MAP551.

Systèmes dynamiques pour la modélisation et la simulation

des milieux réactifs multi-échelles (2023-2024)

PC1: the theory of thermal explosion

according to Semenov and Frank-Kamenetskii

Parts which have to be treated in the report: 2.1.1, 2.1.3, 2.1.4, 2.2.

1 Introduction
In this Petite Class we discuss the mathematical modeling of explosion phenomena. When we study
the evolution of the chemical composition and of the temperature of a tank filled with fluid energetic
material, in order to know if there is a risk of thermal runaway and then of explosion1 , we can tackle
the problem by studying the coupling of hydrodynamics and of the chemistry/thermic associated with
the reactions which take place in the medium [3]. However, before turning to a computer for numerical
simulations, one has to choose within a hierarchy of simplified systems which takes into account various
physical or chemical phenomena; this choice then implies a range of characteristic scales and one must in
general have an idea about the mathematical structure of the system and bring a precise understanding
on the dynamics of the system and its qualitative behavior, before being able to simulate it. However,
this understanding is based on two key ingredients:

1. Knowing how to identify the right modeling hypotheses allowing to conduct a rapid and relevant
analysis of qualitative behavior and therefore to work with the right level of modeling,

2. knowing the theory of dynamical systems in order to understand the possible behavior of the system
under consideration.

Before getting to the heart of the matter, we give some historical background explaining how and
when such theory was introduced, and which difficulties various scientist had to overcome in order to get
there. Students interested in more details are can consult [7] and [1].

1.1 Some historical landmarks

The Petite Class focuses more particularly on explosive reactions. Their main characteristics were studied
in the mid-nineteenth century by Van’t Hoff2 and Bunsen3 . They have the peculiarity of having a reaction
rate which depends in a non-linear and strong way on the temperature (as opposed to slow reactions
whose dependence is low). This property of strong dependence is due to a large activation energy of
the combustion reaction, which usually results in a high thermal power of the reaction. Under these
conditions, a thermal autocatalysis phenomenon can occur, causing the reaction to runaway and a very
sharp rise in temperature; this phenomenon is called thermal explosion. Various experimental works
then led Nikolay N. Semenov4 to develop the theory of thermal explosion in the 1920s for the case of
a homogeneous reactor [2]. He graduated in physics, but ended up working at the interfaces of several
disciplines (mathematics, physics, chemistry ...) in the pure Russian tradition of the time. His work
1
We worked with SME-SNPE a few years ago to study the explosion limits of this type of tank for safety reasons. In
2011, SME (and its subsidiary Roxel) was acquired by Safran and then merged with Snecma Propulsion Solide to create
Herakles, which was then absorbed into ArianeGroup.
2
Jacobus Henricus van’t Hoff (1852-1911), is a Dutch Nobel Prize-winning chemist in 1901. His main research work in
theoretical and physical chemistry concerned the writing and modeling of reactions by systematically taking into account
thermodynamic data, chemical balances and reaction rates. He was a forerunner of rigorous theoretical chemistry and
contributed to the creation of physical chemistry as we know it today.
3
Robert Wilhelm Bunsen (1811-1899) is a German chemist, who gave the Bunsen burner its name.
4
Russian scientist (1896-1986), Nobel Prize in chemistry of 1956, shared with Cyril Norman Hinshelwood, laureate of the
gold medal Lomonosov of USSR Academy of Sciences (1970).

1
mainly covers the study of the mechanisms of chemical transformations and in particular an exhaustive
study of the theory of chain reactions, with applications in particular to combustion processes. David A.
Frank-Kamenetskii5 then extended this theory in the non-homogeneous case and proposed an exhaustive
treatment in a certain number of cases [4, 5]6 . Invited by Semenov, to whom he had sent a letter on
chemical thermodynamics, he worked for seven years from 1934 to 1941 at the Institute of Physical
Chemistry of the Russian Academy of Sciences founded by Semenov on thermal explosion problems and
met another prominent figure of the Russian combustion community: Yakov Borisovich Zel’dovich7 , with
whom he became a very good friend.

1.2 Objectives
The aim of this Petite Class is to introduce these theories and to understand the basic mechanisms
of coupling between the thermal source term resulting from the chemical reaction, thermal conduction
and thermal losses at the boundaries of the domains under consideration. We will then try to obtain
the critical explosion conditions, i.e. the limiting values of certain parameters which determine whether
the explosion phenomenon occurs or not. Finally, the last point we will tackle is the coupling between
thermal explosion and natural convection in the presence of the gravitational field. These studies are
important and find many applications in determining the safety conditions for the storage of hazardous
materials, such as energetic materials that are used in rocket propulsion, for example. The goal of this
Petite Class is also, in the context of a typical example, to understand the various behaviors associated
with a hierarchy of dynamical systems modeling a same physical phenomenon, the structuring of its
dynamics around the critical points and the influence of the notions of stability and bifurcation on the
behavior of the system.
In this study, we consider explosive phenomena located within a reactor, but neglect spatial inho-
mogeneities and in particular do not treat propagation phenomena, which is significantly more difficult
problem, as it requires taking into account the effects of mass density and pressure since shock waves
or detonations can form. We will focus later in the course on constant pressure explosion models which
describe the propagation of a flame at low speed compared to the speed of sound. The founding study
in this field dates back to 1928 with the work of Zel’dovich [2, 7]. For the presentation, we are inspired
by the book [7].
Notation: We will use the following notations for derivatives, dt φ(t) denotes the time derivative of
the function φ(t).

2 Explosion in an homogeneous reactor

2.1 Model 1 - adiabatic explosion
The goal of this part is to investigate, within the framework of a minimal modeling, the time scale of
the problem. We set ourselves within the framework of a simple chemistry, made up of a single global
irreversible reaction “Fuel → Products”. The state of this reaction is given by the scaled fuel mass
fraction: Y (t); which is initially equal to 1 when the fuel has not yet started to react and to 0 when all
the fuel present in the reactor has been consumed into products and heat. The temperature of the reactor
at the initial time t = 0 is denoted T0 and changes due to the chemical reaction: we denote by T (t) the
temperature at time t. The evolution of the mass fraction Y (t) and of the temperature T (t) is described
by a system of two ordinary differential equations. They involve the speed of the reaction which depends
in a non-linear way on the temperature. This model describing the evolution of the reactor writes:
E
dt Y = −B e− RT Y, (1)
E
− RT
dt T = Tr B e Y. (2)
5
Russian scholar (1910-1970) in the fields of theoretical physics and chemistry.
6
People wishing to consult the book can ask M. Massot
7
Russian scholar (1914-1987), is a Soviet astrophysicist and theoretical physicist. He developed the theory of chain
reactions. He played an important role in the development of the Soviet nuclear and thermonuclear weapons. He made
important contributions in the fields of adsorption and catalysis, shock waves, nuclear physics, particle physics, astrophysics,
cosmology and general relativity.

2
B is the frequency factor (homogeneous to the inverse of a time), E is the activation energy of the
reaction and Tr = Tb − T0 , where Tb is the temperature of the mixture when all fuel is consumed. This
temperature is associated with an enthalpy of reaction, with the thermal capacity of the medium and
with a quantity of fuel present in the reactor. We suppose in this problem that the reaction is exothermic
and therefore that Tr > 0. Moreover B, E and R are strictly positive constants. In this model, only
the heating of the medium by the reaction is taken into account and it is assumed that the system is
thermally isolated from the outside; this is then referred to as an adiabatic thermal explosion.

Remark: For this Petite Class, we take for granted that the system (1) - (2), provided with the
initial condition Y (0) = 1, T (0) = T0 , has a unique solution which exists for all t > 0, is of regularity
C ∞ and satisfies Y (t) ∈]0, 1], T (t) > 0 for all t ≥ 0. We will come back to these theoretical aspects in
more detail in the following sessions.
2.1.1 The purpose of this question is to obtain information about the qualitative behavior of the solution
of system (1)-(2).
2.1.1.a Find a quantity H, which is a linear combination of Y an T , and which remains constant
when Y and T solve (1)-(2). Give the value of H in terms of the initial data, and derive an
autonomous differential equation on the variable Y (t) only:

dt Y = Φ(Y ),

for a function Φ that you will explicit. Similarly, derive an autonomous differential equation
describing the evolution of the temperature T (t) under the form dt T = Λ(T ), and give an explicit
formula for Λ(T ).
2.1.1.b Show that T (t) is increasing and bounded from above by Tb . What can you conclude
regarding the limit of Y (t) and T (t) as t goes to +∞?
In general we cannot explicitly solve nonlinear differential equations, and the equation obtained above on
the temperature is no exception. In order to get a more precise description of how temperature evolves
with time, one option would be to try and simplify the model so that we can solve it explicitly, another
one to use numerical simulations. We are actually going to do both, but before that we make some
assumptions on the parameters, based on two characteristic features of combustion reactions.

[H1] The reaction satisfies the assumption of high activation energy, meaning that

E
β= >> 1. (3)
RT0

[H2] The reaction satisfies the assumption of high enthalpy, meaning that Tr = Tb − T0 > T0 .

Notice that [H1] + [H2] implies

Tr Tr
T̄r = 2 =β >> 1, (4)
RT0 /E T0
where RT02 /E is called the Frank-Kamenetskii temperature and is denoted by TFK .
2.1.2 We now derive a simplified version of system (1-2).
2.1.2.a Consider the following rescaled, non-dimensional version of the temperature: Θ = (T −
T0 )/TFK , which amounts to using the initial temperature as the reference temperature and the
Frank-Kamenetskii temperature as the variation scale, and write down the equation it satisfies.
2.1.2.b Assuming [H1] and [H2], show that as long as Θ << β, Θ can be approximated by a
function Θe which solves the following ODE:

1 Θ
dt Θ
e = e , (5)
e
tI

with initial condition Θ(0)

e = 0. The parameter tI , which can be seen as a characteristic timescale,
is called induction time. Express it in terms of (some of) the other parameters of the model.

3
 2.1.2.c Introduce the non-dimensional time τ = t/tI , and show that θ(τ
e ) = Θ(t)
e satisfies

dτ θe = eθ , (6)
e

still with θ(0)

e = 0.
2.1.3 We now study the simplified system (6).
2.1.3.a Solve (6) and write down an explicit solution. Why can the behavior of this solution be
described as “explosive”?
e ) for τ ∈ [0, 1[. Comment upon what you observe,
2.1.3.b Plot the time evolution of the solution θ(τ
in particular with regards to the assumptions that were made to obtain equation (6).
2.1.4 Consider again the equation obtained on Θ in question 2.1.2.a without the approximation Θ  β,
and derive an equation of the form
dτ θ = λ(θ), (7)
where θ(τ ) is the adimensonalized temperature in the adimensionalized timescale, i.e. θ(τ ) = Θ(t).
Show on two graphs the function θ 7→ λ(θ), θ ∈ [0, T̄r ], for β = 30 and β = 100. To what extent do you
expect the explosive behavior that was observed in the simplified system (6) to still be present in the full
system (7), and how is it impacted by the value of β?
2.1.5 – To go further – The goal of this question is to further justify that (6) is a “good approximation”
of (1)-(2), and to better relate the solution obtained in the previous question to the one of (1)-(2).
2.1.5.a Calculate the temperature corresponding to the maximum rate of action, in absolute value,
for the system (1)-(2), where the action rate is the r.h.s. of the equation on Y , namely −B e−E/RT Y .
How does this temperature compare to Tb ?
2.1.5.b What scenario can we deduce about the evolution of the system (1)-(2), given that the
maximal reaction rate is reached in a close neighborhood of tI ?
2.1.5.c Describe qualitatively the evolution of the system in the plane described by the dimensional
variables (Y, T ), Y ∈ [0, 1] and T ∈ [T0 , Tb ] (also called phase space), recalling that H must remain
constant. Highlight three distinct stages in the evolution of the system.
2.1.5.d Consider B = 107 s−1 , T0 = 400K, Tb = 3000K and β = 308 . Plot the functions Y (t) and
T (t). You can use for that the part of the Notebook corresponding to Section 2.2, with α0 = 0.
What happens for the fuel consumption (characterized by the fuel mass fraction Y ) during the
time interval t ∈ [0, tI [? Interpret assumption [H2]. Comment upon the obtained results.

2.2 Model 2 - explosion taking into account thermal losses

In the previous subsection, the absence of thermal losses at the boundary of the reactor systematically
generates the thermal explosion because the temperature can only increase and reach Tb (which is typically
of the order of 3000K for combustion reactions). Under these conditions, the model ceases to be valid
when we arrive at high temperatures within short periods of time, because other phenomena then come
into play (creation of pressure waves, etc.).
In this subsection we remove the adiabaticity hypothesis, but keep the spatial homogeneity. This
provides us with a more realistic model than in the previous part, where thermal losses at the boundary
can prevent the explosion if they compensate for the heat source term resulting from the chemical reaction.
Our main goal will be to characterize the conditions of explosion, that is, to say in which cases thermal
losses at the boundary prevent the explosion and in which cases they are not sufficient to limit the
temperature growth of the system.
In the case of a closed reactor, a simple way of modeling thermal losses is to set the wall temperature
to the reference temperature T0 . The reactor can then be modeled by the following new system, for t > 0,
E
dt Y = −B e− RT Y, (8)
E
− RT 1
dt T = Tr B e Y − (T − T0 ), (9)
tp
8
The value of β can be of the order of 50 or 100 for certain combustion reactions, but we choose here a value of 30 for
reasons of convenience, in order to simplify the numerical simulations.

4
where tp is the characteristic time of heat transfer with the wall, and with the same initial conditions as
before. We again assume that [H1] and [H2] hold, and that there is a unique smooth solution to (8)-(9),
which satisfies Y (t) ∈]0, 1] and T (t) > 0 for all t ≥ 0.
2.2.1 We first study the qualitative behavior of the solution to (8)-(9).
2.2.1.a Show that T (t) > T0 for all t > 0, and that the quantity H introduced previously is no
longer conserved for solutions of(8)-(9) where thermal losses are taken into account. Compare H(t)
with H(0), and show that T (t) < Tb for all t ≥ 0. Can you make a conjecture regarding the large
time behavior of the solution of (8)-(9)?
2.2.1.b Using the same non-dimensionalization procedure as in the previous section, with Θ =
(T − T0 )/TFK and a non-dimensional time τ = t/tI , we can rewrite system (8)-(9) as
1 1+θ/β
θ
dτ y = − e y, (10)
T¯r
θ
dτ θ = e 1+θ/β y − α0 θ, (11)

where α0 = ttpI > 0. Consider again T0 = 400K, Tb = 3000K and β = 30. Plot the time evolution of y
and θ for different values of α0 . Exhibit two very different behavior depending on the value of α0 . For
which approximate value of α0 does the transition between these two regimes take place?
2.2.2 Our aim is now to understand these two drastically different behaviors of the temperature, and in
particular to try and predict when each of them occurs. One difficulty is that, because H is no longer
conserved, we can no longer get a single equation depending only on the temperature. However, we can
still recover such a situation if we first simplify again the system.
2.2.2.a Make some assumptions, under which equation (11) on θ can be approximated by

dτ θe = eθ − α0 θ. (12)
e e

On which time interval can we consider that equation (12) provides a physically meaningful descrip-
tion of system (8)-(9) ? What can we say about the fuel consumption during this time interval?
2.2.2.b Study the stationary solutions of equation (12), namely the solutions of eθe − α0 θe = 0,
depending on the value of α0 . Distinguish between three different cases. In particular, exhibit a
critial value αcr of α0 , and give the temperature θecr of the corresponding steady state. Compute the
corresponding dimensional temperature Tcr and comment upon its order of magnitude (assuming
[H1] and [H2]).
2.2.2.c Classify the long time behavior of the solution of (12), depending of the value of α0 and
of the initial temperature θ(0).
e In particular, what happens when α0 is very small? very large?
Give a physical interpretation for these behaviors. How do the results you obtain for the simplified
model (12) compare with what you observed numerically for the original model with thermal
losses (8)-(9)?

2.3 Model 3 - explosion taking into account convection

In the previous subsection, we have taken into account heat losses at the reactor boundaries in the
form of a term proportional to the difference between the mean reactor temperature, assumed to be
homogeneous and the ambient temperature remaining at T0 . Frank-Kamenetskii removed this assumption
of homogeneity and proposed a model in the form of a partial differential equations on the temperature,
for which one imposes Dirichlet conditions at the boundary. The heat flux at the boundary is related to
the temperature gradient of the solution near the domain boundary and changes over time as a function
of the solution. We will come back in the course to this type of models in the chapter on spatially
extended systems.
In the context of a fluid layer of thickness L and in the presence of gravity, the fluid in the center
heats up due to the exothermic reaction and generates the appearance of natural convection, i.e. the
establishment of convection vorticies [3, 6]. In such a case the heat fluxes at the boundaries depend on
the strength of the convection in the reactor.

5
 We propose here to show the influence of convection on the dynamics of the system by introduc-
ing a new variable, which models the intensity of the convection in the reactor, while neglecting fuel
consumption:

dτ θe = eθ − α(Ψ) θ, (13)
e e

dτ Ψ = −a Ψ (Ψ2 + θec − θ),

e (14)
where Ψ is a non-negative number which characterizes the intensity of convection phenomena (for instance
the maximum of the stream function) and θec is a critical temperature above which the steady state Ψ = 0
of the second equation becomes unstable, corresponding to the onset of convection phenomena. This
critical temperature is associated to a critical Rayleigh number, as for natural convection [3, 6]. The
parameter a characterizes the relaxation rate of natural convection toward its stable state. The rate of
heat transfer at the boundary is assumed to depend on the intensity of the convection Ψ in the following
way:
α(Ψ) = α0 (1 + µΨ2 ). (15)
In the forthcoming numerical experiments, we consider a = 4.3, µ = 0.7 and various values of θec and
α0 , including θec = 1.035, α0 = 2.7 which corresponds to one of the models studied in [3]. Here are some
possible initial conditions (θ(0),
e Ψ(0)) = (0, 0.01) or (θ(0),
e Ψ(0)) = (1.0, 0.5), but feel free to try out other
one to better investigate different regimes.
2.3.1 If Ψ = 0, how does this model relates to what we have done previously. What do we know in that
case?
2.3.2 If α0 > αcr , discuss the behavior of solutions of (13)-(14) and the influence of taking convection
into account using numerical simulations. You can for instance consider α0 = 3 and explain qualitatively
the long time behavior of the solutions based on the results of the previous subsection.
2.3.3 Repeat the same experiment with α0 = 2.7 and θec = 1. Again, comment upon what you observe,
try to interpret it physically, and to compare with what happens when there is no convection (Ψ = 0).
2.3.4 Finally, keep α0 = 2.7, and consider only the initial condition (1.0, 0.5), but study the influence
of the parameter θec . Comment upon what you observe, and try to give a physical interpretation for the
different regimes.

References
[1] H. Berestycki, C.M. Brauner, P. Clavin, and C. Schmidt-Lainé. Modélisation de la Combustion.
Images des Mathématiques. CNRS, 1996.
[2] A.A. Berlin, Y.V. Frolov, and Y.G. Isaevich. The contribution of the Semenov Institute of Chemical
Physics to the Science of Combustion: A historical review. In Chemical Rocket Propulsion, pages
1055–1068. Springer, 2017. Springer Aerospace Technology book series, De Luca L., Shimada T.,
Sinditskii V., Calabro M. (Editors).
[3] T. Dumont, S. Génieys, M. Massot, and V. A. Volpert. Interaction of thermal explosion and natural
convection: critical conditions and new oscillating regimes. SIAM J. Appl. Math., 63(1):351–372,
2002.
[4] D. A. Frank-Kamenetskii. Towards temperature distributions in a reaction vessel and the stationary
theory of thermal explosion. Doklady Akad. Nauk SSSR, 18, 1938.
[5] D. A. Frank-Kamenetskii and N. Thon. Diffusion and Heat Exchange in Chemical Kinetics. Princeton
University Press, 1955.
[6] A. V. Getling. Rayleigh-Bénard convection, volume 11 of Advanced Series in Nonlinear Dynamics.
World Scientific Publishing Co., Inc., River Edge, NJ, 1998. Structures and dynamics.
[7] Y. B. Zel’dovich, G. I. Barenblatt, V. B. Librovich, and G. M. Makhviladze. The mathematical theory
of combustion and explosions. Consultants Bureau [Plenum], New York, 1985. Translated from the
Russian by Donald H. McNeill.