at a velocity
close to the CJ detonation velocity.[9]

Instability Mechanism: Loop and Phase Shift

The origin of the one-dimensional instability can be roughly understood from the ZND
structure of a detonation wave, sketched in Fig. 1.9 (see also Fig. 4.29), when the variation

[1] Boris J., Oran E., 1987, Numerical simulation of reactive flow. New York: Elsevier.
[2] Bourlioux A., Majda A.J., 1992, Combust. Flame, 90, 211–229.
[3] Mevel R., et al., 2014, J. Hydrogen Energy, 39, 6044–6060.
[4] Guilly V., et al., 2006, C. R. Acad. Sci. Paris, 334(11), 679–685.
[5] Joubert F., et al., 2008, Combust. Flame, 152, 482–495.
[6] Clavin P., Denet B., 2002, Phys. Rev. Lett., 88(4), 044502–1–4.
[7] Clavin P., Williams F., 2009, J. Fluid Mech., 624, 125–150.
[8] Fickett W., Wood W., 1966, Phys. Fluids, 9, 903–916.
[9] Lehr H., 1972, Acta Astronaut., 17, 589–597.

of the induction delay with the temperature is taken into account. A disturbance that rein-
forces the lead shock increases the temperature at the Neumann state, shortening the induc-
tion delay and causing the heat release zone to move upstream relative to the lead shock.
The upstream motion of the heat release zone acts as a semi-transparent piston generating
an acoustic pulse that runs upstream and, in turn, further strengthens the lead shock, thereby
intensifying the initial disturbance. The time delays for the temperature perturbations to
be transmitted back and forth between the lead shock and the exothermic zone induce a
phase shift that is responsible for the oscillatory nature of the instability. The transmission
upstream is due to acoustics but the transmission downstream involves both an acoustic
wave and an entropy wave that propagates with the subsonic velocity of the flow relative to
the shock. The latter carries the strongest disturbances and imposes the longest delay in the
loop. It is thus the dominant mechanism in the one-dimensional oscillatory instability, and
also of the cellular instability described in Section 4.5.2, at least for strongly overdriven
detonations in the limit (4.4.9). This one-dimensional analysis[1] of galloping detonations
is presented now.

General Formulation (Mass-Weighted Coordinate)

According to the discussion below (4.2.45), molecular transport mechanisms are negligible
(λ = 0 and Di = 0) and Equations (12.2.1)–(12.2.2) govern the unsteady structure of ZND
detonations. Working in the reference frame of the unperturbed detonation located at x = 0
(with x < 0 in the fresh mixture) and denoting by α(t) the trajectory of the lead shock of
the perturbed detonation, it is convenient to introduce the reduced mass-weighted distance
from the shock, x, and the time, t, reduced by the unperturbed reaction time, tN ,
x
1 t
x≡ ρ(x , t)dx , t≡ , tN ≡ τr (T N ), (4.5.1)
ρu DtN α(t) tN

where the subscript N denotes the Neumann state. According to Section 4.6.3, the particu-
late derivative D/Dt ≡ ∂/∂t + u∂/∂x takes the reduced form

D D ∂ ∂ ρ(x, t)[u(x, t) − α̇t ]
= tN = + m(t) , m(t) ≡ , (4.5.2)
Dt Dt ∂t ∂x ρu D x=α(t)

where m(t) is here the reduced mass flux across the moving shock,

m(t) = D(t)/D, where D(t) ≡ D − α̇t . (4.5.3)

The velocity of the perturbed detonation relative to the quiescent fresh gas, D − α̇t , and m(t)
are the unknown functions of time characterising the dynamics of the planar detonation.

[1] Clavin P., He L., 1996, J. Fluid Mech., 306, 353–378.

Equations (12.2.1)–(12.2.2) then take the form

D ρu ∂ u D ρN ∂ u
= ⇔ = , (4.5.4)
Dt ρ ∂x D Dt ρ ∂x uN
. /
D u ∂ p
=− , p = (cp − cv )ρT,
Dt D ∂x ρu D 2
1 DT (γ − 1) 1 Dp qm Dψ
− = ẇ, = ẇ, (4.5.5)
T Dt γ p Dt cp T Dt
where we have used a shorthand notation for the species and the chemical kinetic scheme
by introducing the progress variable ψ ∈ [0, 1], as in (4.2.49), and also the reduced heat-

release rate, ẇ = tN Ẇ, ρqm ẇ = tN j Q(j) Ẇ (j) ; see (15.1.42). For the simplest model of
a one-step irreversible reaction governed by an Arrhenius law (1.2.2), one has ẇ(ψ, θ ) =
(tN /τcoll )(1 − ψ) exp(−E/kB T).
The boundary conditions at the Neumann state, x = 0, ψ = 0,
x = 0: ρ = ρN (t), p = pN (t), T = TN (t), (4.5.6)
are functions of m(t), since uN , ρN , pN and TN are given by the Rankine–Hugoniot relations
(4.2.14)–(4.2.16) in which the propagation Mach number Mu is a function of time Mu (t) =
(D − α̇t )/au = m(t)M u , δMu = −α̇t /au . The boundary condition for the velocity u in the
laboratory frame (relative to the unperturbed front) is given by mass conservation
x = 0: ρN (t)(u − α̇t ) = ρu (D − α̇t ). (4.5.7)
The intrinsic longitudinal dynamics develop on the induction time scale and are not very
sensitive to the rear boundary condition (in the burnt gas). In particular few differences are
observed between an overdriven detonation and a CJ detonation in the numerical analyses.
For an overdriven detonation with the piston at infinity (isolated detonation), the boundary
condition is
x → ∞: u = ub . (4.5.8)
Similar results are obtained using a radiation condition at the end of the reaction zone for
the acoustic waves propagating in the burnt gas (see the discussion in the original article[1] ),
x → ∞: (p − pb ) − ρ b ab (u − ub ) = 0. (4.5.9)
In the induction zone where the heat release is negligible, qm ẇ/cp TN 1, the flow, the
solution to (4.5.2)–(4.5.5) with the boundary conditions (4.5.6), may be decomposed into
an isobaric entropy wave and acoustic waves, at least for small fluctuations. However, in
general, this splitting is not possible in the exothermic zone. One option to overcome this
technical difficulty would be to consider the limiting case of a reaction sheet of negligible
thickness compared with that of the induction layer, namely the square-wave model, used
in Section 4.3.3 for quasi-steady state regimes. This model is not convenient here because it
leads to singular dynamics, as we shall see later. Other approximations must be considered
to obtain analytical results.

Overdriven Detonations in the Newtonian Approximation

Generally speaking, the quasi-isobaric approximation is valid for a sufficiently subsonic
flow when the evolution is slow on the time scale of acoustics; see Section 2.1.1. This is the
case in the shocked gas of a strong shock in the Newtonian approximation (4.4.9)–(4.4.12),
2
2 ≡ γ M N 1,

2 (γ − 1)
M u = O(1/ 2 ), (γ − 1) = O( 2 ), h≡ = O(1). (4.5.10)
2
In the same limit, the flow of shocked gas is also quasi-isobaric throughout the detonation
structure if the heat release is of same order of magnitude as the thermal enthalpy at the
Neumann state,

qN ≡ qm /cp T N = O(1), (T b − T N )/T N ≈ qN . (4.5.11)

This is the case for strongly overdriven detonation, as shown now.

Conservation of momentum in (4.2.47) and mass, ρu = ρN uN , lead to a linear relation
between velocity and pressure downstream of the lead shock expressed in terms of the
Mach number at the Neumann state, MN ,

2
p/pN − 1 = − 2 (u/uN − 1) , 2 ≡ γ MN , (4.5.12)

where the relation a2N /γ = (pN /ρN ) has been used. The temperature ratio T b /T N , the
velocity ratio ub /uN and the density ratio ρ b /ρ N being of order unity while the relative
pressure variation is of order 2 , the flow in the compressed gas satisfies the quasi-isobaric
approximation. However, the two conditions (4.5.10) and (4.5.11) are compatible only for
strongly overdriven detonations: according to (4.4.12), the ratios aN /au and T N /Tu are of
order unity, consequently, according to (4.2.44), the condition in (4.5.11) implies that MuCJ
is also of order unity, as is qm /cp Tu , so that M u MuCJ .
The problem of overdriven detonations in the Newtonian limit then reduces to solving
the equations for conservation of energy and species in (4.5.5), which form a closed set of
equations since the term (γ − 1)Dp/Dt is negligible,

∂T ∂T qm ∂ψ ∂ψ
+ m(t) = ẇ(ψ, T), + m(t) = ẇ(ψ, T). (4.5.13)
∂t ∂x cp ∂t ∂x

Using the boundary condition (4.5.6),

x = 0: ψ = 0, T = TN (t), (4.5.14)

the solution to (4.5.13)–(4.5.14) yields the spatial distributions of temperature and heat-
release rate in terms of TN (t). The boundary condition in the burnt gas yields a nonlinear
equation for the dynamics of the front. A useful integral equation is obtained by spatial
integration of the equation for energy conservation in (4.5.13) written in the quasi-isobaric

approximation, ρT = ρ N T N using the boundary conditions (4.5.7),

∂ ∂ T ∂ u
+ m(t) = , (4.5.15)
∂t ∂x T N ∂x uN
∞
ub ρu D ρu α̇t
− − 1− = qN ẇ dx, (4.5.16)
uN ρN (t)uN ρN (t) uN 0

where the reduced heat release qN is a parameter of order unity; see (4.5.11). The remaining
problem is to find the instantaneous distribution of rate of heat release ẇ as a function of
x and t. This is more easily done with the additional simplification of a large activation
energy. In ordinary gaseous mixtures, the reaction rate is strongly sensitive to temperature,
and, more specifically, the induction delay varies with order unity for small fluctuations
of the temperature TN at the Neumann state x = 0. Attention is then limited to small
fluctuations of TN (t), that is, to small fluctuations of the shock velocity δD/D = O(1/βN )
where βN is a large parameter, βN 1, that characterises the sensitivity of the distribution
of heat release rate to the Neumann temperature,

N (t) ≡ βN (TN − T N )/T N = O(1), (uN − uN )/uN = O(1/βN ), (4.5.17)

and the solution is sought in the limit βN → ∞. We assume that the inner structure of
the detonation varies by order unity when the variation of Neumann temperature is small,
of order 1/βN , δTN /TN = O(1/βN ). In the limits (4.5.10)–(4.5.11) and for βN 1, the
leading order of the left-hand side of (4.5.16) is (ub /uN − 1) − α̇t /uN , and a simplified form
of the nonlinear equation for the front dynamics is obtained:
∞
ub α̇t
−1 − ≈ qN ẇ dx. (4.5.18)
uN uN 0

Since the variation of D is small, δD/D = O(1/βN ), the linear approximation of the
Rankine–Hugoniot relations (4.2.14)–(4.2.16) can be used. To leading order, the variations
of the profiles [T(x, t) − TN (t)] /TN (t) and ψ(x, t) are of order unity, whereas the variations
of the mass flux at the Neumann state remain small δm(t) ≈ O(1/βN ), and the reduced
mass flux m(t) can be replaced by unity in (4.5.13),
∂T ∂T qm ∂ψ ∂ψ
+ = ẇ(ψ, T), + = ẇ(ψ, T), (4.5.19)
∂t ∂x cp ∂t ∂x
the dominant effect coming from the unsteady boundary condition (4.5.14) for the small
temperature fluctuations at the Neumann state, δTN (t)/TN = O(1/βN ).

Limitation of the Quasi-isobaric Approximation

The CJ regime is excluded by considering qN to be of order unity in the limit (4.5.10).
However, the discrepancies are not very important in real CJ detonations. For a typical
detonation propagating in an ordinary gaseous mixture such as that considered just below
(4.2.57) with qm /cp Tu = 8, γ = 1.3, and MuCJ = 6.22, MN2 CJ = 0.176, T NCJ /Tu = 6.66,
the quantity qN ≡ qm /(cp T NCJ ) = 1.2 is of order unity, as is assumed in (4.5.11). The

increase of temperature in the shocked gas is of order unity, (T bCJ /T NCJ − 1) = 0.93.
The difference with the quasi-isobaric approximation, 1.2, is not large, 25% lower due to
compressible effects. The temperature change due to compressible effects represents less
than 25% of that due to gas expansion in the quasi-isobaric approximation. According
to (4.2.53) and (4.5.12), the increase of the flow velocity and the decrease of pressure
across the exothermic reaction zone are of order unity, abCJ /uNCJ = 2.88, pbCJ /pNCJ ≈
0.5, as is well known for CJ waves. A pressure variation of order unity in the reaction
zone does not mean that the quasi-isobaric approximation is irrelevant for the dynamics
of real detonations near the CJ regime. The large variations of the induction length, due
to thermal sensitivity, is the essential mechanism modifying the reaction rate at a fixed
distance from the lead shock. The induction zone being very subsonic, M NCJ = 0.135 in
the case considered above, the variation of the induction length is accurately described by
a quasi-isobaric approximation. The effects of the compressible phenomena in (4.5.5) may
be decomposed into two parts, unsteady and steady. The quasi-steady effect of the pressure
on the temperature profile is not very important from a quantitative point of view and can
easily be taken into account in a modified version of the quasi-isobaric theory presented
at the end of this section. Concerning the purely unsteady effects of pressure on ordinary
galloping detonations in the self-propagating CJ regime, the acoustic waves introduce a
small additional time delay and a small additional phase shift in the loop mentioned at the
beginning of this section. The phase shift is small because of the subsonic character of the
induction zone, and a perturbation analysis[1] shows that it produces a stabilising effect.
However, the pressure fluctuations become important in the dynamics of strongly unsta-
ble detonations and also in the transonic CJ regime for small heat release studied in Section
12.2.2.
To summarise, in ordinary detonations the quasi-isobaric propagation of the entropy
wave across the induction zone controls the position of the exothermic layer relative to the
shock. It is the basic mechanism that drives the galloping instability of real detonations.
The accuracy of the results in the quasi-isobaric approximation is increasingly better for
smaller values of (γ − 1).

Integral Equation
Using N defined in (4.5.17), let T = T (N , x) and ψ = Y(N , x) denote the steady state
solution to (4.5.19), dT /dx = (qm /cp )ẇ(Y, T ), satisfying the boundary condition x = 0:
ψ = 0, T = TN . The solution to (4.5.19) satisfying (4.5.14) is
T(x, t) = T (N (t − x), x) , ψ(x, t) = Y (N (t − x), x) . (4.5.20)
In this retarded solution, the quantity (t − x) is the reduced time at which a fluid particle,
located at position x at time t, had crossed the lead shock. In other words, x is the reduced
time lag taken by a fluid particle to go from the lead shock to the position x, since, according
to (4.5.20), the reduced propagation velocity is unity. The general solution to (4.5.13) for

[1] Clavin P., He L., 1996, Phys. Rev. E, 53(5), 4778–4784.

(a) (b)
2 3

2
1.5

1
1
0

0.5
−1

0 −2
0 2 4 0 2 4

Figure 4.25 Profiles of 0 (x) and N (x) for a modified Arrhenius law (4.5.25) with qN = 1 and
βT = 1, 2, 4. From Daou R., Clavin P., 2003, J. Fluid Mech., 482, 181–206, with permission.

m(t) = 1 takes the same form as (4.5.20) when the delay x in t − x in the right-hand side is
replaced by τ (x, t), the solution to Dτ/Dt = 1 satisfying the condition x = 0: τ = 0.
Denoting (N , x) the distribution of the reduced reaction rate of the steady state
∞
solution to (4.5.19) for T|x=0 = TN , 0 (N , x)dx = 1, (Tb − TN ) = qm /cp , the
instantaneous distribution of reaction rate ẇ(T, Y) for a fluctuating Neumann temperature
TN (t), that is for (4.5.20), is the retarded function (N (t − x), x). In the limit (4.5.10)–
(4.5.11) Equation (4.5.18) then reads
∞
α̇t
− = qN (N (t − x), x) − (N , x) dx.
uN 0

Neglecting (γ − 1) in front of 2γ Mu2 , the linearised Rankine–Hugoniot relation (4.2.16)

yields
δTN /T u ≈ [4γ (γ − 1)/(γ + 1)2 ]M u δMu , δTN /T N ≈ −(γ − 1)α̇t /uN .
Expressing α̇t in terms of N then leads to a nonlinear integral equation for N (t),
∞
1 + bN (t) = (N (t − x), x)dx, b−1 ≡ βN (γ − 1)qN , (4.5.21)
0
where the parameter b is of order unity for a high temperature sensitivity,
2
(γ − 1)βN = O(1), that is, βN = O(M u ). (4.5.22)
In these conditions, the instability threshold for planar disturbances is described by the
linear integral equation obtained from (4.5.21)
∞
δN (t) = b−1 N (x)δN (t − x)dx, N (x) ≡ ∂/∂N |N =0 , (4.5.23)
0
∞ ∞
where 0 N (x)dx
= 0 since 0 (N , x)dx = 1; see Fig. 4.25. Looking for the solution
in the form δN (t) ∝ eσ̂ t with σ̂ = ŝ+iω̂, the reduced complex growth rate σ̂ is the solution

to the integral equation

∞
b= N (x)e−σ̂ x dx. (4.5.24)
0

For a given regular function N (x), the detonation becomes unstable when b−1 is
increased: above a critical value b−1c the linear growth rate becomes positive (b
−1 > b−1 :
c
ŝ > 0) and the imaginary part of the growth rate is nonzero, describing an oscillatory
instability. This is a Poincaé[1] –Andronov bifurcation (ŝ = 0, ω̂ = 0), also called a Hopf
bifurcation.
For a given value of the parameter b−1 , the bifurcation occurs when the function N (x)
becomes stiffer. Two simple examples describing two extreme cases are instructive:

• Consider first a reaction rate that increases rapidly as the Neumann temperature is
increased, but which is independent of changes of temperature downstream from the lead
shock. The progress variable then increases as an inverse exponential with the distance
−1 −x/lN
from the shock, 1 − ψ = e−x/lN , (N , x) = lN e , and the length scale lN can
be assumed to vary exponentially with the Neumann temperature, lN = e−N so that
N (x) = (1 − x)e−x . Equation (4.5.24) reduces to σ̂ 2 + (2 − b−1 )σ̂ + 1 = 0 with an
instability threshold (ŝ = 0, ω̂ = 1) at b−1
c = 2.
• Consider now the opposite situation, described by the square-wave model, for which
the reaction time is so sensitive to temperature that the heat release is located in a thin
reaction zone located at the reduced distance lN = e−N from the shock, (N , x) =
δ(x − lN ), N (x) = δ (x − 1), where δ(x) denotes the Dirac distribution and δ (x) is its
derivative. Equation (4.5.23) takes the form of an advanced-time difference-differential
equation, dN (t − 1)/dt = bN (t), which is known to be quite singular, as shown by
(4.5.24), σ̂ e−σ̂ = b. This transcendental equation has an infinite set of discrete unstable
modes with unbounded amplification rates increasing with frequency (ŝ → ∞, ω̂ → ∞),
whatever be the value of the parameter b. This square-wave pathology is also observed if
compressible effects (acoustic waves) are taken into account.[2]

Modified Arrhenius Model

A chemical kinetic model, representative of gaseous detonations and free from the spurious
square-wave pathology, must represent both an induction delay that is very sensitive to
changes of the Neumann temperature and a heat-release rate that remains finite. The second
condition is not fulfilled by the one-step Arrhenius model in the limit of a infinitely large
activation energy, which leads to the square-wave model. A convenient and accurate model
for gaseous detonations is an extension of the ZFK model, involving two different activation
energies, EN and ET , describing two thermal sensitivities, one for the induction delay and

[1] Poincaré H., 1908, Revue d’électricité, 387.

[2] Clavin P., He L., 1996, J. Fluid Mech., 306, 353–378.

Modified 1-step model

Arrhenius law

Reduced growth rate,

Reduced frequency,

Figure 4.26 Numerical results for the frequency spectrum of a galloping detonation for the modified
one-step model (solid lines) with b−1 = 2.03 and βT = 8.44. Time scales are reduced by the
unperturbed induction delay. For comparison, the numerical results from the full equations for an
Arrhenius law (βN = βT = 8.44) are plotted with dotted lines. Adapted from Clavin P., He L., 1996,
J. Fluid Mech., 306, 353–378, with permission.

the other for the heat release.[2] The corresponding reaction rate
−1
Ẇ(ψ, T) = Bτcoll (1 − ψ)e−EN /kB TN eβT (T−TN )/TN (4.5.25)

is investigated in the limit βN ≡ EN /kB TN → ∞, βT = O(1). In the distinguished limit

(4.5.10)–(4.5.11) and (4.5.22), this modified one-step model leads to the nonlinear integral
equation (4.5.21) with (N , x) = (l˜ N , x), where

˜ N , x) = l−1 0 (x/lN ),
(l lN ≡ e−N , N (x) = d(x0 )/dx, (4.5.26)
N

and where 0 (x) is the profile of the reaction rate for N = 0, that is, TN = T N .
For particular examples of functions 0 (x), Equation (4.5.24) becomes polynomial and
σ̂ can be obtained analytically; see (12.2.51)–(12.2.56). The distribution 0 (x) can also
be computed from the steady state solution to (4.5.19) for the boundary condition x = 0:
T = T N and in which

ẇ(ψ, T) = (1 − ψ)eβT (T−T N )/T N . (4.5.27)

In this limit, the one-dimensional nonlinear dynamics are thus fully controlled by two
parameters of order unity, b−1 and βT . The first parameter, defined in (4.5.21), characterises
both the degree of overdrive through qN and the thermal sensitivity of the induction length
through βN . The second characterises the increase of heat-release rate with the increase of
temperature inside the exothermic layer. This model has been investigated numerically.[2]
A typical result for the spectrum of the linear equation (4.5.23) with the kinetic model
(4.5.25)–(4.5.27) is shown in Fig. 4.26 for βT = 8.44 and b−1 = 2.03. The number of
unstable modes, the linear growth rate and the frequency of oscillation increase without
upper bound when βT is increased, leading to the spurious square-wave pathology in the
limit βT → ∞. For the sake of comparison, the spectrum of the linear version of the full
−1
equations (4.5.4)–(4.5.5) with the Arrhenius law Ẇ(ψ, T) = Bτcoll (1−ψ)e−E/kB T is plotted

III
Dynamical
quenching

I II
Stable Unstable

Figure 4.27 Stability thresholds of galloping detonation in the b−1 , βT plane. Squares and solid
line: stability limits obtained from (4.5.23) with the model (4.5.26) where 0 (x) is computed with
(4.5.27). For comparison, the numerical stability limits are given by the circles plus dashed line for
an Arrhenius law (βN = βT ) and for different values of f and γ , the heat release being given by
the expression of b−1 in (4.5.21). Diamonds plus dotted line: limit of dynamical quenching for an
Arrhenius law. Adapted from Clavin P., He L., 1996, J. Fluid Mech., 306, 353–378, with permission.

in the same figure for βN = βT = 8.44, γ = 1.2 and a overdrive factor corresponding to
the value of qN yielding b−1 = 2.03. Comparison shows a qualitative agreement but also
quantitative differences. The stability limit in the plane b−1 –βT is plotted in Fig. 4.27. Near
the instability threshold, the time scale of the dynamics of a galloping detonation is of the
same order as the transit time of a fluid particle across the inner structure of the detonation.

Nonlinear Dynamics. Dynamic Quenching

Beyond the stability limit, the numerical results for the nonlinear behaviour, obtained from
(4.5.21), show period doubling and a transition to chaos as b−1 is increased. Intermittent
bursts of large-amplitude high-frequency oscillations, separated by long periods of almost
constant velocity propagation with a velocity and a Neumann temperature both well below
their values in the steady state solution, are also observed as b−1 is further increased; see
Fig. 4.28. Such large fluctuations of the Neumann temperature lead to dynamic quench-
ing of planar detonations. This phenomenon has also been observed in direct numerical
simulations of planar detonations, sustained by a one-step Arrhenius reaction, if the acti-
vation energy has sufficiently high value.[1] In real detonations the dynamical quenching is
reinforced by the crossover temperature T ∗ below which the exothermic reaction is quasi-
quenched; see Section 1.2.2 and Chapter 5 for more detail. This is easily understood: if the
dynamically induced fluctuations are sufficiently strong to bring the Neumann temperature
TN (t) below T ∗ , the coupling of the nonlinear galloping instability to the nonlinear chemical
kinetics leads to quenching of the planar detonation. The multidimensional case is not yet
known.

[1] He L., Lee J., 1995, Phys. Fluids, 7, 1151–1158.

Figure 4.28 Galloping detonation: evolution of N (t) obtained from (4.5.21) with βT = 5 for
increasing values of b−1 showing period doubling, chaotic bursts and quasi-quenching. From Clavin
P., He L., 1996, J. Fluid Mech., 306, 353–378, with permission.

Concluding Remarks. Further Studies and Extended Models

A comparison with direct numerical simulations shows that the dynamics of the lead shock,
obtained by the model equation (4.5.21) represents the essential mechanism of the gallop-
ing detonation, at least for the unstable domain not too far from the instability threshold.
Compressible effects have been taken into account in a perturbation analysis.[2] The
result takes the same form as (4.5.21) but with a correction of order to the coefficient b
and to the time lag inside the integral equation. Moreover, acoustic waves are found to be
stabilising, so that the instability does not arise from thermo-acoustic instability, contrary
to what was often thought before.
Extended analyses have been performed[3] using a three-step chemistry model that
describes chain-branching processes discussed in Section 5.2.1. They are seen to success-
fully reproduce many of the characteristics observed in computations employing detailed
chemistry for hydrogen–air mixtures.[4]
More accurate quantitative results can be obtained with the model equation (4.5.21) by
using a distribution 0 (N , x), computed from a numerical simulation of the steady
state solution for a complex kinetics scheme (e.g. the one used in Fig. 2.2 for the

[2] Clavin P., He L., 1996, Phys. Rev. E, 53(5), 4778–4784.

[3] Sanchez A., et al., 2001, Phys. Fluids, 13(3), 776–792.
[4] Yungster S., Radhakrishnan K., 2004, Combust. Theor. Model., 8, 745–770.

(a) (b)

m/s, K
m/s, K
m/s, K

Figure 4.29 (a) Spatial distribution of the reaction rate in overdriven detonations propagating in a
stoichiometric hydrogen–oxygen mixture for different Neumann temperatures (different degrees of
overdrive). The reaction rate and the distance are scaled by those of the detonation propagating at
3100 m/s. The calculation is performed with a detailed chemical kinetic scheme of hydrogen–oxygen
combustion; see Section 5.3. (b) Same data rescaled, showing that the extended ZFK model is a fairly
good approximation. From Clavin P., He L., 1996, J. Fluid Mech., 306, 353–378, with permission.

structure of a methane flame). For better accuracy, compressible effects may easily be
included in the computation of (N , x). For example, the inner structure of hydrogen–
oxygen detonations, computed with a detailed kinetic scheme for different propaga-
tion velocities (different overdrive factors), is shown in Fig. 4.29a. The scaling law
˜ N , x) = l−1 0 (x/lN ) in (4.5.26) is shown to be a reasonable approximation;[1]
(l N
see Fig. 4.29b.
The instability threshold of galloping detonations in the CJ regime has been
investigated[2] for a small heat release. This extreme condition corresponds to the opposite
limit of the quasi-isobaric approximation since the dominant mechanism is now acoustics.
However, the nature of the galloping instability is not very different from the one in the
limit (4.5.10)–(4.5.11). It still results from a phase shift in the loop involving the lead
shock and the reaction zone, but the time delay is now due to the upstream running
acoustic wave instead of the downstream running entropy wave. The sensitivity of the
heat-release rate to the Neumann temperature is still the mechanism responsible for the
one-dimensional instability, and the detonation becomes stable against planar disturbances
when the thermal sensitivity is turned off. The corresponding analysis[2] is presented in
Section 12.2.2.

4.5.2 Cellular Detonations. Diamond Pattern

Gaseous detonations are known to exhibit transverse structures that are larger than the
detonation thickness. The underlying mechanism of the ‘diamond’ or ‘fish scale’ pattern,

[1] Clavin P., He L., 1996, J. Fluid Mech., 306, 353–378.

[2] Clavin P., Williams F., 2002, Combust. Theor. Model., 6, 127–129.

Figure 4.30 Time evolution of a cellular detonation showing how the diamond pattern is produced
by transverse propagation of triple points coupled to the longitudinal pulsation of the internal and
quasi-planar structure.

recorded on soot-coated foils (see Fig. 4.24), is sketched in Fig. 4.30. The cellular pattern
results from the coupling of the one-dimensional pulsation of the internal structure, studied
in Section 4.5.1, and the formation of Mach stems on the wrinkled shock front discussed
in Section 4.4.2. Using a pulsation time of the order of the induction delay tN and the
transverse velocity of the Mach stem aN , the order of magnitude of the cell size in the
transverse and longitudinal directions would be (aN /uN )lN and (D/uN )lN , respectively.
These two lengths are larger than the thickness of the induction zone lN ≈ uN tN by the
factors aN /uN and D/uN , respectively, in rough agreement with observations. The real
problem is more complicated since planar waves can be strongly unstable to disturbances
with transverse components, even when the detonations are stable to purely longitudinal
disturbances. In this sense the cellular instability is stronger than the galloping detonation.
However, the underlying oscillation, even when it is damped in planar geometry, is essential
to the cellular structure.
As for most of the nonlinear patterns, purely analytical studies cannot be performed
in the general case. An analytical description of cellular detonations has been obtained for
strongly overdriven regimes by a weakly nonlinear analysis at the threshold of instability.[3,
4] This analysis is summarised below; the details are given in Section 12.2.4.

[3] Clavin P., 2002, In H. Berestycki, Y. Pomeau, eds., Nonlinear PDE’s in condensed matter and reactive flows, 49–97. Kluwer
Academic Publishers.
[4] Clavin P., Denet B., 2002, Phys. Rev. Lett., 88(4), 044502–1–4.

Instability Threshold of Strongly Overdriven Detonations

The preliminary step is to carry out a multidimensional linear analysis to find the instability
threshold. The pioneering analyses of linear stability[1,2] were performed as initial-value
problems, using a Laplace transform and numerical analysis. The existing weakly nonlinear
analysis[3,4] is based on linear analyses[5,6,7] performed for overdriven gaseous detonations
in the limit (4.5.10)–(4.5.11), using a normal mode decomposition (4.4.1). The dispersion
relation, namely the relation linking the complex linear growth rate σ and the wavenumber
k, is obtained for a polytropic gas by imposing the Rankine–Hugoniot conditions (12.1.25)–
(12.1.26) at x = 0 and a boundedness condition at infinity, x → ∞ on the solutions to the
reactive Euler equations (12.2.1)–(12.2.2). In the general case there are different branches
of solutions to the dispersion relation σ (k), and the overall picture is rather complicated.[5]
In the limit (4.5.10) the instability threshold occurs at small values of qN , for qN /(γ − 1)
of order unity,[7] that is,
2
2 ≡ γ M N 1, (γ − 1) = O( 2 ), qN ≡ qm /cp T N = O( 2 ). (4.5.28)
Real detonations are always unstable because they do not correspond to such a small
heat release. A weakly nonlinear analysis near to the multidimensional stability limit is
physically meaningful if it retains the dominant nonlinear effects of real detonations, which
is usually the case if there is no secondary bifurcation.
The analysis simplifies near the stability limit but is still complicated. The presentation
is limited here to the discussion of the results, and the details are given in Section 12.2.4. In
a way similar to the one-dimensional analysis in Section 4.5.1 the chemical reaction occurs
only through two distributions, the steady state distribution (x) ≡ (N x) and N (x)
that describes its modification when the detonation speed is modified; see Fig. 4.29a. We
assume that they are still meaningful in the limit of small heat release in (4.5.28).
According to the discussion at the beginning of this section, we introduce the following
scaling of the reduced wavenumber κ and the reduced linear growth rate, σ = s ± iω, s and
ω > 0 being real numbers, scaled using the reaction time tN and the flow velocity uN at the
Neumann state of the unperturbed solution
σ ≡ σ̂ tN , κ ≡ |k̂|uN tN /, σ (κ) = s(κ) ± iω(κ), (4.5.29)
where in this section the hat denotes dimensional quantities. The function σ (κ) is obtained
by a perturbation analysis in the form of an expansion in powers of 2 1 in the limit
(4.5.10) and (4.5.28),
σ = σ0 (κ) + 2 σ2 (κ) + O( 4 ). (4.5.30)

[1] Erpenbeck J., 1962, Phys. Fluids, 5, 604–614.

[2] Erpenbeck J., 1966, Phys. Fluids, 9, 1293–1306.
[3] Clavin P., 2002, In H. Berestycki, Y. Pomeau, eds., Nonlinear PDE’s in condensed matter and reactive flows, 49–97. Kluwer
Academic Publishers.
[4] Clavin P., Denet B., 2002, Phys. Rev. Lett., 88(4), 044502–1–4.
[5] Clavin P., et al., 1997, Phys. Fluids, 9(12), 3764–3785.
[6] Clavin P., He L., 2001, C. R. Acad. Sci. Paris, 329(IIb), 463–471.
[7] Daou R., Clavin P., 2003, J. Fluid Mech., 482, 181–206.

The leading order, σ0 (κ), corresponds to the oscillatory modes of the lead shock (see
(4.4.15)), s0 = 0, ω0 = κ. The linear growth rate appears at the following order Re(σ ) =
2 s2 (κ), and, according to (12.2.87)–(12.2.88) takes the form[7]
Re(σ )/κ, MN
2 = −Im S (i) (κ) − √ S (a) (κ), (4.5.31)
qN qN
where by definition
(i)
S (i) (κ) ≡ βN (γ − 1)sβN (κ) + sq(i) (κ), (4.5.32)
∞ ∞
sβ(i)N (κ) ≡ N (x)e−iκx dx, sq(i) (κ) ≡ (1 + iκx)(x)e−iκx dx, (4.5.33)
0 0

(γ − 1)

S (κ) ≡ 2 Im
(a)
+ S (κ) − 1 .
(i) (4.5.34)
2qN

The result σ0 = iκ has been used in (4.5.31)–(4.5.33). Equation (4.5.31) describes an

√
isobaric mechanism, Im S (i) (κ) , and a compressible effect proportional to M N / qN ,
involving S (a) . In addition to the functions of order unity (x) and N (x) controlled by the
chemical kinetics, the reduced growth rate Re(σ )/qN depends on three positive parameters
√
of order unity, βN (γ − 1) measuring the thermal sensitivity, (γ − 1)/qN and M N / qN . A
small correction to the oscillatory frequency is also obtained at this order ω/κ = 1+ O( 2 ).
Using (4.4.10), it is useful to rewrite (4.5.34) in the form

M N (a) 2 (γ − 1) 2qN (i)
√ S = 1+ Im 1 + S (κ) − 1 . (4.5.35)
qN (γ − 1)M
2 qN (γ − 1)
u

The function S (i) (κ) describes quasi-isobaric mechanisms of instability resulting from the
coupling of the vorticity–entropy wave to the rate of heat release. There are two different
contributions:
(i)
• The function sβN (κ) comes from the sensitivity of the induction delay to the Neumann
temperature, responsible for the one-dimensional instability; see (4.5.23)–(4.5.24). In the
conditions of (4.5.28), the parameter b−1 = βN (γ −1)qN is small, of order 2 , well below
the critical value b−1 −1
c in (4.5.24), since bc is typically of order unity (for a thickness of
the exothermic reaction layer of the same order as the induction length). Therefore the
detonation is stable against one-dimensional longitudinal disturbances.
(i)
• The function sq (κ) describes the interaction of the heat release rate with the deflection
of the flow velocity across the wrinkled wave. This leads to a strong instability against
transverse disturbances that does not involve the sensitivity to temperature. This instabil-
ity mechanism has some analogy with the Darrieus–Landau hydrodynamic instability in
flames.

The quasi-isobaric instability is strong in the sense that the overall growth rate
remains positive and reaches its maximum in the limit of small wavelengths,

7
6
5
4
3
2
1
0
−1
0 2 4 6 8 10

Figure 4.31 −κIm[S (i) (κ)] versus κ for an Arrhenius law with βN qN = 0.1 and βN (γ − 1) = 0, 1,
5. From Daou R., Clavin P., 2003, J. Fluid Mech., 482, 181–206, with permission.

limκ→∞ −κIm[S (i) (κ)] = cst. > 0; see Fig. 4.31. This is also true in the absence of
(i)
thermal sensitivity, βN (γ − 1) = 0, limκ→∞ −Im[sq (κ)] > 0.
The positive function S (a) (κ) > 0 describes a stabilisation mechanism that results from
the coupling of the acoustic waves to the heat release rate. The origin of the stabilisation
is the modification of the longitudinal flow velocity at the exit of the reaction zone due
to radiating acoustic waves in the burnt gas, leading to a weakening of the piston effect
mentioned at the beginning of Section 4.5.1. This is called a ‘negative velocity coupling’
in the thermo-acoustic literature.
For small qN /(γ − 1) the damping mechanism is stronger
than the isobaric instability
√
at all wavelengths, (M N / qN )S (a) (κ) > |Im S (i) (κ) | ∀κ, as shown from an expansion
2
of (4.5.35) for small qN /(γ − 1) 1 by noticing that 1 + 2/[(γ − 1)M u ] > 1. The planar
front is unconditionally stable in this condition. In the opposite limit, for large qN /(γ −
1), the isobaric instability mechanism dominates the acoustic damping in an intermediate
range of wavelengths corresponding to κ = O(1). Compressible effects stabilise the small
wavelength disturbances,

lim S (a) (κ) → 2, lim κS (a) (κ) = 2κ,
κ→∞ κ→∞

as shown from (4.5.34) when qN /(γ − 1) 1 using limκ→∞ S (i) (κ) = 0. Thus, as
qN /(γ − 1) is increased, a Poincaré–Andronov (Hopf) bifurcation occurs at a finite wave-
length, larger than the detonation thickness by a factor 1/; see (4.5.29).
An example of the dispersion relation near the onset of instability, obtained from
(4.5.31) for an Arrhenius law, is shown in Fig. 4.32. The expression for the linear growth
rate in terms of the wavenumber, σ (κ), is a combination of the two roots of a quadratic
equation; see (12.2.87). Only the branch of the solution satisfying the boundedness
condition of the acoustic waves in the limit x→ ∞ is retained. The overall result consists in
two pieces of two different branches of the solutions. This is responsible for the small jump

(a) (b)
0.005 2.5

2.0
0.0025 Unstable branch
Nonphysical
1.5
0
1.0

−0.0025 Stable
branch
0.5
Unstable branch
−0.005 0.0
0 0.5 1.0 1.5 2.0 2.5 0 0.5 1.0 1.5 2.0 2.5

Figure 4.32 (a) Growth rates from (4.5.31) with (γ − 1) = 0.1, βN = 10 and Mu2 = 50 for an
unstable case, qN = 0.04 (solid line) and a stable case, qN = 0.03 (dashed line). The nonphysical
parts of the solutions are shown by thin lines. (b) Frequency as a function of κ for the unstable branch
of the solution. The thin lines correspond to the nonphysical part. From Daou R., Clavin P., 2003,
J. Fluid Mech., 482, 181–206, with permission.

0.002

Numeric
0

−0.002 Analytic

−0.004
0 0.2 0.4 0.6 0.8 1.0 1.2

Figure 4.33 Determination of the threshold of linear instability. The dispersion relations at the
threshold, σ (κ), given by (4.5.31) (dashed line) and by numerical analysis of the linear equation (solid
line) for βN = 0, γ = 1.05, (γ −1)Mu2 = 1. The numerical result corresponds to 2qN /(γ −1) = 10.4
and the analytical result to 2qN /(γ − 1) = 9.6 at threshold. Adapted from Daou R., Clavin P., 2003,
J. Fluid Mech., 482, 181–206, with permission.

in frequency at κ ≈ 0.75 in Fig. 4.32b. A simplification occurs for a purely ‘hydrodynamic

instability’, βN (γ − 1) ≈ 0, for which Equation (4.5.31) corresponds to a single branch of
solution. Such an example of bifurcation is shown in Fig. 4.33. The comparison with the
branch of solution computed numerically shows a good accuracy of the analytical result.
To summarise, near the instability threshold, the multidimensional instability of over-
driven gaseous detonations is mainly due to a quasi-isobaric mechanism that is hydrody-
namic in nature. In contrast to galloping detonations, the instability develops even in the
absence of sensitivity of the induction length to temperature variations because the induced
transverse velocity modifies the longitudinal distribution of heat release rate. Compressible

phenomena have a stabilising effect that is essential to stabilise the disturbances at small
wavelengths. The competition of these two mechanisms, quasi-isobaric and compressible,
results in a Poincaré–Andronov (Hopf) bifurcation at a wavelength larger than the detona-
tion thickness with an oscillatory frequency of the order of the inverse of the transit time
across the inner detonation structure.

Nonlinear Model Equation for Cellular Detonations (2002)

A weakly nonlinear analysis,[1,2] performed for values of parameters in (4.5.28) that are at
the instability threshold, leads to an equation for the evolution of the detonation front that
combines the nonlinear term in (4.4.28), responsible for singularity formation on an inert
shock front (representative of Mach stems), studied in Section 4.4.2, and the linear integral-
differential equation corresponding to the dispersion relation (4.5.31)–(4.5.34). Using the
same reduced coordinates as in (4.5.1) with nondimensional transverse coordinates reduced
by uN tN /, as in (4.5.29), and a nondimensional amplitude of wrinkling reduced by uN tN ,
α = α̂/(uN tN ), the equation takes the form in the limit (4.5.28)
∂ 2α ∂|∇α|2 √ ∂
2
− c ∇
2 2
α + = −2M N qN L(a) (α) + qN L(i) (α), (4.5.36)
∂t ∂t ∂t
where c2 = 1 + 3(γ − 1)/2, L(a) (α) is a linear term corresponding to the damping term
(4.5.34) due to compressible effects, and L(i) (α) is a linear term corresponding to the quasi-
isobaric instability, represented by S (i) (κ) in (4.5.32),
(i)
L(i) (α) = βN (γ − 1)lβN (α) + lq(i) (α), (4.5.37)
2 ∞ ∞
∂
N (x)α(t − x)dx, lq(i) (α) = ∇ 2
(i)
lβN (α) = 2 !(x)α(t − x)/dx,
∂t 0 0

where !(x) ≡ (x) + d(x)/dx, and where the transverse variable of α(t, y) is not written
(i)
explicitly. The second derivative with respect to time in lβN (α) can be well replaced by
∇ 2 since to leading order, ∂ 2 α/∂t2 = ∇ 2 α. The form (4.5.37) is obtained directly from
(12.2.86) when ∂α(t − x)/∂t is replaced by −∂α(t − x)/∂x and using integration by parts.
The linear operator L(a) (.) is defined in Fourier space as κS (a) (κ)/2 in (4.5.34), so that
the small wavelength limit yields L(a) (.) → κ/2 in Fourier space. The left-hand side of
(4.5.36) corresponds to the nonlinear equation for Mach stem formation (4.4.28). Without
the nonlinear term, third term in the left-hand side, Equation (4.5.36) yields the dispersion
relation (4.5.31)–(4.5.34). In one-dimensional geometry this linear equation reduces to
(i)
(4.5.23). The term qN lq (α) describes a hydrodynamic instability that is produced even
in the absence of temperature sensitivity.
The numerical studies of (4.5.36) using periodic boundary conditions in the transverse
direction have solutions that are periodic in time, with pulsating cells delimited by cusps
propagating at constant velocity on the front; see Fig. 4.34. In large boxes with many

[1] Clavin P., 2002, In H. Berestycki, Y. Pomeau, eds., Nonlinear PDE’s in condensed matter and reactive flows, 49–97. Kluwer
Academic Publishers.
[2] Clavin P., Denet B., 2002, Phys. Rev. Lett., 88(4), 044502–1–4.

0.4

Growth rate 0.2

0.0

–0.2

–0.4
0 2 4 6 8 10 12 14 16 18 20
n

Figure 4.34 (a) Growth rate as a function of mode number n in a periodic box of size L, k = 2π n/L,
for a detonation with Arrhenius kinetics, βN = 6, γ = 1.2, q = 0.25, 2π/L = 0.5. The curves
under the graph show the two-dimensional solutions for the front at three reduced times, τ ≡ t/tN .
(b) One period of propagation of a cell of a two-dimensional detonation front with the same kinetics.
Propagation is from bottom to top; the grey lines are the trajectories of the cusps. Reproduced with
permission from Clavin P., Denet B., Physical Review Letters, 88(4), 044502–1–4. Copyright 2002
by the American Physical Society.

unstable modes, the final periodic solution is obtained after a long transient time and
presents a cell size larger than the most amplified wavelength. The case presented in Fig.
4.34 corresponds to a box with 12 unstable modes, L = 4π . The wavelength selected in the
final solution corresponds to a wavenumber kn = 2π n/L with n = 6. The band of linearly
unstable modes ranges from n = 4 to n = 16 and the most unstable mode corresponds to
n = 8. The nonlinear selection of the final cell size is illustrated by the numerical solution
for an initial condition of a sinusoidal perturbation of small amplitude at n = 10, plus a
much smaller level of noise. After a few periods of oscillation, well-ordered pulsating cells
with a small size, corresponding to n = 10, are first observed. This regime lasts roughly
50 tN and is followed by a chaotic phase during which the number of cells decreases down
to the final stable state of six pulsating periodic cells. The relaxation time towards the final
solution is typically 2 × 102 tN . The pattern constituted by the trajectory of the cusps on
the front in the final solution is similar to the ‘diamond’ pattern observed in experiments;
see Fig. 4.34. Because of the nonlinear terms, the shape of the averaged front is not planar,
and a mean streaming flow is associated with the mean front.[3] To summarise, the model
equation (4.5.36) reproduces experiments and direct numerical simulations qualitatively
well. Based on the ZND structure, the model reduces to a simple superposition of the linear
oscillatory instability (due to the heat release) and the nonlinear dynamics of the lead shock

[3] Clavin P., 2002, Int. J. Bifurcation & Chaos, 12(11), 2535–2546.

that leads to the formation of triple points. This is true for weakly unstable detonations in
sense that they are still one-dimensionally stable (no galloping instability).

Strongly Unstable Detonations

Strongly unstable detonations are more complicated. A galloping instability can be
superimposed on the multidimensional instability so that pockets of fresh mixtures can
be engulfed in the shocked gas and reignited when the triple points cross each other.
Experiments show that this phenomenon can take the form of local micro-explosions
with an eventual intermittent character.[1,2,3] It should be possible to simply model this
phenomenon by introducing the nonlinear equation (4.5.21) for dynamical quenching into
(4.5.36).

Model Equation Near the CJ Regime

A complementary study of weakly unstable gaseous detonations near the CJ regime has
also been performed. These conditions concern small heat release[4] and the analysis is
presented in Sections 12.2.2 and 12.2.3. In such extreme (and nonrealistic) conditions the
situation is contrary to the previous case; the entropy–vorticity wave plays a negligible role
and the multidimensional instability results from the coupling of acoustic waves to the heat
release rate.
The results have similarities but also differences with the preceding study. The form
of the result for galloping detonations (one-dimensional oscillatory instability) is similar,
namely an integral equation involving a function of space N (x) but with a delay due
to the upstream running acoustic waves; see Section 12.2.2. The multidimensional study
presented in Section 12.2.3 also shows the existence of a Poincaré–Andronov (Hopf)
bifurcation at a finite wavelength. The instability is promoted by increasing the sensitivity
of the heat release rate to temperature or by approaching the CJ condition. But at the
leading order of the perturbation analysis, in contrast to the strongly overdriven cases,
the nonlinear model does not involve the dynamics of the weak lead shock. As a conse-
quence of the transonic character of the flow, the multidimensional dynamics of the deto-
nation wave is fully dominated by the modification of the heat release rate; see the end of
Section 12.2.3.

4.6 Appendix
4.6.1 Gaseous Shock Waves and Boltzmann’s Equation
Consider a steady planar shock wave perpendicular to the x-axis and propagating at constant
velocity. In the reference frame attached to the shock wave (Galilean frame) the shock struc-
ture is given by the steady solution to Boltzmann’s equation (13.3.2), (p1x /m)(∂f1 /∂x1 ) =

[1] Radulescu M., et al., 2007, J. Fluid Mech., 580, 31–81.

[2] Shepherd J., 2009, Proc. Comb. Inst., 32, 83–98.
[3] Bhayyacharjee R., et al., 2013, Proc. Comb. Inst., 34, 1893–1901.
[4] Clavin P., Williams F., 2009, J. Fluid Mech., 624, 125–150.

B(f1 ), where B(f1 ) is given in (13.3.5) and f1 ≡ f (r1 , p1 ), where r1 = (x1 , 0, 0) and
p1 = (p1x , p1y , p1z ).

Rankine–Hugoniot Relations
The Rankine–Hugoniot relations, studied in Section 4.2.1, are derived here directly from
Boltzmann’s equation using the notations of Section 13.3.1. According to (13.3.15), the
divergence of the flux of conserved quantities, mass, longitudinal momentum and energy,
is zero:

∂ ∂
d3 p1 p1x f1 = 0, d3 p1 p21x f1 = 0, (4.6.1)
∂x1 ∂x1

∂
d3 p1 p1x (p21x + p21y + p21z )f1 = 0. (4.6.2)
∂x1

Far from the shock wave, for x1 → ±∞, the gas is at equilibrium so that the distribution
(0)
function is the Maxwell–Boltzmann distribution f1±∞ in (13.3.19) with a set of param-
eters (n±∞ , u±∞ , T±∞ ) corresponding to the equilibrium states at x1 → ±∞, u±∞ =
(u±∞ , 0, 0).
Integrating the first equation in (4.6.1) from x1 = −∞ to x1 = +∞, using isotropy
and the normalisation of f (0) , d3 p(px − mu)f (0) = 0, d3 pf (0) = n, yields the mass
conservation (4.2.1) in the form

n+∞ u+∞ = n−∞ u−∞ . (4.6.3)

Integrating
3 the second
3equation in (4.6.1) from x1 = −∞ to x1 = +∞ and using the
relation d ppx f = d p(px − mu) f + n(mu)2 yields the momentum conservation
2 (0) 2 (0)

(4.2.3) in the form

n+∞ kB T+∞ + n+∞ mu2+∞ = n−∞ kB T−∞ + n−∞ mu2−∞ . (4.6.4)

Proceeding with (4.6.2) in the same way as before and using the relation

d3 p px (p2x + p2y + p2z )f (0)
(4.6.5)
= mu 3 d p(px − mu) f + n(mu) + d p(py + pz )f
3 2 (0) 2 3 2 2 (0)
,

which is obtained using isotropy of f (0) after replacing px by (px −mu)+mu, (4.6.2) yields,
according to (4.6.3) and (13.3.20),

5kB T+∞ + mu2+∞ = 5kB T−∞ + mu2−∞ . (4.6.6)

This relation corresponds to (4.2.4) where h = cp T and cp = (5kB /2m).

Irreversibility
We show now that the entropy must increase across a gaseous shock wave, which is the
proof that steady planar rarefaction shocks cannot exist. Consider the quantity

−1 ∂ −1 ∂f1
I≡m 3
d p1 p1x [f1 ln f1 ] = m d3 p1 p1x [1 + ln f1 ]. (4.6.7)
∂x1 ∂x1
Using Boltzmann’s equation, the H theorem (13.3.12) then shows that I cannot be positive,

p1x ∂f1
= B(f1 ) ⇒ I = d3 p1 B(f1 )[1 + ln f1 ] 0. (4.6.8)
m ∂x1
By definition, the variables x1 and p1 are independent variables so that the derivative with
respect to the space coordinate
3 x1 can be commuted with the integral over p1 , mI =
(∂J/∂x1 ) with J(x1 ) ≡ d p1 p1x [f1 ln f1 ]. Integrating I 0 from x1 = −∞ to x1 =
+∞ leads to J+∞ J−∞ , where J±∞ ≡ J(x1 = ±∞). Isotropy of f (0) , d3 p(px −
mu)[f (0) ln f (0) ] = 0, d3 p px [f (0) ln f (0) ] = mu d3 p[f (0) ln f (0) ], shows

(0) (0)
x1 → ±∞: J±∞ = mu±∞ d3 p1 [f1±∞ ln f1±∞ ]. (4.6.9)

According to the definition of entropy s in (13.3.14), ns = −kB d3 p[f (0) ln f (0) ] + kB n,
the relations J+∞ J−∞ and (4.6.3), n+∞ u+∞ = n−∞ u−∞ , then imply s+∞ s−∞ if
u±∞ > 0. In other words the entropy of the gas behind the shock cannot be smaller than
the entropy of the gas into which the shock propagates.
The detonation thickness d is estimated from the definition (4.6.7) of I, kB I =
−∂(nus)/∂x, and from the order of magnitude of the collision operator, B(f1 ) = O(δf1 /τcoll ),
where 1/τcoll is the collision frequency; see (13.3.24). Equation (4.6.8) leads to −kB I =
O(nδs/τcoll ), yielding u∂δs/∂x = O(δs/τcoll ) and d = O(uτcoll ). This corresponds to a
thickness of order of the mean free path d/ = O(1), = vτcoll , because the mean velocity
of molecules v is of order of the sound speed, v/a = O(1), and because the flow velocity
relative to the shock front is also of order of the sound speed, u/a = O(1), u > a on the
upstream side of the shock and u < a on the downstream side.
The large thickness (4.2.40) of weak shocks, d/ = O(1/(M − 1)) for a Mach number
M = u/a close to unity (M − 1) 1, can be explained by the small variations of the
thermodynamic parameters across a weak shock. For example, according to the Rankine–
Hugoniot relation (4.2.15), δp/p = O(M − 1), the nondimensional gradient, reduced by
the detonation thickness d, is small p−1 d ∂p/∂x = O((M − 1)), and not of order unity
p−1 d ∂p/∂x = O(1), as in an ordinary shock.

4.6.2 Reference Frame Attached to the Front

The nonlinear analysis is more conveniently carried out by working in a system of coordi-
nates attached to the wrinkled front, ξ = x − α(y, t),
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
→ , → − α̇t , → − αy ,
∂x ∂ξ ∂t ∂t ∂ξ ∂y ∂y ∂ξ

the boundary condition (4.4.23)–(4.4.24) being applied at ξ = 0. It is also convenient to

introduce the flow splitting,

p = pN + δp(a) + p̃, u = uN + δu(a) + ũ, w = δw(a) + w̃,

where, according to the text below (4.4.16), the flow velocity of the acoustic waves (δu(a) ,
δw(a) ) is smaller than the linear part of (ũ, w̃) by a factor 2 . The continuity and Euler
equations may then be written for H = 0 in the form

∂ ũ ∂ w̃ ∂ ũ ∂ ũ 1 ∂ p̃ ∂ w̃ ∂ w̃ 1 ∂ p̃
+ = X̃, + uN = Ũ − , + uN = W̃ − ,
∂ξ ∂y ∂t ∂ξ ρ N ∂ξ ∂t ∂ξ ρ N ∂y
where the residual variation of density can be neglected since it introduces nonlinear cor-
rections that are smaller than ε. When the linear solution (4.4.18) is introduced into the
terms X̃, Ũ, W̃, they take the form of source terms that introduce nonlinear corrections of
order ε to the linear dynamics,

D
X̃ = − α (y, t)α̇ty (y, t − ξ/uN ),
uN y
α̇t (y, t)α̈tt (y, t − ξ/uN ) D
Ũ = − , W̃ ≡ − α̇t (y, t)α̇ty (y, t − ξ/uN ),
uN uN

where, using the relation DuN ≈ a2N and (4.4.15), ∂ Ũ/∂ξ + ∂ W̃/∂y = ∂ X̃/∂t + uN ∂ X̃/∂ξ ,
so that the pressure term is a solution to Laplace’s equation ∂ 2 p̃/∂ξ 2 + ∂ 2 p̃/∂y2 = 0. The
acoustic pressure being excluded from the pressure term, p̃ is a quadratic term which must
(i)
be at least of order ρ N w̃2 ≈ ρ N (w2 )2 to be nonnegligible. The boundary condition (4.4.23)
(i) 2
introduces a term of order ρ N a2N αy2 , which is smaller than ρ N (w2 )2 ≈ ρ N D αy2 by a
factor of 2 . The bounded solution of Laplace’s equation with a zero boundary condition at
ξ = 0 is null, p̃ = 0. Therefore, retaining the correction of order ε and neglecting terms of
order 2 , the flow (ũ, w̃) is a solution to
∂ ũ ∂ w̃ ∂ ũ ∂ ũ ∂ w̃ ∂ w̃
+ = X̃, + uN = Ũ, + uN = W̃. (4.6.10)
∂ξ ∂y ∂t ∂ξ ∂t ∂ξ
The solution to the second and third equations in (4.6.10) subject to the boundary condition
(4.4.25) is obtained by noticing that every function of t and ξ through the grouping (t −
ξ/uN ) is a solution of the homogeneous equations (without second member),

ũ = ũf (y, t − ξ/uN ) + Ũ(ξ , y, t), w̃ = w̃f (y, t − ξ/uN ) + W̃(ξ , y, t), (4.6.11)

[α(y, t − ξ/uN ) − α(y, t)]α̈tt (y, t − ξ/uN )

where Ũ ≡ ,
uN
D
W̃ ≡ [α(y, t − ξ/uN ) − α(y, t)]α̇yt (y, t − ξ/uN ),
uN

and where, according to the definitions of Ũ , W̃, Ũ and W̃,

∂ Ũ ∂ Ũ ∂ W̃ ∂ W̃
+ uN = Ũ, + uN = W̃, ξ = 0: Ũ = W̃ = 0.
∂t ∂ξ ∂t ∂ξ
Using (4.4.15) and H = 0, the flow (Ũ, W̃) is found to satisfy the relation
∂ Ũ ∂ W̃
+ = X̃.
∂ξ ∂y

4.6.3 Mass-Weighted Coordinate

In unsteady one-dimensional flow problems with variable density and when the boundary
condition is given at a moving boundary, x = α(t),
x = α(t): ρ = ρf (t), u = uf (t), (4.6.12)
it is convenient to introduce a system of coordinates (x, t), based on the nondimensional
mass-weighted coordinate, x, and also a velocity u(t) depending only on time,

1 x ρf (t) dα
x≡ ρ(x , t)dx and u(t) ≡ uf (t) − , (4.6.13)
ρ α(t) ρ dt
∂ ρ ∂ ∂ ∂ ∂ ∂
so that → , + u(x, t) → + u(t) , (4.6.14)
∂x ρ ∂x ∂t ∂x ∂t ∂x
where the last relation is obtained by using continuity, the first equation in (15.1.33),
∂ρ/∂t + ∂(ρu)/∂x = 0,
x x
∂ ∂(ρu)
ρ(x , t)dx = −
dx = ρ(x, t)u(x, t) − ρf (t)uf (t). (4.6.15)
∂t α(t) α(t) ∂x

The advantage of the mass-weighted coordinate is to eliminate the variable coefficient

ρ(x, t)u(x, t) in favour of a coefficient depending only on time, u(t).
The equation for conservation of mass (continuity), ∂ρ/∂t + u∂ρ/∂x = −ρ∂u/∂x, takes
the form

∂(1/ρ) ∂(1/ρ) ∂u
ρ + u(t) = . (4.6.16)
∂t ∂x ∂x