Twenty-Third Symposium (International) on Combustion/The Combustion Institute, 1990/pp.
1-19
CHEMISTRY AND
PETER GRAY
Gonville and Caius College
Cambridge CB2 1TA
1. Introduction
The purpose of this introductory lecture is to
speak to an audience much less specialised than the
groups who will split up for our many particular
sessions. Our size is both a source of satisfaction
and a disadvantage for we are now unable to be a
single audience so much have we grown since our
early beginnings. The success of the Combustion
Institute is undoubted, and our sequence of Sym-
posium volumes represents a formidable achieve-
ment. But we must recognize that we are begin-
ning to lose something by the very size to which
we have grown when we attempt to meet in our
traditional style and in our present-day numbers. I
hope the Board of Management will think deeply
about this.
I have chosen as my broad title "Chemistry and
Combustion" because of my own training in chem-
istry. If I expanded upon it, I would say I wish to
concentrate on what is called discontinuous behav-
iour and on its chemical origins. The type of dis-
continuous behaviour that I have in mind is the
abrupt onset of change--an explosive event, or the
sudden extinction of a reaction in response to a
continuous change in circumstances. Ignition and
extinction are instabilities that mirror one another.
A rather different form of instability, of interest since
early research days, is oscillatory behaviour or re-
petitive ignition and we shall also consider this.
All these exotic phenomena are old friends in the
combustion field, though in other circumstances they
are seen as strange and uncommon events. The
questions I should like to air are: broadly what it
is about combustion that encourages them, and how
is chemistry so naturally endowed as to make them
readily apparent? One sort of answer is engagingly
given by the Russian polymath, Lomonossov, 250
years ago. He said he was convinced by what. he
read and what he did, that "chemical experiments
combined with physical showed peculiar effects."
This is what I wish to explore today. Such general
words can be made to bear many loads of inter-
pretation. However, the balance between the loss
and the production of heat during chemical change,
and the influence of temperature upon chemical
change are just such conjunctions of physics and
COMBUSTION
chemistry. Reactions go faster when they are hotter
and slower when they are cooler. When a chemical
reaction is accompanied by self-cooling, we may see
a self-repressing influence. That is a "negative feed-
back"--such as is exploited in electrical circuits to
enhance stability--the kind of internal connexion
that le Chatelier may have made when attempting
to frame his 20th century synthesizing generalisa-
tions of J. H. van't Hoff's earlier, more quantitative
studies. When the reverse holds, and chemical
change is accompanied by self-heating we meet a
self-enhancing influence, a "'positive feedback." De-
stabilization has been built in. A system apparently
"in equilibrium" may now respond very dramati-
cally to a small perturbation. In the language of le
Chatelier, the changes within the system may tend
to reinforce, not to nullify, the "constraint."
We have thus already recognized and in w we
examine the features that distinguish combustion
from mere oxidation and that give rise to its dis-
continuities: feedback and non-linear responses.
Feedback means the effect of consequences upon
their causes. The non-linearity of thermokinetic re-
sponse is described in the famous empirical rule that
a chemical reaction doubles its rate when the tem-
perature is made ten degrees higher. So a span of
temperature of one hundred degrees celsius might
see a thousand-fold change (2t~ = 1024) in speed;
a span of two hundred degrees produces a million-
fold enhancement.
Chemistry has already entered twice over. Firstly,
the fact that rate-processes are so sensitive to tem-
perature reflects the fact that energy needs of
breaking bonds takes priority over energy release
as we climb the energy contours; and secondly, be-
cause the chemical energy so required is returned,
with more besides, as stabler final products are
formed at the end of the reaction path.
Chemistry can also enter combustion in a still
more disect manner than that above. This is by the
intervention of processes in which the chemical
product of an elementary step can catalyze its own
production without the need for energy release. The
name for this family of self-replicating processes is
autocatalysis, and where the overall reaction is ac-
complished by a chain of elementary processes, it
is called chain-branching. These chains of elemen-
 HOTrEL PLENARY LECTURE
tary steps are of supreme importance in the com-
bustion of hydrogen and they also confer on hy-
drocarbon oxidation many of its striking features.
They are not inevitable in all oxidations, and such
features seem not to matter so much, for example,
in the combustion of hydrazine. We shall have more
to say about them below. It is of some interest to
note that whilst the ideas of simultaneous chain-
branching and self-heating were very firmly grasped
more than forty years ago by Semenov and by Nor-
rish, they were not successfully joined together and
coherently expressed until the late 1960s. Proto-
type model schemes required simultaneous chem-
ical confidence and an adequate mathematical train-
ing: mathematics has often seemed a stony pasture
to western chemists, and that perception was the
reason for the delay.
This lecture will not be much concerned with
chain-branching and detailed chemistry, because
merely to sketch the field would make the treat-
ment too shallow. Even the model schemes pre-
sented will have to be mere indications of what is
a lively field today.
This decision to omit detailed chemical reaction
schemes and to place the emphasis on thermal
feedback does not mean that the importance of 'real
chemistry' is not appreciated. (This point will be
taken up again in the concluding section.) Rather
the development is to be taken as illustrating how
much can be obtained when real chemistry can be
fairly represented by certain characteristic laws.
The arrangement adopted is as follows. First, an
outline is given in w167 of the foundation of what
we now call thermal explosion theory. These are
the elementary treatments of spontaneous ignition
in homogeneous systems, in heterogeneous circum-
stances and in continuous-flow, stirred-tank reactors
(cstr). These are called elementary for several rea-
sons. First, they assume the possibility of stationary
states, and they identify critical ignition with their
disappearance. This is satisfactory for open systems,
but for the first two subjects it implies neglect of
reactant consumption. Secondly, they assume uni-
formity of temperature in the reactant. This is rea-
sonable for small and well-stirred systems but is
otherwise either an approximation or restricted to
systems where reactants are liquid or gaseous and
are mechanically mixed. Nevertheless, these are the
investigations that establish, as it were, the ere-
dentials of the subject and allow us to move on to
many applications: Amongst these are the treat-
ment of parallel reactions, of autocatalytie exother-
mie reactions, and the investigatibn of closed sys-
tems (w167 in which reactant consumption is not
ignored.
Other major topics, which could also be ac-
corded the status of foundations, so basic are they,
include the treatment (in w of systems in which
temperatures are not uniform but where the inten-
sity of self-heating varies from point to point. Again,
unless one has immediate recourse to numerical
computation, reactant consumption must be set on
one side and convenient approximations made. Here
the family of applications subsequently presented
have to be chosen selectively from many. Among
those illustrated are the use of theory to cope with
systematic errors in kinetic parameters (evaluations
of reaction order, activation energy, etc.) in the
presence of self-heating, the question of safety lim-
its on the 'hot assembly' of potentially dangerous
materials (w and the response of exothermic ma-
terials to steadily ramped ambient temperatures,
such as are often employed in thermal analyses.
Inevitably the very diversity of apphcations which
deserve to be sketched collectively means that in-
dividual explanations must sometimes be overeom-
pressed. Some relief can be found by consulting
monographs such as references 2, 18, 25, 32 and
33.
2. Foundations: Representing Non-linearity
and Measuring the Intensity of Feedback
The systematic study of thermal instabilities be-
gins with N. N. Semenov's (1928) treatment I of
"'critical conditions" for the self-ignition of a ho-
mogeneous mass. Ten years later (1938) his pupil
D. A. Frank-Kamenetskii extended the same ideas
to exothermic reactions occurring at a surface such
as burning carbon or a platinum catalyst. In 1941,
Ya B, Zerdovich offered the same service to the
simplest of open systems--the continuous-flow,
stirred-tank reactor (cstr). These three great men
are unsurpassed in combustion, and these
contributions1-3 are fundamental. It must suffice to
sketch them briefly. We indicate their broad sig-
nificance and use a common language to emphasize
their inter-connexions and extensions to more com-
plex situations. Amongst the earliest of complexities
are (1) looking at temporal evolution of temperaure,
(2) considering systems in which temperatures are
not uniform but vary in space, (3) broadening the
enquiries to chemical reaction schemes that are
themselves capable of becoming unstable isother-
mally. Finally, (4), yet overlapping with all these,
comes unstable behaviour of a different kind: os-
cillatory behaviour.
In all of this, we are concerned with essentially
static media and with negligible pressure-differ-
ences.
In his original work in this field neither Semenov
nor his colleague O. M. Todes chose to work with
dimensionless groups: both Frank-Kamenetskii and
Zel'dovich, coming not long after, did. We shall ex-
ploit their choices and express some of the early
 CHEMISTRY AND COMBUSTION
resuhs in these terms. At the present day there are
some tendencies to return to older forms: the Ap-
pendix gathers some comments together on this.
Non-Linearity of Response:
The principal source of non-linearity is the re-
sponsiveness of reaction-rate to temperature. The
Arrhenius form
k ~ exp ( - E / R T )
if not exact, is a splendid approximation over a sub-
stantial temperature range, not only for many ele-
mentary reactions but for a number of complex ones
too. It loses its relevance if T is too low or too high
(but 'too high' here means temperatures of order of
E/R, say 104 K, that are quite out of reach of self-
heating, chemical energy-release).
If k oc exp ( - E / R T ) , the responsiveness of k to
T is given by:
dk k
dT (RTZ/E)
This marks the entry of the group of terms RT2/E
as the natural yardstick of temperature. In ignition
problems, dimensionless temperature-excvsses | are
defined by the relationship:
O
T - Try
(aT~//EI '
and relative reaction rates by:
k(T,f) exp ~-~, ~ exp O
The simple exponential function e ~ is to be
regarded as an approximation to the term exp
[O/(1 + ~O)] with t = RT,-f/E and eO "~ 1.
Thermal Feedback and the Zel'dovich Number:
If an exothermic reaction can heat itself up by
an amount AT,d where
AT,~, = Q~co/crc~or Qvco/~rcv
then the quotient
aTaa
Oad(or B)
(nT~/E)
rather than ATad itself is the logical measure of the
intensity of thermal feedback. The symbol B is
gaining general currency, and in flame theory the
quantity B = (EATad/RT2aa) is called a Zel'dovich
number.
3. Unifying Treatments of Homogeneous
and Heterogeneous Systems
It is handy to set out Semenov's treatment of the
homogeneous exothermic reaction and its stability.
When reactant consumption is neglected, the heat-
balance equation ~ = .~, can be written:
QkcoV = hS(T - T~),
with Q denoting the exothermicity (--AU~ for a
system of constant volume, V) and k the reactiun-
rate constant. S is the surface area of the system,
and h the heat-transfer coefficient per unit area. It
is assumed that:
k = A exp ( - E / R T ) = ka exp{(E/RTa) - (E/RT)}
The use of co expresses one of the major assump-
tions: that reactant-consumption may be neglected.
This allows the problem to be reduced to one of
discovering the existence, multiplicity and stability
of stationary states. The stationary-state condition
can in turn be set out as:
(T-Ta, lEE}
hS(RTa2/E) ~ (Se) = ~ - ~ ) - ~ exp
RT
In dimensionless terms and for small temperature
excesses:
(Se) ~ Oe -~
The group of terms equated to (Se), the Seme-
nov number, may be regarded as a dimensionless
measure of exothermicity or a comparison of char-
acteristic rates (or relaxation times) for temperature
change by chemical heating or Newtonian cooling.
According to the size of (Se), the heat balance
equation has either two solutions (O+ and (9_) or
none. The point of change (bifurcation) is to be
identified with criticality. There, stable (O_) and
unstable (O+) stationary states eualesce, and at this
point:
(Se)~ --- e-l; 0 ~ = 1.
A treatment on these lines is generalized ex-
tremely readily to more complex conditions with
e.g., temperature-dependent heat-losses such as
oc~,ur when h is not constant, or non-Arrhenius rates
such as T m exp ( - E / R T ) . If we do not simplify the
4 HOTrEL PLENARY LECTURE
treatment by neglecting eO compared with unity
(or T - Ta compared with T~) we find:
e(Se)cr = 1 + ~; O c t = 1 + 2~.
Heterogeneous oxidation in practice is concerned
with two main circumstances characterized by two
different examples: the combustion of carbon (coal
or coke) and the catalytic oxidation o[ ammonia on
e.g., a platinum surface. In one, the surface is
eroded; in the other, it is not. Originally, coke
combustion was recognized as occurring either very
slowly at low or moderate temperatures or very
rapidly at elevated temperatures. Emphasis was
formerly placed on a mysteriously different chem-
istry in the two r6gimes. It was an early triumph
of thermal theory to unify the two phenomena, to
predict conditions for extinctions as well as igni-
tions and to reveal the connexions between the two
examples above and their relationship to the ho-
mogeneous system.
The treatment of heterogeneous systems2 begins
in the same way except that two separate ingredi-
ents of stationary states are now needed--the ar-
rival of matter at the reacting surface, and the
transport of heat to the (unreacted) gas. The form
of ~ is altered to a driving force divided by the
sum of two resistances, one chemical and the other
diffusive:
~t = QVco/(k -~ + 13-~).
The separate mass and heat-balance equations:
k c V = ~S(co - c),
QkcV = hs(T - To),
allow us to connect c and T in the stationary state:
c h S ( T - To)
co 13Qco
and to reach the dimensionless equation:
Oako Oe - ~
(Vk) . . . . ergy are the starting point. For matter:
13 1 - o/o~
The quotient O / O d measures how hot the surface
is in comparison with how hot it could become ff
reaction were so fast that chemical resistance was
effectively nil. Amongst the results that emerge are
the existence of a range of conditions under which
multiple solutions are possible, bounded by points
of ignition and extinction. The requirements of ig-
nition are almost the same as before:
Oig ~ 1 + O21 ~ 1, frequently.
The new feature, extinction, occurs when
Td-T-~R~a/E or O ~ t ~ O d - 1
but the basis for dimensionless temperature for ex-
tinction is Ta and Oa relates to how far below Ta
is the extinction.
This marvellously simple analysis was given in
1938 by Frank-Kamenetskii. Not long afterwards,
his pupil N. Ya. Buben extended it to reactions of
order different from unity (with rate = kcm). Their
treatment (which was less than elegant) was im-
proved by Thomas & Bowes (1961) who provide an
economical exposition of the elementary case. Real
systems often involve distributed temperatures and
it is to chemical engineers that we look today for
these more general enquiries: Aris's monumental
monograph33 is the natural starting point.
4. Extensions to Closed and Open Systems
When homogeneous and heterogeneous systems
were considered in the early days, reactant con-
sumption was ignored and the problems of real sys-
tems were approximated by idealized ones in which
stationary states were possible. In open systems,
however, truly stationary states are possible. The
simplest of open systems is afforded by a well-stirred
reactor into which fresh reactants flow continuously
and out of which issues a partially reacted mixture
of products and reactants. Exothermic reaction leads
to self-heating and under certain circumstances to
multistability. In the context of combustion, Long-
well's unforgettable ~ontribution (1952 paper) is the
western landmark. (There was an earlier pioneer,
Liljenroth, who seems never to have been noticed
much even during his own lifetime.) In the USSR,
Zel'dovich and Zysin (1941) were responsible for
opening up the field: they too worked with a view
to combustion problems. After 1955, chemical en-
gineers ploughed these fields.
The treatment is so instructive that it deserves
to be sketched in outline. As with heterogeneous
systems, conservation equations for matter and en-
V(dc/dt) = uco - uc - Vkc m
or equivalently, for a first-order reaction:
dh l-h h exp{ E ET}
dt tres tch ~o "
Most of the symbols have obvious meanings; h is
the fractional reactant-concentration, c/co; T and To
 CHEMISTRY AND COMBUSTION 5

are the exit and entry temperatures. For energy: or

Oe -0
V~cp(dT/dt) = QVkc m - hS(T - Tw) (Ze)-~ - - = g(O)
(1 - O / B )
- uacpff - To).
These expressions have the same structure as those
self- exothermic losses heated for heterogeneous reaction. They can yield either
heating reaction v/a wall outflow one or three solutions (not more) and thus predict
either monotonic responses to changes in conditions
or equivalently: (this is all that happens when AT~a < 4RTOTad/E)
or the possibility of 'ignitions' and 'extinctions' and
a'r hysteresis when ATaU exceeds this limit. In dimen-
QC~ (k~176 ~o exp sionless terms, the approximate condition for mul-
dt cFcp tiplicity is B > 4. For larger values of B, the ap-
T - Tw T - To proximate locations of ignitions and extinctions are
at O_ = 1 and O+ = B - 1;
(Vcr%/hS) (V /u)
T,g - To = RT~o/E; Tad -- Text = RT~aa/E.
If we use dimensionless temperatures and charac-
teristic times (V/u) = t,~ and 1/k o = tch and (Vacp/ In the foregoing, the term written (Ze) denotes
hS) = tN then: the group (BtreJtch), or more explicitly:

1 dO km (Ze)---~ Qco ~ [A E
-~ =--exp - +

n dt tch

This last expression has been simplified in two ways.
First it assumes that inflow temperatures and wall
temperatures are equal: To = Tw. This restriction
can readily be lifted without significant complica-
tion when non-adiabatic operation is considered. It
also ignores any dependence of the various thermo-
physical properties (such as specific heat, density
and heat-transfer coefficient) on either the extent of
reaction or on the temperature. These matters may
need consideration in a computational study but they
are subordinate to the main thrust.
Adiabatic Operation and the Zel'dovich Numbers:
Adiabatic operation is best achieved in practice
by externally controlling the wall temperature so as
always to equal that of the contents so that Tw =
T. (This has the same effect in the equations as let-
ting h = O or tN = ~.) In these circumstances,
concentrations and temperatures are linked, and the
stationary-state relationship:
AT 0 temperatures and concentrations;
. . . . l-k,
~T~a B
is established. Our system is described by a single
variable, so oscillations are impossible. The govern-
ing equation for stationary states is:
Non-adiabatic Systems:
The reason for commemorating Zel'dovich's name
in this context is nowhere more obvious than in the
clear way in which he, with Zysin, set up and ana-
lyzed the governing equations 3 and evaluated their
consequences for non-adiabatic systems. They in-
vestigated only the patterns of stationary states (and
their 1941 work was rather strikingly rediscovered
from 1955 onwards), but what they uncovered was
very novel. Their work is especially relevant to what
is called (by real engineers) 'glass bulb chemistry'
where work in fragile apparatus on a laboratory scale
permits only slow flow-rates and guarantees a major
role to heat-losses. In terms of conservation equa-
tions, that for matter is identical and that for en-
ergy preserves the heat-loss term. (The quotient
(ucr%/hS) or (tN/tres) or (kftN) is the sole extra term:
we call this quotient K and its reciprocal [3). This
similarity of structure means that results can be
written down at sight for the occurrence and lo-
cation of ignition, extinction, hysteresis and multi-
plicity and for the link between stationary-state
First,
KB B
Oss= B * ( 1 - k , 8 ) where B* = - -
l+K I+~'
Next,

(Ze) = (Taa To)(T - To) B*tres Oe -~

(RT~o/E)(Taa- T) exp = Z(T) tch 1 -- O/B* '
6 HOTTEL PLENARY LECTURE
For multiplicity, B* > 4. At points of ignition and
extinction (so long as B* is not too close to 4):
1 temperature cases. This was copied by several au-
| Oext=B*-l.
Although the similarity of structure means that there
is once again only a choice between one and three
steady states at any particular flowrate, the depen-
dence of B* on flowrate means that for any given
excess temperature there are either two flowrates
or none. This in turn generates a greater variety of
responses to changing flow speeds. In particular as
flow-rates are increased from zero, stationary-state
temperatures grow and ignitions can be caused by
making the flow faster. The temperature-jumps that
set in increase quickly to values near the maximum
possible: if that maximum is low, the jumps can be
quite small. The notable work4 of Griffiths & Hasko
on ethane oxidation at low flow rates is a case in
point. In rich mixtures, oxygen consumption jumps
from tiny values to completion, yet the excess tem-
perature after ignition reaches only 60 K.
5. Temperature-time Histories
(for Systems with Negligible Reactant
Consumption or Strong Exothermieity)
The earliest treatment of homogeneous exother-
mic reactions ignored reactant consumption and
concentrated on stationary states. It is readily pos-
sible under these assumptions to deduce the evo-
lution of temperature in time. When approxima-
tions to the Arrhenius law are adequate, sub-critical
and supercritical evolutions are very sharply dis-
tinct. The former lead to finite temperature ex-
cesses (after infinite times); the latter to infinite
temperatures after finite times (induction periods).
The 'critical trajectory' occurs as the limit of the
subcritical family, and dimensionless temperature
evolves according to laws like:
(t/tad)
13 = - when t~ ~-qlcr ~'-
1 + (t/tad)
For marginally supercritical circumstances when
only slightly exceeds ~Jcr, there is a characteristic
dependence of induction period on the degree of
supercriticality, (t~/t~cr) - 1. It is described by an
equation of the form:
t ~ (supercriticality)-1/2
so it lengthens indefinitely as the degree of super-
criticality diminishes. This form of result was dis-
covered by Frank-Kamenetskii for uniform-temper-
ature systems. (Rather distractingly, he set a
precedent of using the same symbol for the critical
Semenov parameter (which we have called 6) as the
one (~) which he had introduced for distributed-
thors and has generated confusion to newcomers for
many years.) In fact the same functional form does
apply to distributed systems, but this was not es-
tablished until long afterwards in notable studies 19
by Boddington and Kordylewski.
The Strongly Exothermic Limit:
To study this, matched asymptotic expansions of-
fer the systematic route. The techniques were
introduced5'7 by Kassoy and Lifian and by Lacey.
They systematize the earlier work2'6 by Thomas and
Frank-Kamenetskii, and open the way to discov-
ering higher-order terms. (Note the misleading use
of 8, however.)
The conservation equations under examination are:
(~rc~V)dT/dt = VQcmA exp(-E/RT) - hS(T - Ta);
T = Ta at t = 0
and
-dc/dt = cmA exp(-E/RT)
c = Co at t= 0
or in dimensionless terms:
r = Cg(w)e ~ - O
B(dw / dr) = g(w)e ~
where the dimensionless time-scale satisfies r = t~
tad. When reactant consumption is neglected and
the concept of criticality is clear, the appropriate
"small quantity" for asymptotic expansions is k, de-
fined by ~ = qJo(1 + h z) and the appropriate time-
scale z = kr is used. When reactant consumption
is taken into account, z = Ixr is used, where 3
= 1/B for a first-order reaction. This scaling is the
1 most economical and allows an efficient solution to
e varied circumstances (generalized temperature-de-
pendences of k; generalized isothermal reaction ki-
netics; ramped ambient temperatures, etc.). The
work of Boddington (Leeds), Kordylewski (Wro-
claw) and Feng (Beijing) has made innovative con-
tributions in this field. All results require an in-
crease in qJc~from its 'primitive' or zero-order value
qJo = e -1 by amounts proportional to (m/B) 2/3.
Roughly speaking, ~cr - ~o = 3(m/B) ~/3. This re-
quirement that (m/B) 1/3 be a small quantity is not
a trivial restriction. For a first-order reaction, we
would expect perceptible deviations from leading-
order correcting terms even if B was as large as 27
= 33; B is unlikely to exceed 100, and 100 1/3 ~ 4.6;
 CHEMISTRY AND COMBUSTION
for a higher-order reaction, requirements on B are
more severe.
Temperature programming (see also w is com-
monly used to study exothermic substances from the
point of view of safety. As ambient temperature is
raised, so the balance between acceleratory effects
of self-heating and deceleratory effects of reactant
consumption is tilted, and a particular heating-rate
may be found at which the latter pair cancel one
another out and produce behaviour identical (to
within first-order effects) with the primitive case of
zero reactant-consumption. These effects were sys-
tematized by Kordylewski (1983) and extended by
Kay. Values for ~ c r below or in excess of the clas-
sical value are possible and the mathematical tech-
niques used are applicable to safety procedures such
as heat-removal by surface cooling of a reactive mass.
6. Thermal Runway in Closed Systems:
Criticality Versus Sensitivity
The ideas explored in the previous section clearly
link "criticality" to a jump from one stationary state
to another. Stationary states just cease to exist at
the point of such a jump and we deal with bifur-
cations. The treatment is appropriate for the truly
stationary states of an open system, like a contin-
uously-fed, well-stirred reactor (cstr). For a closed
system, it matches only the simplified model that
ignores reactant consumption and not the system
itself, though it can be used as a point of departure
for asymptotic analysis.
In a closed system, temperature evolution and
reactant consumption both proceed with differences
only in degree, not in kind. All temperature his-
tories (for a single exothermic reaction) have the
same gross form with a single maximum value. All
histories of extent-of-reaction begin at zero and rise
to unity with only varying degrees of steepness. How
are we to hope to replace the clear-cut contrasts of
criticality by some useful operational criterion? This
question turns out to receive a great deal of help
from numerical computation, but an analytical ap-
proach together with an appropriate choice of re-
duced variables help us to pose the questions ef-
fectively. It is clear that a different concept from
"criticality' needs to be used to express ideas more
satisfactorily. The idea of sensitiveness (or insensi-
tiveness) to initial conditions is what is needed, and
the search is for circumstances where small changes
in initial conditions make the most substantial dif-
ferences in the unfolding of events. We shall see
that the principal parameters in a general treatment
are (i) the Semenov number (which measures initial
rate of heat-release)
(Se) = VQkcom/hS(RTa2/E),
7
(ii) the isothermal reaction order m (which can be
generalised to more complicated kinetics), and (iii)
the Zerdovich number B which measures the sys-
tem's capacity for self-heating
B = Oad ~- (EQco/Cxc~RTaZ).
Analytical solutions are out of reach and results have
to be expressed as dependences of events on the
values prescribed for (Se), m and B; when B is
large, the last two operate together as the quotient
(B/m). In what follows we may write O~r simply to
aid comparison with the literature, but the preced-
ing remarks about the idea of sensitivity as superior
here to the concept of criticality must be borne in
mind.
In the past the following considerations have been
influential: (1) For the primitive case with c = Co,
RTa2/E is the upper bound on safe temperature-
rises, so in a deceleratory system temperature rises
greater than this should exist that belong to the
quiescent family. (2) Temperatures that evolve
without any surge in gradient must also belong to
quiescent systems: points of inflection in tempera-
ture-time curves are essential symptoms of runa-
way.
In the 1960s Adler and Enig s suggested that in-
flexions in the temperature-reactedness curves were
sufficient indicators of runaway and on this basis
produced a lively analysis with some appealing an-
alytical results for extreme cases (zero order; infi-
nite B). For a reaction of order m, inflexion points
set in at Oinfl = 1 + X/m; for an adiabatic system,
a minimal value for B is (1 + ~/m) 2. Their results
were won for the exponential approximation but ex-
tended by P. R. Lee to the Arrhenius form. De-
spite apparent inconsistencies in their develop-
ment, these neat formulae will continue to keep this
work in mind.
A useful present-day study9 of deceleratory re-
actions in closed systems is that due to Kordylewski
and Scott. They begin by adopting the equations:
d c / d t = - k c '~ with k ~ exp ( - E / R T )
cvcrV(dT/dt) = VQkc "n - hS~F - Ta)
but generalize them to arbitrary temperature-de-
pendence of k and arbitrary concentration-depen-
dence of isothermal rate. Then treatment is centred
on the maximum value reached by the excess tem-
perature, which they call O*, and how sensitively
O* responds to initial conditions (i.e. as expressed
by the Semenov number, qJ) for different values of
exothermicity (as measured by B). For any large
value of B there is a region of great sensitiveness
so that d| takes a sharp maximum value (where
d20*/dt~ 2 = 0). When B = 20, and m = 1, this
occurs at d~ = 0.545. The same features persist for
 HOTrEL PLENARY LECTURE
any large value of B, but as B gets smaller, so the
responsiveness of | to 0 becomes less and less
pronounced and the curve O versus 0 loses its in-
flexion between B = 5 and B = 3.
Asymptotic analyses by Lacey7 help to link var-
ious approaches together and it is fair to say that
the Scott-Kordylewski approach is as attractive as
any and promises more rewards from further study--
especially of times taken to reach maximum tem-
perature. This is an obviously underdeveloped area.
There is broad agreement between their results and
Tyler and Wesley's classical numerical computations 1~
at the one extreme, and the Thomas-Kassoy-Lifian
analyses 5"~ at the large B range.
The other approach also deserves attention. B.
F. Gray and Sherrington, who were amongst the
earliest (1975) to stress the need n for replacing
'criticality' by 'sensitivity,' concentrated on how
whole or start-to-finish temperature-time trajecto-
ries were affected by small changes e.g. in ambient
temperature. So far their analysis has afforded rather
conservative estimates of the range of insensitive
('suberitical') behaviour, with seemingly no stabilis-
ing effect being conferred by reactant consumption.
We are here discussing sensitivity to initial con-
ditions and we have chosen to combine the various
initial conditions (V, S, k, Co etc.) into a single pa-
rameter. The rather uninformative description
'parametric sensitivity" (Bilous & Amundson, 1955)1~
could be said to spring from this, though its actual
origin in a study of tubular reactors was different.
We consider tubular reactors next.
7. Tubular Reactors
Continuous-flow processes offer certain econom-
ies in chemical industry and tubular reactors exploit
them. The reactive fluid (gas or liquid) flows down
a tube and emerges partially or completely con-
verted. Though laminar flow down a round tube
produces a parabolic distribution of velocities, the
simplest model imagines plug flow. In this model,
radial dispersion is 100% effective and concentra-
tions and temperatures are beth uniform across the
tube diameter. Heat is lost only to the walls. Axial
mixing is neglected. In such circumstances, the
predicted growth and decay of concentrations or
temperatures along the tube's length exactly par-
allel their evolutions in time in a dosed, stirred
vessel (stirred to preserve uniform temperature and
concentration fields). Although apparently ex-
tremely oversimplified, the model is the natural
prototype for real systems, much of whose behav-
iour is often quite well represented by it.
Accordingly the chemical reactor engineering lit-
erature throws light on batch reactors as well as tu-
bular ones and parallel studies have been made. For
exothermie systems, Bilous & Amundson 12 open
modern studies, although K. B. Wilson (1946) came
much earlier and recognized the importance of the
group (RTa2/E) as a watershed between moderate
and stable spatial temperature profiles and the on-
set of sensitive, spiky ones. Because of this differ-
ent background, Barkelew's reactor engineering
studies were unknown to Adler & Enig, who
rediscovereds his results, and early thermal-explo-
sion work was not much exploited by Amundson
though he refers to it.
When axial dispersion is taken into account, the
strong positive feedback so made possible generates
multistability (in space). Jumps from one mode (low
conversion, low temperature) to another (high con-
version, higl~ temperature) can be expected. Find-
ing these jumps of ignition and extinction is equally
a clear indication of feedback. They are expected
9 . . 13
to be subject to the same features of long passage-
times as were met before:
time ~ (degree of supercriticality) -1/2.
When radial mixing is not instantaneous, the jumps
that occur are now from one stationary-state, radial
profile to another. Work in this field seems to be
sparse but unifying principles are clearly available.
We have in fact made a link between thermal ex-
plosion theory and the 1-dimensional laminar flame,
which long seemed unconnected.
8. Ramping Ambient Temperature
Differential Scanning Calorimetry:
Early studies of potentially hazardous materials
proceeded by measuring their response to being
placed in surroundings at a sequence of discrete,
elevated temperatures. Some very humble mate-
rials show very reproducible behaviour, and classic
work was done by Walkerzr on the oxidation of fibres
like wool and powders like wood flour. Such se-
quential methods take a long time, and alternative
procedures shortening the search for the sensitive
region and for recognizable ignition temperatures
are therefore desirable. Dynamic methods offer
suitable alternatives. In them a sample of known
size and shape is subjected to a steadily increasing
or 'ramped' ambient temperature. Many automated
and computer-controlled instruments are available
to provide this. One family of procedures operates
by measuring rates of supply of heat required to
keep the temperature of an active sample increas-
ing linearly. This is a form of scanning calorimetry;
when exothermic chemical reaction sets in, less heat
is needed, and so on. Many systematic errors can
be avoided by comparing the energy requirements
of the active sample with those of an inert standard
of the same size and mass; this is the basis of dif-
 CHEMISTRY AND COMBUSTION
ferential scanning calorimetry. By such techniques,
values for thermophysical properties of unreactive
materials can be obtained automatically, and com-
mercial devices can be so programmed as to pro-
vide values for exothermicities of chemical as well
as physical changes. Reaction rates and Arrhenius
parameters can also be extracted when reactions are
not too complex. The traditional methods of anal-
ysis concern themselves with locating temperatures
Tm at which the instrumental signal is a maximum.
This signal measures reaction rate (or heat-evolu-
tion rate) and its location (T,,) depends on the value
of the heating rate (W). There are several methods
of treatment, and the Kissinger relation (1956) is
typical in showing the route to activation energies
from a graph of In W or In(W/Tin 2) plotted against
1/RTm:
In (W/T,, 2) = const. - (E/RTm)
Thermal explosion theory actually offers a very
reliable route to analyses of this sort--superior to
most of the traditional analyses, which are often ac-
tually based on static treatments. It must be clearly
recognized that we are dealing here with systems
that are very far from constant reactant composi-
tion: a slowly increasing ambient temperature is
inevitably accompanied by very extensive decom-
position long before the end of each 'run.
The successful procedure leads to 'critical' or wa-
tershed values for Semenov's parameter t~ distin-
guishing quiescence from runaway. Below Oct, tem-
perature-rate trajectories pass through maxima less
than unity and then decay (temperature excesses
are expressed on a reduced scale like O). Above
Ocr, these trajectories are vertical when the re-
duced temperature-excesses reach unity and tem-
peratures then run away. The value of ~c~ depends
on the kinetic laws for isothermal reaction and must
be found by computation. However, only a single
computation is needed for a particular rate-law: for
a first order reaction, ~Jcr = 1.799.
The analysis leading to this requires a reference
temperature although ambient and sample temper-
atures are changing:
da/dt=-ka with k = A e x p ( - E / R T )
T=Ts+Wt
The relationship that T,-s has to satisfy is:
(RT2rf/EA) exp (E/RTrf) = W
Clearly this must be solved by iteration. An up-to-
date account of this for small samples of uniform
temperature is given by Boddington and Kay (1989)
and they also make remarks on the connexions with
more limited treatments.
Ramping Ambient Temperature in Systems on the
Verge of Runaway (Including Systems With
Distributed Temperatures):
Theoretical studies of temperature-evolution were
first made in systems with uniform temperature in
which reactant consumption was ignored. A sys-
tematic approach5 including consumption was pro-
vided by asymptotic analysis though it is only fair
to say that the form of the leading-order corrections
had been anticipated by the early and perhaps less
9 ..2
mechanmal treatments of Frank-Kamenetskil and
Thomas. 6 Attacks on systems with distributed tem-
peratures began later with the work of Boddington
(1983, 1984) and that work was extended to varying
ambient temperatures in association with Kordy-
lewsky. The end result was a powerful treatment
(1986) that could cope with very general depen-
dences of reaction rate upon temperature, on ex-
tent of reaction and on whether material diffuses or
not. The asymptotic analysis requires a suitable
"small quantity" and the neatest analysis uses a pa-
rameter I~ related to the Zel'dovich number: it is
~ B-l/3.
The development is fairly straightforward though
it involves a whole alphabet of symbols. The time-
dependent amplitude (I)(z) of the temperature
obeys an equation like:
dOP/dz = a~ ~ + bz + c
dp--->oo as z--->O+
The case when the stabilizing tendencies of reac-
tant consumption are just balanced by self-heating
is represented by b = 0. This is the primitive "zero-
order" case. When external heating is insufficient
to compensate for reactant depletion, subcritical and
supercritical behaviour of the first-order perturba-
tion are also clearly separated. When external heat-
ing more than compensates for consumption, high
temperatures are inevitable and all initial condi-
tions lead to runaway.
It should be noted that these results are most
accurate for vey exothermic reactions near to the
point of 'unaided' ignition. The asymptotic treat-
ments spotlight a narrower band of conditions than
that of the general realm of scanning calorimetry
dealt with in the previous section.
9. Parallel Reactions
Although at the outset we instanced the oxida-
tions of carbon monoxide and hydrogen as substan-
tial reaction networks, they are far surpassed by hy-
drocarbon combustion, of which they also form
important elements. All these however are clearly
"complex systems" and for thermal explosion theory
and chain-branching to be applied to them in a
10 HO'I'TEL PLENARY LECTURE
simple fashion, great simplifications must be made.
Before them come the properties of more elemen-
tary prototypes--concurrent and consecutive reac-
tions.
When materials like cellulose or wood are oxi-
dized, or even merely heated in the absence of air,
and logical interpretations demanded, there is a
natural tendency to abandon simple principles and
to expect irreducible complexity. This fear is not
always justified and rather broad chemical princi-
ples based on the behaviour of simpler compounds
can help a great deal. Scientists from several coun-
tries including New Zealand and the Soviet Union
have considered14"-16 the possibilities of dealing with
chemically independent, concurrent reactions that
are linked only by their several responses to heat
evolved communally and hence by their exposure
to a common temperature. When all the reactions
are exothermic, a communal activation energy based
upon the responsiveness of overall power output
(Watts kg-1)
E = RT2 d{ln (power output)}/dT
permits the assignment of a common dimensionless
temperature excess. When the kinetic and thermal
features of the individual reactions are known, it is
now easily possible to go forwards with predictions.
Unfortunately, the route backwards is not easily
mapped.
More interesting than these elementary direc-
tions are those to be followed15"16 when exothermic
and endothermic reactions occur in parallel. If heat-
absorbing reactions are at first slower but more
sensitive to temperature, thermal runaway is curbed
and may even be prevented. Although attention has
been mainly directed at systems without any reac-
tant depletion, the results are full of interest, as
they may provide primitive models for exothermic
oxidations which simultaneously drive off moisture
(endothermic). They may turn out to be primitive
because the manner of the transport of moisture may
play more of a key role in real systems.
Some dry materials absorb moisture with heat
evolution, and another complication is now possible
because ff the dry material is already on the verge
of thermal instability, additional heat-release due to
its being wetted may take it over the brink and into
instability. Once again simple circumstances can
generate very complex responses and these models
need to be evaluated. Rewarding beginnings16 have
been made by B. F. Gray, so far in terms of a uni-
form composition model; this is another of those
fields in which uniformity is a rather strong as-
sumption, and both the transport of water vapour
and the reversibility of the exothermic wetting are
likely to have significant influences.
10. Distributed Temperatures
Theory:
The most influential development in thermal ex-
plosion theory was its early extension2 to systems
in which reactant-temperature varied from point to
point. This assessment is based partly upon the im-
portance of the problem itselfis and partly because
it was in this context that the dimensionless tem-
perature 0 and the exponential approximation to
the Arrhenius equation were first introduced:
exp [(E/RTo) - (E/RT)]
= exp [O/(1 + r ~ exp O.
These steps were taken in order that analytical
solutions to describe how temperature depended on
position in systems of simple geometry when heat-
flow was purely conductive might be derived. Such
solutions were found analytically for the infinite slab
and later for the infinite cylinder, whilst tables ex-
isted for an allied function that subsequently en-
abled an accurate account to be given for the sphere.
All these important results relate to circumstances
in which reaction-rates depended only on temper-
ature: reactant-consumption is neglected.
The problems are described by Fourier equa-
tions of the type:
(r%(dT/dt) = V(KVT) + Qkcmo
For the 'class A' geometries (infinite slab, infinite
cylinder, sphere) when K is constant, stationary
states satisfy
d20/dp 2 + (j/p)dO/dp + 8e ~ = O,
where j = 0, 1, 2 for slab, cylinder and spherical
geometries. Temperature is a function of one vari-,
able (p = r/a) the reduced radial distance, and
Frank-Kamenetskii's parameter 8 is given by
~3 -= [a2Qkacom/K(RTa2//E)]
The boundary conditions chosen require zero tem-
perature-excess at the surface. This is an equation
which, broadly speaking, has no solution when 8 is
too large (8 > 80, say), and has pairs of solutions
| stable and unstable, for 8 < 8o. The merging
of these solution types at 8 = 80 corresponds to a
"critical profile," and 8 = 8o defines criticality:
8o Oo
Slab 0.878 1.199
Cylinder 2 1.369
Sphere 3.322 1.607
There follows a period of study of different ge-
ometries, initiated2 by Frank Kamenetskii himself.
 CHEMISTRY AND COMBUSTION
The 'infinite slab' of thickness 2a is equivalent to a
finite slab also of thickness 2a with its four edges
perfectly insulated; the infinite cylinder is equiva-
lent to a finite cylinder with insulated ends. Taking
away the insulation affects the condition for criti-
cality. Various attacks have been made on the
problem--ab initio numerical computation being
common from very early days. More satisfactory are
intuitive or reasoned analogies such as the idea of
the equivalent sphere. New Zealand workers were
prominent in these developments, and a survey of
their inter-relationships and the existence of rigor-
ous (and sometimes very useful) bounds was later
given 17 by Boddinigton in 1971.
The next elaborations were provided around 1958
by Thomas and Merzhanov who generalized the so-
lutions to systems with a temperature-step at the
surface, corresponding to heat flow affected by sur-
face-resistance as well as by internal responses to
thermal conductivity. Many of these treatments is
are not exact, but serviceable general approxima-
tions are always enlightening and satisfy most needs
so far as practical applications are concerned.
After this there is a period of study addressed to
the problem of abnormally small activation ener-
gies. It had been noted by Semenov in 1928 that
since critical temperatures and ambient tempera-
tures were related by the quadratic
RTZcr/E = Tcr - Ta
any activation energy E < 4RTa offered no solu-
tion. This result (expressible as err = R T J E < 0.25)
may be expected to have analogies in the distrib-
uted-temperature case. The problem has intrigued
many mathematicians and many papers have con-
cerned themselves with determining ~-tr. Some of
these studies were incorrectly made and most were
undertaken as intellectual exercises rather than for
practical utility. But a watershed in the subject is
marked by Kordylewskii's 1978 study 19 which of-
fered a generalized attack on the loss of criticality
in distributed-temperature systems (usually when
the quotient RT~/E approaches values near 1/4).
It must be emphasized that many of these are stud-
ies of the pathology of thermal runaway: at such
temperatures the Arrhenius expression has lost much
or all of its distinctive curvature, and there is no
point in persisting with the exponential approxi-
mation. Moreover, as these temperatures are often
around 104K and chemically inaccessible, their im-
portance for serious practical study is slight.
Distributed Temperature Systems: Unifying
Treatments:
The manner in which critical central tempera-
tures, etc. vary from one geometry to another often
11
seems perplexing to those with a purely chemical
background. An attempt to show the underlying
similarities was made by Boddington and Scott us-
ing a treatment (a second-order reversion) which
gives analytical results that are not exact but are
remarkably precise in terms of a single parameter
of the system. This treatment is closely linked to
questions of the efficacy of catalyst pellets used in
exothermic reactions and it affords instant extension
to endothermic systems. The type of result can best
be seen in tabular form.
The results are 'exact' for the infinite cylinder--
i.e., as exact as is the Frank-Kamenetskii approxi-
mations (exp O) which usually implies errors only
of order e = RTa/E "-~ 1/4o; they are close to 'ex-
actness' for the sphere and the slab.
Distributed Temperatures: Experimental
Verification:
Despite the universal nature of the acceleratory
mechanism, effective searches for excess tempera-
tures that vary from point to point were long de-
layed. Some early attempts using mercury-in-glass
thermometers were unlikely to succeed but when
fine-wire resistance thermometers and thermocou-
pies were used the difficulty of a too-sluggish re-
sponse to temperature was overcome. The bench-
mark studies. 21 are still those of Gray and Fine on
the thermal explosion of gaseous diethyl peroxide,
but pioneering workzz by Vanp6e (Thesis 1956) was
done considerably earlier and richly deserves to be
celebrated.
Briefly, the peroxide studies established that ig-
nition was always preceded by self-heating, that
temperature excesses were greatest at the vessel
centre and there was little or no detectable excess
at the walls. Stable decomposition ceased to be
possible at a temperature-excess around 20~ C (close
to 1.61 RT2a/E). The level of precision attainable
is more severely tested by comparing 'experimen-
tar and 'theoretical" values for ~cr: these are 4.0 and
3.32. The shape of the temperature-position curve
was roughly parabolic but thought to be more spiky.
Seen in the perspective of twenty years elapsed,
TABLE I
Unifying relationships for slab, cylinder and
sphere (j = 0, 1, 2 respectively)
At criticality
Dimensionless rate 2(j + l)(j + 3)/(j + 7)
Central temperature | 2 In [(j + 7)/4]
Boundary gradient 2
Effectiveness factor (j + 7)/(j + 3)
12 HOTrEL PLENARY LECTURE
TABLE II
Unifying treatments for different geometries:
parametric results for the infinite cylinder
(N.B. x > O, exothermic; x < 0, endothermic)
8x(1 + x)2
Oo ln(1 + x)z
F 4x/(1 + x)
(1 + x)
AE/E x/(1 - x)
8/fi~ 4e-~176 - e -~176
O(p) 2 In[(1 + x)/(1 + xp2)]
some of our earlier certainties sl seem now at the
edge of confidence limits but the totality of the evi-
dence is still convincing.
Two or three other systems may be mentioned
here. Ditertiary butyl peroxide has been studied by
Griffiths and his co-workers. It is a very satisfactory
illustration of a 'dark' thermal explosion, for there
is none of the visible light emission associated with
the oxidation of organic compounds. By working in
stirred and unstirred systems, Grifliths has been able
to show experimentally the relationship between
O~t = 1 and Oo~t = 1.61 (uniform or averaged
critical temperature excess and central temperature
excess) for an unstirred system. This system can also
be thermo-chemically tuned. Dilution or partial re-
action reduces the value of B (= O~d), obviously
leaving E the same; added oxygen increases B (rather
less obviously also leaving E the same). This is a
matter of great usefulness and importance in testing
consistency of prediction and observation, e.g. in
flow-through systems.
An exothermic oxidation that self-ignites ther-
mally is that of hydrazine. This, too, seems notto
involve chain-branching reactions but because its
effective activation energy is rather low, very large
sub-critical temperature-excesses occur--up to 70
K is common.
Finally, if self-heating shows up in exothermic
changes, self-cooling should be seen in endo-
thermic ones. Here there will be negative feedback
and no discontinuous loss of stability is to be ex-
pected. The system chosen to test this is the en-
dothermic pyrolysis of nitrosyl chloride, studied in
a cylindrical vessel
NOC! --* NO + i/2C12 AH/kJmo1-1 = 37
Self-cooling is always observed; temperatures are
lowest along the axis of the vessel, typically falling
5 K below that (Ta) of the walls. The quasi-station-
ary temperature position profiles are in reasonable
(but not perfect) accord with expectations:
T ( p ) - Ta f 1+ C ]
(RTa2/E----'
~ -- - 2 In /L1 + Cp2J~
where C is predicted to be about -0.18, and found
to lie between -0.08 and -0.22.
The Corrections of Reaction-kinetic Measurements
for Unexpected Self-heating:
Interest has always been strong in certain fami-
lies of 'unimolecular' decompositions because of the
manner in which structure and thermochemistry are
reflected in activation energies and preexponential
factors. For example there are the differing stabil-
ities towards ring-opening displayed by rings of dif-
ferent sizes or by rings of the same size containing
different hetero-atoms. Convenient studies require
the elimination of surface reactions--easier in large
systems than in small, and the desire for a wide
temperature range may be more conveniently met
by measuring short half-lives at elevated tempera-
tures than by accepting very prolonged times of
study at low temperatures. In this case self-heating
is present, and may be sufficient to matter when
high precision is the goal.
The problem was recognized by reliable experi-
mentalists who stressed the need to measure true
reactant temperatures and not merely those of a
vessel's surroundings. An early analysis offered by
Benson illustrates the difficulties felt even by the
ablest of kineticists for it relies on the fortuitous
cancellation of errors. The problem yields imme-
diately to thermal-explosion theory, producing re-
sults of simplicity and expressiveness. Under uni-
form temperature conditions, errors in apparent
activation energy are very strongly related to tem-
perature excess:
AE O T -Ta
E l-O; O=(R,I~a/E)
An iterative procedure is necessary in principle, but
one iteration may suffice.
When thermal conductivity governs heat-trans-
port, and temperatures depend upon position, more
complex but related expressions can be used. These
are approximations but of great utility. For the sim-
ple geometries if the centre temperature Oo is
known then a single parameter x links it to many
practical expressions
Oo = 2In(1 + x)
When ~ is known x is the solution of the expression
= 8x/(1 + x)2. This formula refers to the infinite
cylinder. For the infinite slab and the sphere re-
spectively
4x(1 - x/f) 12x(1 + x/10)
=
and
(1 + x) ~ (1 + x) 2
 CHEMISTRY AND COMBUSTION
The theory can also be made to disclose where
to place a thermometer so as to record a represen-
tative "average" temperature. For infinite slabs, in-
finite cylinders and the sphere, these are at dis-
tances r = 2ao/3, ao/2 or 2ao/5 from a mid-plane,
an axis or a centre respectively. The problem is re-
lated to that of evaluating the non-isothermal per-
formance of catalyst beds or particles.
11. Hot Assembly; Hot Spots;
Thermal Explosion of the Second Kind
Many real problems of thermal runaway arise
when materials capable of exothermic reaction are
assembled at a temperature T well above ambient.
The classical, stationary-state equation for a uni-
form-temperature system
9
Bt._._~N= (Se)= (RT?TE)exp R-T
tch a Ambient temperatures being known, the mis-
admits one or three solutions. When there are three,
the lowest is stable, the middle one is unstable and
the uppermost solution is at an inaccessibly high
value. The equivalent formulation
(Se) = O exp [ - O / ( 1 + eO)] -~ Oe -~
with zero or two solutions to its approximate form,
highlightS the relevance of the two, physically ac-
cessible, stationary branches, the lower being sta-
ble (T~) and the upper branch unstable (Tu). If the
temperature T lies between these two, the system
moves toward the stable state: if T exceeds Tu,
thermal runaway sets in immediately.
Bowes names these circumstances, "thermal ex-
plosions of the second kind." The branch Tu sep-
arates quiescent from explosive states and is a wa-
tershed. In this approximation, the unstable branch
runs from the classical critical point C at Tcr to in-
definitely large values.
The significance of thermal explosions of the sec-
ond kind has never been lost on practical technol-
ogists working, for example, in the chemical indus-
try but does not seem to have been the centre of
attention that the critical point C has been. Of course
the question "how far does a system need to self-
heat before it runs away" with the dramatic answer
"often only by RTa~/E--whieh may be no more than
ten or twenty degrees" deserves to catch the at-
tention, but so does the parallel question about how
safe it is to stack reactive material that is already
preheated. Examples of such circumstances are the
storage of hospital sheets after hot laundering and
pressing, the manufacture and stacking of materials
like fibre-board and the piling of spent, sugar-cane
residues (bagasse) in huge amounts.
13
Hot assembly
: i
o
o~
ee
r~r
Ambient temperature
FIG. 1.
leadingly large safe stacking-sizes for piles at those
temperatures can distract attention from the dan-
gers of stacking materials that are initially hotter than
their surroundings.
Several subordinate features arise in any practical
discussion of large assemblies. First, the materials
presenting these problems are often poor conduc-
tors of heat so that uniform-temperature theory is
not enough. Secondly, when the materials are close
to the watershed stationary state, their dynamic be-
haviour is not the same as at the classical point of
criticality (with the drawn-out time scale obeying
the inverse square-root law). Thirdly, whilst for very
large amounts of material, characteristic times for
temperature-decay are very long, at these more el-
evated temperatures the characteristic chemical times
are abbreviated and there can be much less time
to make sytems safe. (The mixture of fact and fancy
about the effects of moisture, coupled with the vast
amounts of water required simply to cool down such
huge systems encourages premature and ill-consid-
ered opening of self-heating stacks and may ex-
change a conflagration for mere smouldering.)
The problem of timescales has recently been in-
vestigated by B. F. Gray & Merkin, 26 who call the
'hot assembly' problem the critical initial value
problem. It aims first to calculate the highest, safe
temperature of assembly, safe meaning that the sys-
tem remains quiescent and does not run away. The
second aim is to discover a rule for time-to-ignition
for marginally unsafe systems for which the initial
temperature T exceeds its stationary value by a tiny
amount A. The parallel circumstance near the point
of criticality has an exponential growth with a (slow)
start obeying
,'foc ~k-1/2
i4 HOTI'EL PLENARY LECTURE
For the new problem:
x ~ In(l/A)
So escape is brisker partly for this reason but also
because characteristic times for chemical reaction to
accelerate become shorter as temperatures rise.
Bowes' bookz5 describes some of the investiga-
tions of these systems when a large pile of material,
all at the same elevated temperature, is assembled
as quickly as possible from a very large number of
individual small 'packets.' There are some of the
liveliest practical investigations of the thermal ex-
plosion theory ever performed, ranking with Walk-
er's extensive examinations of fibrous, porous and
powdered substrates. Broadly speaking, a uniform
temperature distribution tends fairly rapidly to the
appropriate stationary-state distribution for the same
total heat-release rate and it then either accelerates
or decays away.
12. Oscillatory Behaviour in Combustion
and in Simpler Exothermic Systems
In the combustion field, studies of oscillatory re-
actions begin with the observations by Newitt &
Thornes in 1937 of repetitive cool flames in the
closed-vessel oxidation of gaseous propane. They are
normally to be found (see ref. 28 for a brisk survey)
as oxidations proceeding with a feeble, blue chem-
iluminescence, visible only in a darkened room.
Newitt and Thornes reported sequences of up to
five successive luminous pulses in propane oxida-
tion. Thus they provided the experimental basis for
Frank-Kamenetskii to point to the existence of ho-
mogeneous oscillatory oxidation phenomena that did
not depend, e.g., on diffusion processes. What seems
obvious today seemed strange then and the ines-
capably oscillatory character was still not stressed
elsewhere even when Pease (1940) produced eight
such pulses in propane oxidation or even, many years
later, when Yokely detected eleven pulses by tem-
perature measurements in the same system.
The paths of study divided. To produce and study
indefinitely sustained oscillations, open systems with
a constant supply of reactant are required. To sim-
plify interpretation, temperatures and species con-
centrations need to be made uniform: mechanical
stirring is generally needed for this. Finally, for a
really secure experimental base, more than one
variable should be monitored continuously. All this28
had to wait a long time. In early days, only light
pulses were measured and only the end-products of
oxidation were analyzed. The eventual use of well-
stirred, flow-through systems had constituted one
of the major advances. By the 1960s, interpreta-
tions of real systems could draw on increasingly ex-
tensive mathematical theory of the dynamical be-
haviour of simple, exothermic reactions in a cstr as
provided by work of chemical engineers (especially
of the Minnesota schooll though this was a river
with many tributaries).
The other streams of work have their origins in
early Russian studies. One dealt with closed sys-
tems, or at least furnished a basis for modern closed-
system studies. The other sought to make the re-
lationship between single explosive events in closed
systems and repetitive explosive events in open
systems (in a rather dismissive way). The serious
study of thermokinetic oscillations in closed systems
goes back to Sal'nikov, who examined the behav-
iour of the system:
P --> A kop
A~B+heat ka
i.e., precursor ~ intermediate --+ inert product +
energy release. In the simplest case, (a) the pre-
cursor was assumed to generate the intermediate at
a constant rate (precursor itself being maintained at
constant concentration), (b) the second step was both
exothermic (Q # O) and responsive to temperature
with
k cc exp(-E/RT).
The scheme was broadened3~ by B. F. Gray and
C, H. Yang and has recently received rather in-
tensive investigation in Leeds and in Sydney. It is
a remarkable example of how a simple model can
generate rewardingly rich behaviour. (It does not
adequately explain hydrocarbon oxidation but it
contains many lessons: it was the distinction of
Sarnikov's original work that it insisted on a rig-
orous proof of oscillatory behaviour.)
It is instructive to set out the conservation equa-
tions governing mass and heat2balance in this ele-
mentary system. For a concentration of P main-
tained constant at Po with a constant ambient
temperature Ta, the two conservation equations are:
dt kopo - ka kopo - kaa exp ;
a
dT Qka hS
-dt- = -eCv
- Vcrc~(T - Ta).
The energy-conservation expression can be writ-
ten in dimensionless form:
dO
- - ~ ore O -- O.
d,
We choose O = (T - Ta)/(RTa2/E) as usual and
eL = a/aft, where arf = (hS/QVka)(RTa2/E) = earn.
Here aign is the concentration of the intermediate
 CHEMISTRY AND COMBUSTION
A that, if present alone, would just become explo-
sively unstable.
In these terms the mass-balance expression be-
comes
d~
- Ki~ - ~e~
d~
The two parameters K and r satisfy K = kat N
and o" = (kotNPo/a~f), so o" = (kopo/koaq). (In their
study, Scott & Kay work in terms of K and ~,
tr, so ,~ = IJ,/K). This system gives a unique sta-
tionary state with elegant parametrie expressions:
O~s=,X and ass=,re-L
"lq~e stability of this stationary state is conditional.
When loss of heat is slower than consumption of A
so that
-> (~r - 1)e-" -> e -z ~- O. 13,
there is stability everywhere. When K < e -2 there
is a middle range of unstable behaviour between
the two values of ,~ satisfying
(o" - 1)e -r = K.
At these values, oscillatory behaviour sets in or ends.
It should be noted that although values of the sta-
tionary-state concentrations do not exceed those
predicted to be explosively unstable by simple
thermal explosion theory, values of the (stationary,)
excess temperature 0 certainly do exceed the nor-
mal stable limit (Oig, = 1). It is also worthy of note
that the amplitude of the oscillations is very sub-
stantial and that oscillatory temperatures may range
from near zero to more than 19 = 10 when their
(unstable) focus is at | ~ 4.
Precursor Decays: p = Po exp (-kot):
All this development accepts the idea that a sta-
tionary state is achievable and that in turn depends
on assuming a constant concentration for the pre-
cursor. If the precursor decays slowly from an ini-
tially large value to zero, then an evolution in time
can be described as oecuring in four phases. In terms
of temperature excess we may find:
I First, a fast jump of temperature 0 to the sta-
tionary state value appropriate to initial con-
ditions. The approach may be monotonic, or by
overshoot or by overshoot plus damped oscil-
lations.
II Secondly, an exponential downward drift of 0
in time, hugging the stationary-state locus more
closely as time passes.
III Thirdly, as the 'upper' Hopf bifurcation point is
15
passed and this locus ceases to attract, the 3-
dimensional trajectory winds out on to a sur-
Face and travels around it in a helical t~shion.
This is the oscillatory phase.
IV Next, after the second (lower) llopf bifurcation
point, the stationary, state locus again becomes
attractive. Oscillations die out well before all
the reactant is consumed.
V Finally, eventual exponential precursor-decay is
mirrored by the exponential decays of the con-
centration of 'intermediate' and excess temper-
ature towards zero.
It is quite remarkable how so simple a model so
graphically represents the major features of many
real systems in closed vessels. Moreover, it links
many properties to the parameters of the system
and so to one another. Amongst these are such
properties as the duration of the oscillatory and pre-
oscillatory modes, the frequencies and amplitudes
of oscillations at birth and death, and the values of
the various concentrations and chemical reaction-rate
coefficients, activation energies etc.
13. Chaotic Behaviour of Thermokinetic Origin
Even systems of only one variable (such as the
adiabatic cstr) can show multiplicity of stationary
states and hence ignitions and extinctions. These
are called static instabilities. Systems of two vari-
ables--a single reaction in a non-adiabatic cstr--
can show not only nmltiplc stationary states hut also
sustained oscillations. These are dynamic instabili-
ties. In recent years much attention has been fu-
cussed on oscillations that show period-doubling se-
quences and chaotic behaviour. For this kind of
versatility, a mininmm of three variables is re-
quired. Certain prototype systems deserve to be
listed. They all relate to theory rather than exper-
iment. In practice a key question is, what are the
effects of the small but inescapable variations in
conditions that always occur in "'real life"? Ulti-
mately, of course, there is an irreducible minimum
of fuctuations that is related to the moleeular na-
ture of matter, and most theoretical eflbrt has gone
into studying that. However, real exothermie sys-
tems in thermal contact with their surroundings will
show, e.g., temperature fluetuations far in excess of
this minimum. Useful analyses of these important
questions seem scarce: perhaps this admission of ig-
norance will generate a lively response. The tahle
lists a few early examinations of "unforced' systems.
Period-doubling leading to chaos in response to
forcing had been established earlier, but the un-
forced examples are, even ff only at first sight, more
surprising. One of the most suggestive of recent in-
vestigations is a numerieal-plus-ana[ytieal study of
virtually isothermal autocatalysis
16 HOTrEL PLENARY LECTURE

TABLE III
Minimal systems showing chaotic responses to thermal feedback

System Variables Investigators Date

CSTR Concurrent reactions CA, Ca, T Lynch D. T., Rogers T. D. and

A---> C, B---> C Wanke S . E . 1982
Math. modelling 3, 3, 103.
CSTR Consecutive reactions CA, Ca, T Aris R. A. and Jorgenson D. V.
A ---> B --* C Chem. Eng. Sci. 38, 45 1983
CSTR Single reaction Two CA, TI & T2 Planeaux J. B. & Jensen K. F.
heat capacities Chem. Eng. Sci. 41, 1491 1986
Semi-batch reactor CA, Ca, T Scott S. K., Kay, S. R. & Gray P.
Autocatalysis plus self- J. Phys. Chem. 94, 3394 1989
heating

P~ A ko = koo exp[(E/RTo) - (E/RT)] 14. Unifying Themes

A---~ B ka This Review Lecture began with the intention to

A + 2B---~ 3B k~
B --~ C ka AH = -Q; Q.~ RTo
The strictly isothermal system (with Q = 0) is one
of only two variables when p is kept constant. It
shows oscillations conditionally (so long as k, <
(1/8)ka) over a range of values of Ix satisfying
(ixz) +_ = 1 + 2K -+ (1 - 8K) 1/2
where K = k,,/kd and
IX = kopo(kdk 3d)1/2
These oscillations are of simple form. As soon as
the reaction becomes non-isothermal we have a 3-
dimensional system and if any of the steps in the
scheme is made sensitive to temperature, we also
have a feedback mechanism. The intensity of feed-
back is measured by the group ~ = ATad/(RTo~/E).
As this quantity increases, so does the possible va-
riety of response. At first oscillations remain simple
but a value of ~ is then reached at which com-
plexity sets in. For such values of ~j, period-dou-
bling sequences exist in the oscillatory region as IX
is varied. These sequences can lead into chaos and
out again. What is perhaps unexpected is the very
small value of exothermicity or responsiveness to
temperature that is needed to cause these compli-
cations, because all real chemical reactions are ac-
companied by heat effects and the displays calcu-
lated here are probably waiting to be found in many
real systems.
emphasize broad lessons in the relationships be-
tween chemistry and combustion and to offer a var-
ied set of illustrations. In fact, "real chemistry" has
taken rather a back seat as the value of simplified
representations and model schemes has been ex-
ploited. It has been commonplace in the technical
exploitation of combustion to hear it stated that the
chemistry does not matter, and that physical pro-
cesses (like mixing) hold the key to understanding
and to advances in applications. To the extent that
chemical processes are not only multiple, full of al-
ternative routes and above all rapidly increasing in
rate with temperature, neglect of chemical details
in engines and furnaces is not a bad first simplifi-
cation, especially for a chemist, to make. However,
as limits to performance are approached, so deeper
understanding is called for. Problems of knock, of
unwanted emissions and of substitute fuels all de-
mand this depth. One of the most interesting chal-
lenges at present is the need for some convenient
but not oversimplified "global kinetics", interme-
diate between schemes involving hundreds of ele-
mentary steps (these are already with us), and op-
timistic mass-action laws involving only expressions
like rate ~ [fuel] ~ [oxygen] t3.
Most of the discourse has concentrated on the
interplay between heat-release, self-heating, reac-
tant consumption and reaction rate, and thus on
topics that are lumped together under the umbrella
of thermal explosion theory. The first unifying theme
is the concept of feedback: consequences of the
change having effects upon the change. We have
not said much about the chain-branching reactions
that offer isothermal acceleration. They take the form
of free radical multiplication, and different steps are
 CHEMISTRY AND COMBUSTION 17

important over different ranges. In heptane com- symbols and ideas of thermodynamics, constant use
bustion, for example, all of the following branching casteth out fear and a sympathetic notation cer-
steps can be important, each under different con- tainly eases the process.
ditions:

HO2CTH1402H ~ HO + OC7H1402H Nomenclature

H202+ M~2HO+ M
Symbol Meaning SI unit
H + O2--~ HO + O a species concentration mol m -3
ao radius, half-width m
B, B* dimensionless excess
Instead we have concentrated upon thermal feed-
temperatures
back, with energy release serving to heat the sys-
c, Co species concentration mol m -3
tem, so causing the energy-releasing reactions to go
cv, cp specific heat capacity J K -1 kg -1
faster. The combination of heat-release and respon-
E activation energy J mo1-1
siveness to temperature has been expressed by a
k reaction velocity coeffi- s -1 (conc) 1-m
dimensionless group B, the Zel'dovich number. The
cient
constituents of B are ATad which involves the exo-
rate of heat loss W
thermicity, and RTrfZ/E which involves the acti-
m reaction order
vation energy. To each reaction corresponds its own
p concn of species P mol m -a
B and it is their collective outcome that we must
Qv, Qp exothermicities J mol -I
establish in complex systems.
R universal gas constant J K-lmo1-1
The second is less a unifying principle than an
rate of heat production W
observation of the near-universality of non-linear
S surface area m2
responses in chemistry, so that feedback is non-lin-
t time s
ear, our rate-equations are non-linear and the tex-
T temperature K
ture rich. But this pervasive quality also means that
u flow-rate mas- 1
quite simple models can enjoy a rich pattern of re-
V volume m3
sponses. The Sal'nikov2~ scheme as the prototype
W steepness of temperature K s -1
redeveloped here for oscillatory behaviour, and ex-
ramping
tended to the circumstances of closed systems is just
X parameter
such an illustration.
Ze Zeldovich number
More fundamental in its implications than this is
the choice of starting point that we should take.
Greek symbols
We begin with the conservation laws of matter and
energy. If we choose to suspend them it should be ot dimensionless concentra- --
briefly. One example is the primitive foundation for tion
studying thermal explosion, where reactant-con- 13 diffusive transfer m s -1
sumption is at first ignored and ignition treated coettqcient
simply as a sudden jump away from a stationary F boundary t e m p e r a t u r e - --
state. This jumping between stationary states can, gradient
of course, be realized for an open system like the e (=RT~/E)
cstr. Otherwise it is an invaluable first approxima- ~l effectiveness factor --
tion but not the whole story. A increment
Chemical prejudices in the combustion field en- O d i m e n s i o n l e s s excess --
courage us to disregard the reversibility of chemical temperature
reactions because combustions very obviously tend 8 Frank~Kamenetskii --
very near indeed to completion. So in combustion, parameter
any approach which required overall reversibility as kO Semenov parameter (Se) --
a vital ingredient of understanding unstable behav- dimensionless time --
iour would deserve little encouragement. The K thermal conductivity W m -x K -1
opposite attitude has been fostered in the isother- tr parameter
mal oscillatory field without perceptible gain: the I~ dimensionless reaction- --
approach from oscillatory combustion is not a bad rate
preparation. On a lower level of generality but with intensity of thermal feed- --
a bonus of convenience comes the use of dimen- back
sionless groups. There is something in a chemist's p dimensionless distance --
education that impedes their adoption. Like the k fractional concentration --
18 H O T r E L PLENARY LECTURE

Subscripts and superscripts pertaining to different P. AND HARVEY, D. I.: Phil. Trans. Roy. Soc.
conditions A270, 467 (1971).
18. GRAY, P. AND LEE, P. R.: Thermal Explosion
rf reference, as in Tff Theory in Oxidation and Combustion Reviews,
ad adiabatic, as in ATad Vol. 2, p. 1, Elsevier 1967.
cr criticality
19. KORDYLEWSKY,W.: Kommunikat 1-20/K-020/78
ign ignition
Tech. Univ. Wroclaw (1978); BODDINGTON, T.,
ext extinction
FENG, C. G. AND GRAY, P.: Proc. Roy. Soc.
a ambient, as in Ta
A390, 247 (1983).
o initial value, as in Co 20. BODDINGTON,T., GRAY, P. AND SCOTT, S. K.: J.
res residence, as in tres Chem. Soc. Faraday Trans. Vol. 78, p. 1721 &
ch chemical
1731 (1982).
N newtonian cooling
21. FINE, D. H., GRAY, P. AND MACKINVEN, R.:
Proc. Roy. Soc. A316, 241, 255 (1970).
Acknowledgments 22. VANP~E, M.: Compt. rend. Acad. Sci. Paris 241,
951 (1955).
I am grateful to Dr. Richard Gann for his un- 23. GOODMAN,O. AND GRAY, P. : Trans. Faraday Soc.
failing courtesy as an editor and to the Combustion 66, 2772 (1970).
Institute and its British Section for their support. 24. BODDINGTON,T. AND GRAY, P.: Proc. Roy. Soc.
A320, 71 (1970); BODDINGTON,T., GRAY, P. AND
SCOTT, S. K. : Proc. Roy. Soc. A378, 27 (1981).
25. BOWES, P. C.: Self-heating: Evaluating and
REFERENCES Controlling the Hazards, London H.M.S.O.,
1984.
1. SEMENOV, N. N.: Z. Physik. 48, 571, 1928. 26. GRAY, B. F. AND MERKIN, J.: J. Chem. Soc.
2. FRANK-KAMENETSKII,D. A. : Diffusion and Heat Faraday Trans. 86, 597 (1990).
Transfer in Chemical Kinetics, 2rid ed., Plenum 27. WALKER,I. K.: New Zealand J. Sci. 4, 775 (1961)
Press, 1969. and many subsequent papers.
3. ZEL'DOVlCH, YA. B.: Zh. Tekh. Fiz. 11, 501 28. GRAY, P., GRIFFITHS, J. F., HASKO, S. M. AND
(1941); ZEL'DOVICH,YA. B. AND ZYSIN,: Zh. Tekh. LIGNOLA, P. G.: Proc. Roy. Soc. A374, 313
Fiz. 11, 510 (1941). (1981); A396, 227 (1984).
4. GRAY, P., GRIFFITHS, J. F. AND HASKO, S. M.: 29. SAL'NIKOV,I. E.: Zh. Fiz. Khim. 23, 258 (1949).
Proc. Roy. Soc. A396, 227, 198ft. 30. GRAY, B. F. AND YANC, C. H.: J. Phys. Chem.
5. KASSOY, D. AND LINAN, A.: Q. J. Mech. Appl. 69, 2747 (1965); 73, 3395 (1969).
Math. 31, 99 (1978). 31. GRAY, B. F. AND ROBERTS, M. J.: Proc. Roy.
6. THOMAS, P. H.: Proc. Roy. Soc. A262, 192 Soc. A416, 391 (1988); GRAY, P., KAY, S. R. AND
(1961). Scoa'r, S. K.: Proc. Roy. Soc. A416, 321 (1988).
7. LACEYA. A.: Int. J. Eng. Sci., 21, 501 (1983). 32. GRAY, P. AND SCOTT, S. K.: Chemical Oscilla-
8. ADLER, J. AND ENIC, J. W.: Comb. Flame 8, tions and Instabilities, Clarendon Press, Oxford
97 (1964). 1990.
9. KORDYLEWSKI,W. AND SCOTT, S. K.: Proc. Roy. 33. Ares, R. : The Mathematical Theory of Diffusion
Soc. A390, 13 (1983). and Reaction in Permeable Catalysts, Claren-
10. TYLER, B. AND WESLEY, T. A. B.: Eleventh don Press, Oxford, 1975.
Symposium (International) on Combustion, p.
1115, The Combustion Institute (1967).
11. GRAY, B. E. AND SHERRINGTON, M. E.: Comb. Appendix: Limitations on the Use of O
Flame 19, 435, 445 (1972). and the Form exp O
12. BILOUS, O. AND AMUNDSON,N. R.: A. I. Chem,
Eng. J. 1, 513 (1955). Recently B. F. Gray, J. Merkin and G. C. Wake
13. GRAY, P., KORDYLEWSKIW. AND KRAJEWSKI,Z.: (GMW) have urged the avoidance of O and exp O
Proc. Roy. Soc. A412, 45 (1987). and urged the use of the Arrhenius expression in
14. BODOINGTON, T., GRAY, P. AND WAKE, G. C.: the form exp-(1/u), where the variable u stands for
Proc. Roy. Soc. Lond. A393, 85 (1984); A401, RT/E. This is the choice made around 1956 by Vulis
195 (1985). (following Todes, 1936) and later by Hardee et al.
15. GRAHAM-EAGLE, J. G. AND WAKE G. C.: Proc. The point of view of GMW is that of the student
Roy. Soc. A407, 183 (1986). of differential equations rather than that of com-
16. GRAY, B. F. AND WAKE, G. C.: Comb. Flame, bustion but they have interesting comments to make
79, 2 (1990). that should not be ignored. For example they cor-
17. See e.g. references in BODDINGTON, T., GRAY, rectly suggest errors of order RTa/E in many so-
 CHEMISTRY AND COMBUSTION
lutions. Vulis' book is the best starting point
for a host of examples for uniform-temperature
systems. For distributed-temperatures, the variable
u is less user-friendly, leading immediately to
numerical computation--analytical solutions are
completely out of reach.
An interesting recent example of features high-
lighted by the new variable u and not by Frank-
Kamenetskifs O is the discussion of "disjoint
solutions" in the simplest of open systems--a first-
order exothermic reaction in a cstr. The exact
expression for turning points on a stationary state
locus (see section 2) is:
T,~a - To~ T2 = tion scientist not only to avoid absolute zero as a
1+ E?R J - (Tad'4- W~ + W~ O.
The physically realistic ordering is To ~ Taa
E / R since To ~ 500 K, Tad ~ 3000 K and E / R
19
30,000 K. Under these circumstances, which are
the norm, ignition and extinction s~tisfy:
T_ - To ~- RToZ/E; Taa - T+ ~- R T 2 a d E
If, with GMW, we go outside the normal realm of
combustion conditions (abandoning strongly T-de-
pendent reaction-rates) and allow Tad - To to be
as big as E/R, an extinction point can be located
at T+ = E/2R, when To is absolute zero. Values
of T+ on the "wrong side of the To axis" are also
solutions. These lively and unexpected results (or
forgotten results--see Vulis) remind the combus-
reference temperature for O but also to be sure that
his chemical system is indeed strongly responsive
to temperature, and not up in the region of Te =
E/R.