Downloaded via UNIV OF HYDERABAD on August 26, 2022 at 11:10:29 (UTC).

See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

OF
THE
THEORY
Though the history of oscillating reactions is intimately connected with modern
biochemistry, particularly through the studies of microbial ecology, the treat-
ment of oscillating reactions is applicable to a wider class of phenomena.
The author presents the theory of oscillating reactions in a highly biological
context, but generalizes as much as possible for the benefit of the general
technical community.

in the theory of oscillating reactions has itself oxidant and reductant. While the overall reactions
Interest
fluctuated greatly and has been largely derived from are simple, the detailed mechanism is still not known,
associated problems in biology. Prehistoric man was though homeography would dictate that periodic acid
undoubtedly familiar with the diurnal and annual should certainly be involved. There is still a question
oscillations in gross biological behavior which he pre- whether the reaction takes place in a homogeneous
sumably attributed to the “sun god.” More recent phase and, more generally, whether oscillating reactions
studies have demonstrated the existence of internal can actually occur in homogeneous systems (7, 27).
biological clocks synchronized to the external rhythms. It was not until the 1950’s that interest was renewed
With the modern attempt to establish a physical and in the relationship between chemical mechanisms and
chemical basis for biology, the possibility of oscillating kinetic oscillations. Important results, discussed below,
chemical reactions serving as a source for such rhythms were obtained by Prigogine (20) through the application
has received renewed attention. In turn, the relations of irreversible thermodynamics and by Bak (7) through
between kinetics and the types of chemical reactions analysis of first-order and pseudo-first-order reactions.
involved have become increasingly important. However, their results demonstrate that some reaction
In view of the close association with biology, it is not schemes and conditions could not lead to sustained
surprising that the first plausible mechanism was oscillations.
proposed by Lotka (75), in 1920, to explain ecological More positive results, though largely qualitative,
oscillations. The mechanism can be written as were derived from the application of engineering con-
[G] + A A 2A) cepts to biochemical systems. In the area of cellular
A + B A 2B > (1) reactions, the use of feedback concepts and experimental
B A ) evidence for chemical feedback were firmly established
where animal A eats the grass G, which is kept in a con- by 1961. Jacob and Monod (72), in particular, recog-
stant supply as indicated by the brackets [ ], and re- nized that chemical circuits involving cross feedback
produces. In turn, animal B eats animal A to reproduce,
could have bistable states similar to electronic multi-
and finally animal B dies off. As Lotka obtained the vibrators, and might thereby account for cellular dif-
dynamical equations by applying the Guldberg-Waage- ferentiation. In addition, they realized that astable
van’t Hoff law of mass action, the chemical mechanism states yielding oscillations, were also possible.
has frequently (7, 20) been studied. The equations Both theoretically and experimentally, however, the
can be written in the form: detailed relationship between chemical mechanism and
x = x (ei —

y) oscillatory response remained elusive. In 1961 there

was no clear cut experimental evidence for oscillations
y y (*
=
(2) in cellular reactions. In addition, the theory was largely
“

c2)
where ei =
k\ [GJ/fo, e2 k3/k2,
= and t k2t\ where= tf qualitative. Quantitative attempts to mathematically
is real time and x =
dx/dt. As shown by Volterra prove the possibility of oscillations for known chemical
(22), the equations can be solved in closed form and yield reactions were curtailed by the nonlinear differential
semistable rotations around the singularity. Analog equations, and when the mathematics was made tract-
computer solutions are given in Figure 1. For details able the chemistry was no longer satisfactory.
of this and other figures, see Appendix III. Though Both problems have now been partially resolved.
there are no reactions known which satisfy the Lotka The work of Chance and his colleagues has definitely
mechanism, the latter dominated virtually all theo- established the existence of oscillatory kinetics in cells
retical studies for nearly 50 years because of the mathe- and cell extracts (6, 5), while the combination of elec-
matical tractability of the equations. tronic computers and more advanced mathematical
While experimental examples of periodic reactions techniques has made it practical to examine the known
in heterogeneous systems have long been known, it was chemistry.
only in 1921 that Bray (5) discovered a periodic reaction Figure 3 shows a train of damped oscillations in the
in a homogeneous solution. As indicated in Figure 2, concentration of DPNH (reduced dihydrodiphospho-
that reaction involves a catalytic decomposition of pyridine nucleotide) as observed by spectrophotometric
hydrogen peroxide by iodine and depends on the simul- studies on intact yeast cells under appropriate conditions.
taneous behavior of hydrogen peroxide as both an Assays (3) of other metabolite intermediates in the

VOL. 59 NO. 5 MAY 1 967 19

 glycolytic pathways demonstrated that they all ex-
hibited an oscillatory response. Still more interesting
results have recently been obtained on apparently
homogeneous (single phase) extracts (7 7) derived from
both yeast and heart muscle cells as shown in Figure 4.
The fact that the oscillations are virtually sustained and
exhibit both increasing and decreasing amplitude
should be carefully noted.
Such experimental results provided the impetus for
the theoretical studies discussed here. While it is not
Figure 7. Analog computer solutions of the Lotka mechanism, showing
typical kinetics and phase plane trajectories the intention of this paper to discuss the detailed inter-
Note that the phase For the reaction
relation between those specific experiments and the
plane consists of semistable cycles.
scheme shown, Ki = 0.086, K2 0.9, and Kd
=
0.1. =
Scales: A, B = theory, it is, nonetheless, convenient to use some aspects
10. T = 2.13 sec. [M] 1.0.
=
of the glycolytic reactions as a model for illustrating
the general application of the results. Extensions to
other types of reaction mechanisms are relatively
straightforward.
Figure 5 gives an abbreviated form of the major
reaction steps in glycolysis. Several forms of chemical
feedback are evident. The coupled reactions involving
DPNH and ATP provide feedback through stoichio-
metric reactions. In addition, the reaction steps are
catalyzed by enzymes whose specific activity may
TIME (DAYS)
depend on the metabolites as well as various ions, thus
Figure 2. Experimental results obtained by Bray (5) for the catalytic
providing nonstoichiometric (or catalytic) feedback
decomposition of hydrogen peroxide by iodine
pathways.
Overall reactions considered by Bray are 5 H202 + 12 2 HIOs + =

4 H20, and 5 H202 + 2 HIO-& 5 02 -f- /2 -f- 6H20. Overall re-

=

action: H202 H20 + V2 02. Periods depend on the pH and temper-

=
DESCRIPTIVE ASPECTS OF OSCILLATORY
ature, and can be varied from minutes to days KINETICS
In view of the relationship to mechanism, it is im-
portant to distinguish and characterize some particular
types of oscillatory and nonoscillatory kinetic response,
and to define carefully various descriptive features.
The basic types are listed below.

Nonperiodic Forms
Pure Monotonic—as illustrated in Figure 6A. The
kinetics are described by a rising or decaying curve, and
PLUS can generally be represented in the form
5.5mM GLUCOSE

y(t) = c ± D(t) (3)

Figure 3. Experimental results obtained by Chance et al. (6) for
the kinetics of DPNH in intact yeast cells under conditions of anero- where C is a constant, D(t) is a monotonically decaying
biosis function such that /)(<») 0. The kinetics are non-
=

Observations were made with a double-beam spectrophotometer as indicated oscillatory.

Overshoot-Undershoot—as illustrated in Figure 6B.
In this case the kinetics appears to exhibit an oscillatory
response. However, the response is not periodic. Such
kinetics can be represented in the form

y(t) = C + X) (4)
i =» 1

where Di(t) are distinct monotonically decaying func-

tions such that Dt(co) 0. In general, such curves
=

Figure 4. Experimental results by Frenkel (77) for the kinetics of

DPNH in heart muscle cell extracts. Observations were made by can exhibit only a finite number (at most n 1) of —

jluorometry overshoots and undershoots relative to the final value.

Note the increasing and decreasing amplitude and frequency variation.
In addition, they are not periodic—i.e., the time between
These are previously unpublished data maxima is not constant.

20 INDUSTRIAL AND ENGINEERING CHEMISTRY

 Figure 5. Abbreviated diagram of the glycolytic pathway 3-phosphate; PYR =
pyruvate; PGA = phosphoglyceric acid; AA =
acetaldehyde; EtOH =
ethanol; ATP, ADP, AMP = adenosine
All reactions catalyzed by enzymes, which are designated by boxes in some
are
(tri, di, mono) phosphate; DPN = diphosphopyridine nucleotide; hex =
cases. Dashed arrows indicate known or suspected inhibition or activation hexokinase; pfk = phosphofructokinase; aid aldolase; gapdh =
=

of the specific enzyme activity by the metabolites. Nomenclature: GLU- glyceraldehyde 3-phosphate dehydrogenase; pgk phospho glycer ate
=

glucose; G1P glucose-1-phosphate; G6P

— =
glucose-6-phosphate; kinase; pyk =
pyruvate kinase; car = pyruvate decarboxylase; adh =
F6P =
fructose-6-phosphate; FDP =
fructose diphosphate; GLY =
alcohol dehydrogenase; mk = adenylate kinase; Pi = phosphate; PEP —

glycogen; DAP =
dihydroxy acetone phosphate; GAP =
glycer aldehyde phosphoenolpyruvate

Periodic Forms normalized amplitude. The function A(t) may be

increasing or decreasing, and may approach a constant.
Damped Oscillatory—as illustrated in Figure 6C. Such an analysis, based on the apparent characteristics
In this case, the kinetics can be described as
of the kinetics, is not analytically unique, but it is con-
y(t) = C + D(t)P(t, co) (5) venient in the qualitative sense.
More complicated forms of kinetic responses exist and
where P(t9co) is some periodic function with funda-
may require a sum of several periodic functions for their
mental frequency co and normalized to unit amplitude. to be
description—e.g., a doubly periodic response
It may be noted that the kinetics exhibit an infinite discussed later. These can be described as
number of maxima and minima, although the amplitude
decreases to zero as t — co. In practice, it is often y(t) =
y(t) +
i
Xi Ai(t)Pi(t, coi) (8)
difficult to distinguish this case from the overshoot-
=

The periodic function may also exhibit a slowly

undershoot case. Thus, if D(t) becomes very small,
further oscillations may not be observable» and the varying frequency as can be seen in Figure 7. In such
cases the frequency, co, can be treated as a function of
kinetics may appear to be of the previous form (Equa-
tion 4). time. While the changes in frequency are continuous,
it is usually sufficient for graphical purposes to evaluate
Sustained Oscillatory—as illustrated in Figure 6D.
In this case the kinetics can be represented as the period centered about some particular point in the
oscillation for each cycle.
y(t) = C + A Pit, co) (6) Emphasis is placed on graphical analysis of the kinetics
in terms of observable components such as frequency,
where A is a constant and the periodic oscillations amplitude, and mean values due to the nonlinear
represented by P(t, co) persist as t — <» . nature of the oscillations, i.e., derived from nonlinear
General Oscillatory—as illustrated in Figure 7. In differential equations. For systems of linear equations
this general case the kinetics can usually be represented with constant coefficients, Fourier analysis of the kinetics
as provides considerable information as to the order of the
differential equations and other properties. How-
y(t) =
y{t) + A(t)P(t, co) (7)
ever, there are no such simple relations for nonlinear
where y(t) is the short time average (i.e., the apparent systems; a complete Fourier analysis of the kinetics
mean value); A(t) is the amplitude modulating func- would provide little, if any, information. Similar limi-
tion which is slowly varying compared to the periodic tations apply to other descriptive aspects of sinusoidal
component; and P(t, co) is the periodic function with (linear) oscillations.
VOL 59 NO. 5 MAY 1967 21
Phase Differences

As will be seen, the variables of a nonlinear oscillator

do not generally exhibit a constant phase shift over one
period. Several different measures of phase shift can
be given. If x(t) and y(t) are two oscillatory variables
having the same frequency, and if the respective normal-
ized (amplitude 1) periodic components, Px{t) and
=

Py(t) are resolved out, then we can define

= =
phase difference of the maxima (9)
--

&copy;max
r

where r is the period and tm time difference between =

maxima of Px(t) and Py(t). Normally, the range for 0

is restricted to tt < 0 < 7r, and tm is taken as plus or
—

minus accordingly.
Similarly, one can define
&copy;min
=
phase difference of minima (10)
A crude average value can be defined as

0 crude average phase difference

2
(n)
However, by comparison with sinusoidal oscillations
the following definitions appear more relevant. If a Figure 6. Sketches of typical types of kinetics and associated phase
plane behavior
phase plane graph is constructed with Px as the x-axis
,

component and Py as the y-axis component (/ is treated A—monotonic, B—overshoot-undershoot, C—damped oscillatory, D—
sustained oscillations (limit cycle)
as a parameter) then a closed curve is generated in the
plane. If the curve is approximately elliptical so that
a major axis (of length a) and a minor axis (of length b)
slight disadvantage that it is only approximately correct
can be defined, then an average difference can be meas- for true ellipses although the error is never very large.
ured by The error derives from the approximation of the circum-
ference of an ellipse by
&copy;axis
=
.2 arctan b/a =
average phase difference
(12) based on elliptical axis C = 2 rV (a2 + b2)/ 2

which has the advantage of reducing to correct values for rather than through the use of an elliptic integral which
linear oscillations. would lead to undesirable complications in the formula
A final measure of average phase difference can be for 0. The advantage resides in the fact that it can be
obtained through the ratio R A/C where A is the area =
applied to simple closed curves for which approximate
enclosed by the phase plane curve and C is the circum-
elliptical axis may not be clearly defined. For these
ference of the curve. Then last two definitions, the proper quadrant for 0 can be
0/2 = arcsin (2R) =
average phase difference
defined as follows:
With Py(t) considered as reference—i.e., Py(t) ~

based on openness (13) sin cot and Px(t) ~ sin (cot + 0) then:
This definition of 0R is essentially a measure of the rela- (a) If the representative point in the phase plane
tive openness of the phase plane curve. It has the moves counterclockwise (as a function of t)
0 < 0 < 7r/2, if the major axis has positive slope

7r/2 < 0 < 7r, if the major axis has negative slope *

(b) If the representative point moves clockwise (as

AUTHOR Joseph Higgins is Assistant Professor of Bio-
a function of t)
physics at the Johnson Research Foundation, University of
Pennsylvania. He acknowledges the assistance of Mr. G.
—

7r/2 < 0 < 0 if the major axis has positive slope

Rangazas in carrying out the associated computer studies, of —

7r<0< —7t/2 if the major axis has negative slope

Mrs. J. Summers in preparing the manuscript, and of W.
(14)
Nelson, Miss J. Chance, and Mr. Rangazas in checking equa-
tions and proofing. He also acknowledges the help of Prof. B. While the last definition of average phase shift appears
Chance in encouraging the development of the analysis, and is to be the most useful for nonlinear systems, it is still not
grateful for his stimulating criticism. adequate. In some cases, particularly with triangularly

22 INDUSTRIAL AND ENGINEERING CHEMISTRY

shaped cycles, it becomes difficult to define a proper For nonlinear systems, however, d, or c/i/4, may itself
be a function of (t). Save for purposes of comparison
major axis.
with linear systems, there is relatively little advantage
Damping Factors to plotting d(t) over a direct graph or a semilogarithmic

For damped oscillatory kinetics, a damping factor graph of D(t) or A(t). The dissection of arbitrary
is often used for descriptive purposes and is defined as oscillatory forms can be effectively completed when the
the ratio of the amplitude of the oscillations one period individual functions, y(t), P(t, co), co(t), and A(t) have
been resolved [e.g., see Chance (7)] as illustrated in
apart. In terms of Equation 5, it appears as
Figure 7.

damping factor = d = (15) THERMODYNAMIC CONSIDERATIONS

Thermodynamics and general kinetic considerations
It is also convenient to define a similar quantity based place certain constraints on the conditions for sustained
on a one-quarter period difference—i.e., oscillations, as well as providing a useful breakdown of
distinct types of kinetic states. As is well known, for
a closed system of chemical reactions, as t —> °°, the
final state achieved is the equilibrium state.
For an open system, in which substances consumed
For linear differential equations with constant coeffi- by the reaction are supplied from an external source
cients, D(t) is a simple exponential function and the while those produced are removed by an external sink,
damping factors are constants. Thus, for the final state (t °°) is a stationary state of the system.
The equilibrium state can be viewed as a special sta-
D(t) =

tionary state for which the net flux through the system
2 7rX 7rX is zero. As noted elsewhere (20), the stationary state
—

In (d) -

and —
In dii (17) of a system may still be time dependent as in the case of
4
co 2 co

sustained oscillations. An analysis of the steady state,

In general, it is convenient to define a damping “con- its existence and meaning, has been given elsewhere
stant” as
(15). It may be useful to point out that a system may
damping constant =
5 = —In (d) (18) involve more than one steady-state region defined
according to the particular subset of chemicals which
For oscillations with no damping, 5 0; if there is a
=

satisfy steady-state relation.

a
strong damping, 5 is positive and large; if the oscillations In general, if an open or closed system is perturbed,
are increasing in amplitude (undamping), then 5 is such as by the addition of one or several chemicals, the
negative. kinetics of the transition to the new final state can be
divided into several regions between an initial transient
region, in which the various chemical components
exhibit rapid changes, to a steady-state region which
gradually evolves into the final state. The steady-state
region (15) can be viewed approximately as a sequence
of stationary states for a subset of the chemicals in much
the same way as a reversible process can be viewed
as a sequence of equilibrium states.

An important theorem concerning the possibility of

oscillating kinetics has been developed by Prigogine.
Theorem: For chemical systems at constant tempera-
ture and pressure, rotations are thermodynamically
possible around nonequilibrium stationary states, but
they are not permitted around an equilibrium state.
(19)
The conclusion given by Prigogine assumes the appli-
cability of the Gibbs formula for entropy changes due to
nonequilibrium situations. There appears to be no
reason to doubt this assumption for normal chemical
Figure 7. General type of oscillatory response (top graph) and its reactions.
visual breakdown Because the conclusion is basedon an extension of the
In the center graph is a plot of the visual {or short time) mean value, y{t) thermodynamics of irreversible processes, it is applicable
and the amplitude function A{t). In the bottom graph, the frequency cc{t)
to systems involving any type and any number of
of oscillations in radians I min. is shown as a function of time. There is
also an indication of double periodicity in the center graph physically realizable chemical reactions. It follows

VOL. 59 NO. 5 MAY 1967 23

that sustained oscillations can only occur in open sys-
tems. Although closed systems may exhibit oscillatory
m
kinetics as they pass through the transient or steady-
state region, such oscillations cannot persist in time.

FIRST-ORDER AND PSEUDO-FIRST-

ORDER SYSTEMS
Thor Bak (7) has made an excellent, comprehensive
study of the possible kinetics of first-order (FO) and
pseudo-first-order (PFO) reactions. His results are
reviewed here together with some additional remarks.
For systems composed of only first-order and pseudo-
first-order reactions, the resulting differential equations
are linear and possess solutions of the form

yi(t) =
X) ai}eXit for i =
1, . . .n (20)
; = i

where yt represents the concentration of chemical (z),

the aij are constants which depend on the initial con-
centrations and the rate constants, while the depend
only on the rate constants. The are the eigenvalues
derived from the characteristic equation associated with
the original differential equations. For open or closed
systems, one of the eigenvalues (say, \i) is zero and the
Figure 8. Computer solutions for first-order and pseudo-first-order
corresponding coefficient (an) is the final value (as reactions
t —> oo) for the chemical (y*). The kinetic behavior
Case A: first-order reaction; Wegsheider conditions applicable. Note the
of such systems is largely, but not entirely, determined monotonic decay of the perturbed chemical (A) and the single overshoot in
by the values of the A*, which may be real or complex B in accordance with Theorem 22. Cases B and C: pseudo-first-order
reactions; all reverse rate constants are zero. The forward rate constants
numbers. are set for damped oscillatory solutions (X* complex) in Case B and for
nonperiodic functions (Xt- real and negative) in Case C. Note the compar-
able overshoots and undershoots in both cases. The distinction between
First-Order Systems
them is more evident in the phase plane which exhibits stronger rotational
character for Case B. Appropriate constants are shown below:
For first-order reactions, application of the principle
of detailed balance yields the Wegscheider conditions,
which require that the product of equilibrium constants
around any closed loop in the reaction scheme be equal 7H 824 ][3
D 6 C
to one. For example, the reaction scheme
5
A. K\ 0.9, K2
=
0.6, Kz = =
0.6, A4 =
0.9, Kh =
0.4, A6 =

A A, 0.4. Scales: substrates 10. T 0.13

2
0.3, K7 0.3, Ks
= = = =

S N X V sec.

B. Ki =
1.0, K3 =
0.7, K*> =
0.8, K7 =
0.9. Scales: substrates
= 10. T = 0.26 sec.

C. K\ =
1.0, Ks =
0.5, Ks —

0.15, K7 =
0.05. Scales: substrates
must satisfy the equations = 10. T =
0.5 sec.

K i 2 3 Af aK
3 41
= 1

and

KmKkKvKuK4* = 1
Consequently, the kinetics of first-order systems take
the nonperiodic form (Equations 3 and 4). As recog-
(note that Kz4 =
l/Ahs) nized by Hearon, this does not imply that the kinetic
It is clear that the Wegscheider condition imposes a
curves are represented as a monotonic decay to the
constraint on the possible choices of rate constants. stationary state. Bak’s statement that real, negative
eigenvalues (the A^) imply a monotonic decay to the
Extending a technique of Hearon’s (13), and assuming
stationary state is only valid for small perturbations.
only the Wegscheider conditions and the positivity In general, such nonperiodic forms may exhibit
of the rate constants, Bak has proved that:
overshoots and undershoots according to the values of
Theorem: For systems composed only of first-order the initial conditions and rate constants. It is easily
reactions, the eigenvalues are always real and negative. realized that a sum of n decaying exponentials (as given
The result does not depend on whether the system is in Equation 20) can exhibit at most (n 1) extrema
—

open or closed. (21) (either maxima or minima). For both the FO and

24 INDUSTRIAL AND ENGINEERING CHEMISTRY

PFO systems, since one of the X is zero, one of the ex- The interrelationship between the Wegscheider con-
trema is simply the final value (as t °o) and at most ditions, closed loops, and oscillatory response is easily
(n
—

2) relative extrema can be obtained. Conditions recognized for nearly irreversible reaction cycles. Thus,
for the actual occurrence of such extrema have not been the mechanism
mathematically analyzed; however, computer studies
suggest the following relevant result:

Computer Theorem: If a system of first-order reactions,

in a given stationary state, is perturbed by the addi- C
tion of only one chemical (say, Xi), then in the transi-
tion to the new stationary state (which may be the same formed by pseudo-first-order, nearly irreversible
as the old one), the kinetics for any chemical (Xi for
reaction steps presents a closed cycle. Since each of
i 1) can exhibit at most one relative extrema. In the equilibrium constants is infinity, the Wegscheider
addition, the perturbing chemical (Xi) exhibits a conditions are maximally violated—i.e., KiKzKz °°. =

monotonic decay to its new value. (22) To satisfy the Wegscheider conditions, the nearly irre-
Under the same conditions, if the system is perturbed versible steps must be arranged as
by the addition of two chemicals (say X\ and A2), then
the Xt (for i ^ 1 or 2) can exhibit at most two relative
extrema. At least one of the perturbing chemicals
will exhibit monotonic kinetics and the other can exhibit
C
at most one extrema. The extension to a larger number
of perturbing chemicals is obvious. Computer solutions
for a typical case are shown in Figure 8. and it is evident from the direction of the arrows that
the cycle is effectively not closed. Thus, we may con-
Pseudo-First-Order Reactions with Closed Loops clude that the fulfillment of the Wegscheider conditions
If any or all of the first-order reactions are treated as of effectively destroys any closed reaction paths which
the form might appear to exist on the basis of the stoichiometry.
Based on the previous discussion and computer studies,
[A*] + Y, —Ym + [B*] the following theorems may be concluded and compared
(for reaction k) (23) to the previous results for FO and PFO-NCL systems:

where Ak and B* may be different for each reaction step, (a) Necessity: For systems composed of pseudo-first-
and the bracket [ ] indicates that they may be con- order reactions to exhibit multiple (more than one) ex-
sidered constant over the course of the reaction, then the trema in the transition from one stationary state to
reactions between the Yi are pseudo-first-order reactions. another and when perturbed by addition of only one
The concentrations of the A* and B* can be incorpo- chemical, it is necessary that the reaction scheme pos-
rated into the rate constants which are then treated as sess closed loops and that the Wegscheider conditions

Thus, the system of be violated.

pseudo-first-order rate constants.
pseudo-first-order reactions appears identical to the sys- (b) k-Sufficiency: (1) When the conditions of a are ful-
tem of first-order reactions except that the Wegscheider filled, there exist values of the rate constants for which
conditions need no longer apply. For such systems, Bak all the eigenvalues are real and negative and overshoot-
(/) has demonstrated the admissibility of complex eigen- undershoot response can be exhibited. In such cases,
values with negative real parts. Consequently, such for a system of n chemicals, at most (n —

2) relative ex-
trema may be exhibited by any one chemical. For
systems can exhibit damped oscillations. It is important
to realize that pseudo-first-order reactions without certain schemes of pseudo-first-order reactions, the rela-
closed loops (PFO-NCL) are dynamically identical to tion between the number of extrema actually exhibited
first-order systems. Consequently, all the remarks and and the rate constant has been discussed elsewhere (16).
theorems applicable to FO also apply to PFO-NCL. (2) When the conditions of a are fulfilled, there exist
values of the rate constants for which some or all of the
Pseudo-First-Order Systems eigenvalues are complex (the real part is always nega-
The details of the pseudo-first-order reactions with tive) and consequently, a damped oscillatory response
can be exhibited.
closed loops (abbreviated PFO-CL) are worth further
scrutiny. For such systems, the eigenvalues may be It should not be concluded that the converse is true.
either all real and negative or partly complex with The existence of closed cycles and violation of the
negative real parts, giving rise to either nonperiodic or Wegscheider conditions does not necessitate an oscilla-
damped oscillatory forms (Equations 4 and 5). tory response. Three types of kinetic response are pos-
Depending on the particular values of the rate constants, sible according to the values of the rate constants: pure
the kinetics may therefore be monotonic, exhibit only monotonic; overshoot-undershoot (including only one
one maxima, exhibit overshoots or undershoots, or extreme); damped oscillatory.
exhibit damped oscillations. In spite of that, based on the previous discussion we

VOL. 59 NO. 5 MAY 1967 25

 may expect that the extrema of oscillatory behavior theoretical conclusion can be summarized as follows:
must occur for such irreversible cycles. It is also clear
Theorem: In order for a system of pseudo-first-order
that the symmetrical cycle (i.e., all rate constants equal)
should provide a minimum possible value of the damping reactions, when perturbed by the addition of only one
constant. These conclusions are easily verified for a chemical, to exhibit any form of oscillatory response,
whether overshoot-undershoot or damped oscillations,
three-component system. then it is necessary that the system possess closed loops.
For the general rcth order, irreversible cycle, with
In no case can such a system exhibit undamped oscilla-
equal rate constants (all k, where k may be taken as
=

tions. (29)
one by appropriate time-scale change) the eigenvalues
are easily derived to be As will be seen in the following sections, any oscillatory
response requires the existence of feedback pathways.
( 2 Closed reaction paths provide one method by which
X* = ( —
A
1 + cos -17r/\ + i such feedback can be introduced.

for l =
0, 1 ...» -
1 (26)
COMPLEX REACTION SYSTEMS
The study of chemical systems involving bimolecular
and, consequently, the damping constant is given by
or other types of complex reactions presents considerably
8 —

In d = 2 7r greater difficulty. The simplicity of analyzing PFO

and FO systems is entirely due to the fact that the
associated differential equations are linear with constant
for l ^ 0, «/2 (27)
coefficients. The bimolecular reactions introduce non-
For l 1 (or —1) the smallest possible magnitude of
= linear differential equations for which there is no known
—

In d is obtained. Note that this represents the smallest method of general analysis. For such systems, the use of
possible value for any rcth order system of pseudo-first- computers, linearized approximations and reduction of
order reactions, including those with reverse reactions. chemical complexity play a necessary and important
For the third-order system (» 3), of 2 X 10~5 = ~
role.
and di/4 0.066; thus, the very least possible damping
=

constant is still extremely large in magnitude and, for all Reduction of Complex Systems
practical purposes, only the first max (or min) can be Consider for a moment a complex system, such as that
observed at 6.6% of a0/3, where a0 is the amplitude of the indicated in Figure 5, involving any number of chemicals
perturbing chemical. As n increases, the damping engaged in first- and second-order reactions. It is
constant decreases and more cycles will be observed. generally possible to reduce the mathematical complexity
For very large », the value of 8 — (2 7r2/n) and it may be
by the following techniques:
verified that a damping factor of 0.9 would require a
(a) Conservation Constraints. Recognition of chemi-
system of about 180 chemical components. cal conservation leads to a reduction in the number of
It is of some interest to compare the true overshoot
dynamically independent variables.
phenomena, for which there can be at most (n 2) (b) Time-Scale Reduction. With single reactions or
—

extrema, with the damped oscillations. For the oscilla- subsets of reactions it is possible to associate a charac-
tory case to exhibit (n 2), extrema requires (» —

2)/2 —

teristic time (t) for the transition from some particular

cycles. Hence, initial state to some other state, such as the stationary
state. Where there are large factors between charac-
(%) (-% -) (28) teristic times for different subsets of reactions, the com-
-*
5(f°r cycles)
plexity of the system can be reduced (75). Thus,
and <7(n_2)/2 5.5 X 10~5 so that the last extremum
=
slowly evolving chemicals can be treated as constant,
would be near the practical limit. The additional which is indicated by a bracket [ ] placed around the
extrema would not be observable. We may conclude chemical. Exceedingly rapid reactions (i.e., small r)
that the distinction between overshoot-undershoot and can be ignored after proper modification of the rate law

damped oscillatory kinetics cannot, in practice, be based with simultaneous reduction of the number of variables.
on the number of observable extrema for PFO. Some A simple enzyme reaction provides one example.
distinction might be possible on the basis of the period The basic chemistry is given as
which is not constant for the overshoot-undershoot case.
Figure 8 shows computer solutions for both cases.
S + E ^ ES
X q p
More important from the standpoint of this paper is
that mechanisms based on first-order or pseudo-first- ES X E + P (30)
y
order reactions cannot, in any realistic sense, explain
the experimentally observed oscillations discussed pre- and requires four independent variables (*, q, p, y)
viously. Recognizing that only one extremum does not representing the concentration of the various species.
an oscillation make and that fulfilling the Wegscheider By conservation, q + p where e is the total enzyme
=
<?,

conditions effectively destroys any closed loops, the concentration and is constant in time. It is shown else-

26 INDUSTRIAL AND ENGINEERING CHEMISTRY

where that the transition time to achieve the steady fact that such variables determine the structure of the
state is usually much smaller than the transition time phase plane for the dynamical variables.
from the steady state to the equilibrium state. Under a
steady-state condition {i.e., p q 0) one obtains
= =
Activation and Inhibition

where Km (31) The concepts of activational and inhibitory reactions

p
_ -
=
k2/k\
x + Km play an important role in the description of complex
and systems, particularly in the discussion of oscillatory
reactions. As discussed in more detail elsewhere (75),
Vmx these terms can be given an analytical definition in
—

x = v =
-, Vm =
k2e (32) reasonable accord with their common
x + Km usage.
In general, if vx is the net rate of production of x
Thus, the variables p and q can be considered as through a particular set of pathways which depend on the
dynamically dependent variables completely determined concentration of other chemicals, say z, then we define
by the dynamically independent variables (in this case, the pathways as “activational in z” (or “inhibitory in z”)
x). Further, the mechanism can be abbreviated accordingly as an increase in z causes an increase (or
4. decrease) in vx with all other variables considered con-
S p (33) stant. Thus, the activation and inhibition refer to the
with the
new rate law given by Equation 32. It should signs of the partial derivatives of the flux functions such
be emphasized that the reduction to steady-state equa- that
tions depends primarily on the relative values of the
characteristic times and not on the steady-state condi- vx is said to be activated by z if > 0
tion {p 0). The exact reduced equations are referred
=

to as characteristic equations and represent limit lines

in the associated phase space. Such equations are only vx is said to be inhibited by z if < 0 (34)
approximated by the steady-state equations, but the
differences are generally small. A complete discussion said to be independent of z if
vx is = 0
of these aspects is in preparation. As discussed else-
where (75), such reduced equations are applicable to
Partial derivatives of vx with respect to x are referred
oscillating systems where the steady-state conditions are to with the additional phrase “self.” Thus, if
apparently violated. In regard to notation, reaction
steps depicted with — imply the mass action rate law; > 0
those with imply rate laws of the form given by
Equation 32 and, more generally, implies an arbitrary then the pathway is said to be self-activating.
rate law as expressed by a specific equation v v{x,y). =
The definition, as given, can be utilized in describing
(c) Self-Consistent Control Approximation. In many specific reaction pathways or the collected sets of all
cases the reduced rate law for a given reaction may be reaction pathways which refer to the net flux. For
independent of certain variables under certain condi- example, the simple first-order reaction
tions. Thus, in the above example, Equation 30 for
x Km, v(x) « Vm, and x is said to be out of control of A B
the flux; in that case the dynamical equations can be for which
further simplified. If the dynamical equations are
solved with such an assumption, and the motion of x is Va =
—V\, Vb
=
+Vi =
kid
always such that x{t) Km, then the approximation is can be characterized by noting that A is self-inhibitory,
considered self-consistent. A reaction for which one of since increasing a increases the rate of removal of a—
the variables is out of flux control is indicated by a slash

through the arrow. In this example, S P.

i.e., (—^
\da )
< 0, but activates b since (—>
\ da)
0.

By employing such approximations, the dynamical As a second example, consider the two reactions
complexity of a mechanism can be reduced. The
variables then divide into three types: the basic, in- [M] + A -i- 2 A
dependent variables; the dependent dynamical variables A C (35)
(such as p, q, above) which do not enter into the reduced
differential equations and whose motion can be deter- Note that the concentration of M is one of the SCV.
mined once the motion of the independent variables is The first reaction is clearly self-activating in a, while
known; the structural control variables (abbreviated the second reaction is self-inhibitory. The net flux,
SGV) which enter the differential equations as param- (va) for production of a is given by
eters, and include the rate constants, all chemicals con-
sidered constant, and conservation condition constants, va =
vi —

v2 =
{k\m —

k2) a (36)
such as e in Equation 32. The name derives from the and can either be self-activating or self-inhibitory,

VOL 59 NO. 5 MAY 1967 27

according to the values of the SCV. definition is convenient in practice. The reader should
The preceding may be considered as a definition of have no difficulty recognizing which is implied.
absolute activation or inhibition. In more common

usage, the terms are applied in a relative sense, usually Dominance Relations
with respect to simple first-order reactions or simple As will be shown, the pertinent mathematical quanti-
enzyme reactions as a standard. In the relative sense ties relate to the absolute activation or inhibition of
those two reactions are neither activational nor inhibitory the net flux. Since net fluxes are derived from a sum-
and would be indicated as mation of single reaction rates, the net activation or
A L B or A 4- B inhibition will also be a summation of the character of
However, the reactions individual reaction steps. The interrelationship be-
tween the net character and the individual reaction
[M ] + A — 2A may be described as
character can often be realized by inspection based on a
act
*r-N knowledge of dominance relations. A list of some
[Ml
L J
A act
specific types of reactions, the condensed reaction nota-
k
tion, and the associated rate laws is given in Table I.
while the reaction A + A B is equivalent to A —> B The number given under Dynamical Characteristic
in the relative sense. While the absolute definition refers to the corresponding graph in Figure 9.
has the greater mathematical bearing, the relative The reader should become familiar with the con-

TABLE I. CHARACTERISTICS OF RATE LAW

Form
Equa- Abbreviated Rate Law {Table In Graph Dominance
Reaction Mechanism Notation v{x,y) II) Variable {Figure 9) Magnitude Remarks
k k
X —k
x kx a * Linear 1

k actn
X + x x —
kx2

act
k k )
[M] + Y —'2 Y [M]
—
Y kmy Linear 1

k
f\act
X+Y-i2Y X —“
Y kxy a x or y Linear 1

fa
X + E —EX
E Vm =
k2e
fa X Y
x + Km
EX —kE + Y Km —

fa/ki

6 As (5) but adjoin 1 0 Fra, Km as in (5)

lr*E
X —'

X + E E+{Ki) Ki =
h/k,
fa x + Km{ + Kix) 1

Dynamically equivalent to Case

5

7 As (5) but adjoin 1 <-» 0 As for (6) but K2 =

k$/k%
x + Km( + Kix +
1 KiK2X2)
(inh)
X + E E+{Ki) See graph D for Figure 9 and
X —v
Y
Case d of Table II for defini-
fa
tion of Xm. Note change of
X + E+ E++{K2)
fa character according to x > or
<Xm

8 As (5) but adjoin inh(EX) 1 <-> 0 Note remarks for Mechanism 7

X + EX ;=± EX+(7fi)
n
X —^
Y
x -f- Km Kix2

9 As (5) but adjoin B 1 <-» 0 Dynamically equivalent to

X —^
rrY Km + x + Ki Km y Case 5

Y + E ^ E+(/G) C 1 <-» 0 Same as Case c with d =

0;
0 < -

\rc' < 1

28 INDUSTRIAL AND ENGINEERING CHEMISTRY

densed reaction notation which, for this paper, will above or below the value (xw) corresponding to the maxi-
always imply the associated rate law and mechanism mum rate.
given in Table I. The basic features of the correspond- Since most chemicals will be engaged in several
ing graphs should also be noticed. They can be divided reactions, the question of whether there is net activation
according to first order (linear); second order (qua- or inhibition becomes important. In general, this will
dratic) ; and zero order (independent of the variable, as in depend on the specific values of the SCV and the values
the case of the simple enzyme reaction for x Km). of the concentrations of the basic dynamical variables.
The double back-activation reaction (Table I, Equation However, for simple reactions, of a sequential form, the
15) exhibits second-, first-, and zero-order regions. The problem of interest centers on the intersection of any
forward inhibition reactions (Table I, Equations 7 and two dynamical characteristics. Such intersections de-
8) have two distinct regions of positive and negative termine the location of the singularity (stationary state)
slope, while the others have only positive slope. Whether as can be realized in the sequence
such characteristics are interpreted as inhibition or Vl(x) V2(x)
activation depends on whether they are involved in the —> X-> (37)
removal or production of the variable as illustrated in where the stationary state requires Vi v2. = The nature
the previous section. Those based on forward inhibi- and behavior of the systems will depend to a large
tion (Table I, Equations 7 and 8) may exhibit either extent on the values of the slopes of the dynamical
character depending on whether the variable (*) is characteristics in the region surrounding the singularity.

Form
Equa- Abbreviated Rate Law (Table In Graph Dominance
tion Reaction Mechanism Notation v(x, y) II) Variable (.Figure 9) Magtitude Remarks

10 As (5) but adjoin inh(ES) b X B 1 <-> 0 See remarks for Case 9

Km + x(l + Ki y)
Y + EX E+(/Ci) X —Y c' y C 1 0

Vmx
11 As (5) but adjoin (inh)2(EX) x Km -}- Kixy -J- K\K2y2
B 1 0

Y + EX ;=± EX+(/Ci) X —^
nY c V C 2 <-» 0

Y + EX+;=±EX++(/C2)

12 As (5) but adjoin act(EX) b x B 1 0 Dynamical equivalent to Case 5

o
X “Y
Km x + —

X + EX+ EX(ivi)

Vmx2
13 As (5) but adjoin act2(EX) E 2^0
1

o
x —»
Y
KaK2
+ *( Km + ~

) + *2

X + EX+ ;=± EX(7Ci)

X + EX++^±EX+(K2)

14 As (5) but adjoin b X B 1 0

act (* + Km)y + Km/Ki
Y + E+ ^ E(Xi) X —Y b y B 1 0

Vmxy2
15 As (5) but adjoin b x B 1 <- 0
{x + Km)f +
^+^
Y + E+ —
E(Ki) E 2 0
Y + E++ —
E+(K2)

16
b x B 1 <-> 0
As (5) but adjoin act(EX)

X
n
E
Y
+ Kmy

Y + EX+— EX(Ei) b B <-> 0

y 1

VOL 59 NO. 5 MAY 1967 29

 For the zzth-order rate law, vn =
K"xn, it is obvious that
1Rn = n (39)
All other equations are treated similarly. Thus,
for the simple enzymatic reaction (Case b, Table II);
the rate law of the form
ax
vE (40)
hx + c

yields
dz;

dx

and consequently
h
Rve (41)
h + cx

where the E refers to equations of the form (40).

The other forms given in Table II can be treated by
the same technique. The results are summarized in
Figure 9. Sketches of the dynamical characteristics associated with Table II where the symbol
different rate laws and comparison with a linear law / slope of rate law, case Q \
Note particularly the possible reactions between slopes at interactions. evaluated
See Table II for particular rate laws associated with graphs (slope of first-order rate law)
at intersection (42)
Dominance relations refer to the net effect in that and Q refers to the case letter associated with a par-
neighborhood. An examination of the graphs (Figure ticular form of rate law as given in Table II. It should
9) shows that this depends on the possible curvature of be noted that the form may relate to either the x or y
the dynamical characteristics. For example, the inter- dependence of the corresponding rate laws (Table I) and
section of a quadratic form with a linear form always that the constants (a, b, c, d) may be functions of the
requires that the slope of the quadratic form will be other variable. While the ratios of slopes are given
greater. Intersection of linear forms with enzymatic with respect to a first-order rate law, the ratio with
forms always requires that the linear form will have the respect to other rate laws is easily obtained from the
greater slope. Reactions involving forward inhibition equation
or back-activation of enzymes are more complicated
and the dominance relations may depend on the location
of the intersection, which is determined by the values of TABLE II
the SCV. In such cases, the maximum possible domi- Form of Rate Law 1R Q Remarks
nance is usually of interest, although the fact that it need (a) v =
kzn lfla = n Independent of rate con-

not necessarily occur should be recognized. Table II stants

gives a listing of the dominance relations and a break- (b) = 0 lRb -such that
lRb -

down of the range of reaction order as indicated by the

1 as z -* 0
arrows. Other reactions, such as third order, can be 0 as z —- °o
lfl6
easily included. cz “f- 2 dz^
lie, 0 “^Rc < 2 such that
While the dominance domains are qualitatively evi- (c) =
v —

-r~7“9
+ N + +
-
°
b cz dz2 b cz dz1
for z 0 0
dent by inspection of the graphs and equations, a precise
—*-
\-rc
1 for z =
sffb/d
lftc =
—

analysis is easily carried out and provides some interest- —

1
Rc
—2 for z —00

ing results. For the first- and second-order rate laws

k'x2 one finds (c') V = —

1/e Same as c except d =

0
vi —

kx, and v2 = b + cz
0 < —

1
Rc, <1 such that
dzu vi bv2 -

lRcf
--
0 for z —
0
,
and = 2 k'x =

~bx —

lae/ 1 for z — co

b —

dz2 0 < |lflj < 1 such that

Since the intersection of the two curves occurs at some (d)
b ~t~ cz -(- dz2 r+cz + dz2 1 Rd
—
+1 for z ^
0

value of x for which v\ v2, the ratio of the slopes at the

=
1^ =
0 for z =
\/b/d =
zm
interaction is given as \Rd = —1 for z — co

az2 2b + cz 0 < lRg < 2 such that

2 (e)
slope 2 (v/x) b -f- cz -f- dz2 b -f- cz + dz2 1 Re 2 for z 0 —- —*•

(38) for
1 1 z =
%/b/d
(v/x)
—

1 ue
slope
\Re
—^
0 for z — co

and does not depend on the values of the rate constants.

30 INDUSTRIAL AND ENGINEERING CHEMISTRY

 ~9 The collection of the signs of the four partial derivatives
=
(43)
PRq 1 RP (Xz,Xy,Yx,Yy) is referred to as the character of the reaction
where P may also be any of the case letters in Table II. scheme. The character may be stated in matrix form
It must be recognized that the values of the constants with the signs of the partial derivatives substituted for
the partial derivatives in the matrix:
(a, b, c, d) must be taken as different for the different
cases (jP and Q). Values of the derivatives are easily Xy\
obtained from the table as (Xx
\ Yx YJ
(lfi9)
~
(44) Thus the character matrix for Xx and Yy positive, and
dz z
Xy and Yx negative is
As an application consider the reaction sequence
*'

„act

X 4Ty -i
\\ (45)
where both subtractions are listed as first order in the
(- +)
It important recognize that the partial derivatives
is to
table. Whether activation or inhibition dominates for themselves depend on the particular values of x and y.
Y requires closer analysis. For the sequence Their signs, and hence the character, may change in
different regions of the (x,y) phase space, nevertheless,
x (46)—
it is possible to divide the phase space into regions which
it is clear from the table that the self-activation of Y have specific character.
will always dominate its self-inhibition, thus giving a The character of a reaction which has been reduced
net activation. Other examples will be treated in later to a two variable problem may be conveniently and in-
discussions. tuitively expressed by a net flux diagram as follows:
GENERALIZED TWO-BODY OSCILLATORY
MECHANISM
Consider now that a complex system of reactions has
been reduced to a two-body system—i.e., only two in-
dependent dynamical variables. Let the two variables
be labeled x and y. Depending on the complexity of the
unreduced systerh, the differential equations for x and y where X and F represent the chemicals, and the large
may be quite arbitrary and take the functional form arrows represent the collected sets of pathways (in the
x X(x, y)
=
vx = original complex mechanism) which produce and re-
move the chemicals. Associated with each large arrow
y=Vy= F(x, y) (47) is a net flux function, vx and vyy respectively. The small
typical of a general two-body problem. However, theif arrows from X and F directed toward the large arrows
original chemistry was based on the mass action law, the indicate influence of chemicals X and F on the net
functions X and Y will be continuous and differentiable fluxes. The terms act and inh written on the small ar-
(in the positive x, y quadrant), frequently taking the rows stand for activational and inhibitional with respect
form of a rational function, though more general alge- to the net fluxes. For such diagrams, these terms are
braic forms may result. When derived from chemistry,
only employed in the absolute sense and are, therefore,
however, these functions retain one very important equivalent to the signs of the partial derivatives
property namely: (X„ X„ Y„ Yv).
F(x, 0) £ 0; X(09y) > (48) 0
Theorems on Nonlinear Oscillations
A proof follows directly from the assumption that the
rate of removal of a substance (say X) by any single In view of the theorems and discussion to follow,
reaction step approaches zero as the concentration (x) it is important to recall some basic aspects of singularities
approaches zero. Since the net rate vx 2^in = = —
and their properties. The singularities of a system
2^out> and the sum of removal rates vanished as x — 0, occur for those values of x and y which make the fluxes
then the net rate, vX9 will be necessarily positive or zero. vanish (i.e., vx vy =
0); we denote these values by
—

This property provides that the motion of the variables x* and y* and recognize that there may be more than
is confined to the positive quadrant if the initial values one singularity for a given system. Expansion of the
of the variables are in the positive quadrant; that is, differential equations around the singularity and neglect
the concentrations of the chemicals can not go negative of higher derivatives leads to a linear system whose co-
for any realizable reactions. efficients are the values of the partial derivatives evalu-
The kinetic behavior is strongly dependent on the ated at the singularity. The treatment of such systems
is the same as given in conjunction with the first-order
partial derivatives of the net flux (vx,vy) defined as fol-
lows (see also Equation 48): reactions and the eigenvalue equations are given as

X* =
&X/dx), % =
(dZ/dy), Yx =
(dF/dy), X2 -

B*\ + A* = 0 (50)

yv =
(dr/d>o where B =
Xx + Yy and A =
(.XxYy
—

XyYx) are

VOL 59 NO. 5 MAY 1967 31

general functions, but the asterisk (*) indicates evalua- In effect this theorem requires that the system ex-
tion at the singularity. hibit the character depicted in the diagram of Theorem
The value of the X’s determines the behavior only 53, but in an average sense. Thus, if a system possesses
in the small neighborhood of the singularity. Thus, if two regions of the form
both X’s have negative real parts, the singularity is
stable, while for positive real parts the singularity
is unstable. If the X’s are pure imaginary then, to
a first approximation the trajectories form closed cycles

(oscillations) around the singularity. Most previous

studies of oscillatory mechanisms have been limited to
these stability properties. However, it must be realized
that in general the stability properties of the singularity
are neither necessary nor sufficient for the existence of
sustained oscillations. For nonlinear systems, such then while neither region alone could possess a closed
oscillations are usually associated with limit cycles and
trajectory, the conditions of Theorem 54 would be satis-
consequently require an examination of the behavior fied for a trajectory passing through both regions. As
in the large. with the Bendixson Theorem, both Theorems 53 and
The analysis of nonlinear systems is severely limited. 54 may be referred to as negative criteria since they are
However, five pertinent theorems are presented below; more useful in eliminating certain mechanisms. Satis-
two are well known and the others are proved in the
faction of the criteria does not guarantee the existence of a
appendices. closed trajectory; nor are the necessary conditions given
Poincare Theorem: If C is a closed trajectory (in by the theorem to be considered the strongest which may
the phase plane), then it must contain at least one be imposed. However, as will be demonstrated below,
singular point, the index of which is +1. For the the theorems are easily applied and provide considerable
two-body problem given here, the index is +1 if insight into the requirements for oscillatory reactions.
A* > 0 and ( 1) if A* < 0.
—

(51) A very weak theorem, concerning the stability proper-

The proof of this theorem is given elsewhere (10): ties of the singularity can be given.

Theorem: If
no antilimit cycle (a limit cycle as t —
Bendixson Theorem (10): If C is a closed trajectory
surrounding the region R, then the function B =
—

exists, then the existence of a limit cycle

oo)
(Xx + Yy) must be either zero or change sign in R. requires that the singularity be unstable—i.e., that B*
(52)
=
(Xx* + Yy*) £ 0. .

(56)
An extension of the preceding theorems, of particular The theorem is trivial and self-evident. While it will
interest here is provided by the following: be applied to some of the mechanisms which follow, the
Theorem: If a region, R, of definite character con- premise of no antilimit cycles will simply be an assump-
tains a singularity and a closed trajectory surrounding tion. Prigogine’s statement that rotations are not
the singularity, then permitted around the equilibrium states requires further
consideration. The current proof rests on the thermo-
(a) The character of the region must be of the
following type dynamics of irreversible processes and the Onsager
relations which are only applicable in the small neighbor-
hood of the equilibrium state. With such assumptions,
it is easily demonstrated that the equilibrium state is
always a stable singularity. While it appears physically
reasonable that Prigogine’s statement should be appli-
cable on the whole, I am not aware of any general proof.
The possibility that the situation illustrated in Figure 10

that is the self-coupling must be of opposite character

and the cross-coupling must be of opposite character.
(b) In addition, the magnitude of the product of
the cross-coupling terms must be greater than the ANTI-LIMIT CYCLE

magnitude of the product of self-coupling terms—i.e.,

LIMIT CYCLE
\XW Yx| >\XxYtI (53) STABLE SINGULARITY

Theorem: If C is a closed trajectory (simple, smooth,

closed curve), it must
(a) Encompass a region, R, which contains sub-
regions such that Xx and Yv have opposite signs and Figure 10. Graphical example of a system with a stable singularity
(b) Must pass through regions for which Xy and which possesses a limit cycle. Whether such phase plane structure
Yx have opposite signs. (54) can occur for chemical systems remains an open question.

32 INDUSTRIAL AND ENGINEERING CHEMISTRY

could exist for certain mechanisms seems to be left open. motion (of y) is always toward this stable point. If the
Such an assumption seems reasonable for relatively self-coupling terms are zero and the cross-coupling
simple mechanisms, but could be wrong for more com- terms have opposite character [see (60)], then concentric
plicated ones. Examples of nonlinear systems with closed trajectories occur in the phase plane and the
stable singularities and limit cycles are easily generated system exhibits semistable oscillations, as shown in Figure
as can be seen in Figure 10. In general, the proof that a 1 for example.

limit cycle actually exists for a given set of equations A system with the form
is not trivial, but can in some cases be carried out by
utilizing the Poincare-Bendixson theorem which follows:
Theorem of Poincare-Bendixson {19). If two simple
closed curves C\ and C2 be obtained such that C\
can
is contained entirely within C2; and such that there
are no singularities in the region D between C\ and

C2; and such that trajectories enter D through every

in which the self-coupling terms are both inhibitional
point C1 and C2, then at least one closed trajectory
(limit cycle) exists. (57)
will typically yield damped oscillations if the cross-
coupling is sufficiently strong (see Figure MB). If
For the studies which follow it is somewhat more the self-coupling terms are both activational, then un-
convenient to use the following corollary which is obvious
damped (increasing) oscillations can result.
from the preceding theorem. If both cross-coupling terms have the same sign, such as
Corollary: If a simple closed curve, C2, can be ob-
tained which surrounds one unstable singularity and
such that all trajectories which cut (or touch) C2
are directed toward the interior (or tangent to C2),
then at least one closed trajectory exists (58)
Unstable is here used in the sense that all trajectories in
the neighborhood of the singularity leave the singularity;
hence, saddle points are excluded as they must be then the singularity will be a nodal point (i.e., no
according to the Poincare Theorem 51. oscillations) if A* > 0 and will be stable or unstable
according to whether the self-coupling terms are both
Discussion of the Generalized Mechanism
inhibitory or activational, respectively.
By comparison to other two-body mechanisms, a great As can be recognized from these results, rotations
deal of insight can be gained as to the necessary condi- derive primarily from the cross-coupling terms and re-
tions for sustained oscillations as given in Theorem 53. quire that they have opposite character. Whether the
The reader should be aware that the following behavior rotations are damped or undamped depends primarily
patterns are only typical of regions of definite character on the self-coupling terms—undamped if both are
with one singularity. Some additional mathematical activational and damped if both are inhibitory; a
conditions and a more complete analysis of both one- stationary rotation (limit cycle) requires that the self-
and two-body problems are in preparation. coupling terms also have opposite character somewhere
In case the cross-coupling terms {Xy and Yx) vanish, in the region. Interference between the self-coupling
the two-body problem reduces to two separate one-body and cross-coupling terms is also present and leads, in
problems which may take the form as indicated by the effect, to the condition that the product of the cross-
diagram coupling strengths be greater than the product of the
'>x self-coupling strengths (at the singularity) if rotations are
to occur.
(59)
For regions of multiple character, the behavior is
generally more complicated. Nevertheless, as can be
TVac seen from Theorem 54, the same basic elements are
For such systems the time-dependent behavior is obvious, required; namely, regions of opposite cross-coupling and
since X and Y are independent of each other. For the regions of opposite self-coupling. In one sense, Theorem
54 can be interpreted as requiring that the character
activational case, the singularity is unstable and the
variable (*) increases or decreases indefinitely. The diagram of Theorem 53 holds “on the average.” Thus,
for sustained oscillations to exist in the region of multiple
inhibitional case provides a stable singularity and the
character it must be possible to obtain the requisite
diagram by an effective “superposition” of regions,
While there is no proof, it seems likely that the product
of cross-coupling strength must be greater than the self-
coupling strengths in some average sense over the region
enclosed by the limit cycle. Unfortunately, the exact

VOL 59 NO. 5 MAY 1967 33

 For a region of definite character, the direction of rotation
is easily shown to obey the following rule:
Rotation Rule: If x(t) is plotted on the vertical axis
andy(/) the horizontal axis (positive to the right),
on
then the direction of rotation depends only on the
cross-coupling terms according to:
(1) Counterclockwise rotation for Yx negative (inh)
and Xy positive (act)
(2) Clockwise rotation for Yx positive (act) and
Xy negative (inh) (63)
The theorem is similar to, but somewhat more general
than, Prigogine’s theorem (20) since it does not depend
on the values of the self-coupling terms and is not re-
stricted to small neighborhoods around the singularity.
Analysis restricted to the neighborhood of the singu-
larity with x(t) A sin cot and y(t) ^
B sin (cot + 0) ^

«"\aAaaa provides the following results

tan 0 = —

co/Xx =
co/Yy (64)

TIME- 4.25 sk. where it is clearly necessary to have (Xx + Yy) 0. =

However, we may. expect, that the result is valid, at least

as to sign, if Xx and Yv are interpreted in some average
20 F6P 10 20 sense. By use of the descriptive analysis given pre-
viously, the following table can be established by the
D
preceding theorems:
FDP
0 0 1 Direc- Direction
0 TIME- 8.5 sw. 10 0 tion of of Major
-FDP (Xy, Yy) (Xy, Yy) e Rotation Axis
>

Figure 11. Computer solutions for back-activation oscillator. Values + ~

+ —

7r/2 < 0 < 7r i u Slope

—

V •§
of SCV were chosen to give: A—stable, nodal singularity; B— >
5
damped oscillations; C—nearly sinusoidal oscillations; D—nonlinear “

+ + —

0 < 0 < 7r/2 3

O
O
O
oscillations U o
Slope +
+ ~ —

T —7r/2 <C 0 < 0 V

Slope +
k oact
E 1
Ex
FDP + E*
5
[GLU] F6P - FDP -* >

FDP + Ei ->
6
EX
~

+ —

T —7r < 0 <C —7T /2 Zg

O Slope
—

O
7 7
[GLU] -> F6P EX Ex
(65)
F6P + eXeS Additional formulas valid only in the neighborhood
3 of the singularity can be derived; they may, however, be
ES -> Ey + FDP
taken as very crude first-order approximations to the
A. Ky 0.0658, K2
actual properties. The frequency of sustained oscilla-
=
0.9, Kz 0.9, K4 = = =
0.9, Kg 0.9, Kg
= =

0.9, K'i 0.07. Scales: F6P, FDP

= = 5. All enzymes = 7. tions is given by:
T 7 sec. [GLU]
=
1.0 =

B. Same as A except for Ky =

0.04, Kg =
0.63. T =
4.25 sec. CO = vV =
A/(XxYy -

XvYx)* (66)
C. Same A except for Ky 0.06, K2 since B*
1.0, Kz 0.2, K4 0, if the oscillations occur in
as 0.8, (Xy* + Yy*)
= = —
=
= =

Kg =
1.0, Kg 0.3, =
Ki =
0.1. Scales: F6P, FDP =
10
T = 4.25 sec. the neighborhood of the singularity. If damped (or
D. Same as A except for Ky 0.01, K2 1.0, Kz 0.3, K4
=
0.3 = = = undamped) oscillations occur, the damping constant is
Kg =
0.3, K’j 0.03. Scales: F6P
=
20, FDP 70, T = = =
given as
8.5 sec.
-tr B*
8 (67)
a;
nature of the average is not clear and appears to relate
to integral conditions. where should preferably be measured or approximated
by co =
[(B*/2)2 A*]1/2.
-

Properties of the Oscillations While the limitations of these latter formulas are
The properties of the oscillations, if they occur, can be severe, it should be realized that correct procedures,
related to certain features of the generalized mechanism. such as the method of Kryloff-Bogoliuboff, are difficult

34 INDUSTRIAL AND ENGINEERING CHEMISTRY

to apply and generally lead to extremely complex (Table I) either one could actually dominate depending
formulas {19). on the value of y and the rate constants. Similarly, for
the self-coupling ofy. The net flux diagram appears as
APPLICATIONS OF THE THEORY
Simple Examples
Theorems 53 and 54 provide a convenient first test,
usually applicable by inspection, as to the potential (72)
oscillatory character of a mechanism. As a first ex-
ample, consider the mechanism
0 1 2

S
--
X —
Y —
(68)
and the positive x, y quadrant has multiple character
with the two variables x and y. Step 0 plays no role as shown in Figure 12. Since regions of opposite self-
except to provide a stationary state. For fixed y, an
increase of x increases the net rate of production of y
coupling exit, by Theorem 54, the necessary character is
fulfilled and the mechanism may exhibit oscillations
while increasing its own net rate of removal through
(as it does).
Step 1. Hence, x activates y and is self-inhibitory in the A nearly direct interpretation of the oscillatory charac-
absolute sense. Similarly, y is self-inhibitory through ter diagram is provided by the Jacob and Monod
both Steps —1 and 2 and activates * through Step 1. —

mechanism {12) for genetic-metabolic control. In a

The resulting net flux diagram is
very abbreviated form the mechanism appears as

[Amino Acid Pool]--- —

Enzyme (-!—-> (73a)
\ /activational

/ \ inhibitory
\
(—L.)
[Metabolic M]--- — Met N (73b)
Since this character prevails throughout the entire
positive (*, y) quadrant, it is clear from Theorem 53 The genes which catalyze the production of enzymes
that such a mechanism can never oscillate. As in- from the amino acid pool can be repressed—i.e., in-
dicated in the previous discussion, if A* > 0 (as it is) hibited by certain metabolites, say N. The enzymes
then the singularity is of nodal character and the motion catalyze the conversion of metabolites. For the typical
is directed toward the singularity—i.e., no rotations of rate laws involved, such a mechanism has cross-coupling
any type. It should also be noted that modification of opposite character and if the decay reactions (73a
of the simple mass action rate laws (e.g., v\ kx) to = and 73b) are ignored—i.e., have fixed rates independent
more general monotonic forms [z>i f{x) ] will not alter
= of the concentrations e and n—semistable oscillations
these conclusions, since would be obtained. If typical decay reactions are
included, then only damped oscillations would result.
A XXYy XyYX Vl\x[v-1\v +
~
= -

v%\i/]

Vl\xV-l\y
=
Vi\xV2\y > 0 (70) General Results for Sequential Reactions
where While the generalized oscillatory mechanism is sym-
V\\x
=
&»!/** metrical in the variables X and F, such symmetry is
The Lotka mechanism provides another simple and lost for sequential systems where X is presumed to precede
illustrative example. From the reaction F. Altogether there exist four distinct diagrams;

[G] + X -V 2X (71a)
X + Y 2Y (71b)
Y ^ (71c)
it is clear that an increase of x for fixed y always increases
the net rate of production of y through Reaction 71b.
Hence, x always activates y. Again by Reaction 71b,
we observe that y always inhibits x. Thus, the cross-
coupling is of opposite character throughout the positive
x,y quadrant and the self-coupling is more complicated.
For fixed y, an increase of x will provide self-activation
by Reaction 71a and self-inhibition through Reaction
71b. Both the self-activation and self-inhibition are
first order. In accordance with the dominance relations

VOL. 5 9 NO. 5 MAY 1967 35

 where act and inh notation is in the relative sense and the
reactions are arbitrary. Either situation can yield the
appropriate diagrams. Thus, in the case of back-
yi ys Y
activation, Ai is self-inhibitory but activates A2 while A2
activates the removal of Ai and hence inhibits the net
Mnh flux for Ai while activating its own net production.
y y* Such a back-activation of the net reaction rate can be
achieved either by activating the forward reaction rate
or inhibiting the back-reaction rate as:

Ol
or
o
X* (78)
X -

[s]
Figure 12. Multiple character of phase plane for Lotka mechanism.
Cross-coupling terms are definite throughout entire plane while self- Similar methods can be used for achieving net forward
coupling terms depend on location. Regions of opposite self-coupling inhibition.
(for Xx and Yv) do exist
Again it should be noted that these modes of feedback
which can be associated with are opposite to the normal behavior of simple reactions

(see Equation 68). When feedback is provided in the

[G] X Y (75) usual sense, as with forward activation or back-inhibition
Since the basic Theorems 53 and 54 only provide a of the net reaction step rate, then the system will not
set of necessary conditions, it is not clear a priori that exhibit an oscillatory response. In that case, depending
their fulfillment will actually lead to sustained oscilla- on the specific nature of the feedback, the system may
tions. In addition, it is not clear that chemical mech- either remain stable or become unstable.
anisms composed of realistic reactions can actually The preceding conclusions are applicable to sequences
be associated with each different diagram. The estab- involving any number of chemicals. It need only be
lishment of “effective sufficiency” and realistic models remarked that the feedback does not need to be applied
for the different diagrams is, of course, most pertinent at the step directly preceding or following a given chemi-
to the usefulness of the theory. The four possible cal. Thus, the following sequence
diagrams divide into three effective problems. Dia-
r i
~

grams a and b differ in the interchange of X and Y.

If either one can be established, the remaining one is [S]
—
At A2
—
A3 A4
^
As (79)
then a problem relating only to the realistic sequential or
chemistry. Diagrams c and d differ as a class from a
and b in that the self-coupling and cross-coupling charac-
ter are the same (as c and d) or opposite (a and b) for ^ — -- ^
[S] Ax A2 A3 A4 A5
the variables; again c and d differ only in the inter-
change of X and Y. can have a potential oscillatory character. A fuller
Some general requirements for the occurrence of discussion of such a system, as well as those involving
oscillations in sequential reactions are easily derived crossed feedback loops is in preparation.
from the generalized schemes. In the previous section, Again, the Lotka mechanism provides an interesting
it was shown that the simple sequence example. In the notation used here it appears as
act act
0 1 2
[S]
—
A2
—
(76) [G] A ^
B (80)
Ax^±
and the back-activation is evident. The character of
can exhibit an oscillatory response of any type even
not this reaction can be viewed as a superposition of the
when monotonic but otherwise arbitrary rate laws are basic diagrams (Equations 74a and 74d).
used. There are two evident chemical means by which the
To obtain the appropriate oscillatory diagram for feedback may be established. In the Lotka mechanism,
sequential reactions, it is necessary to provide feedback it is through the use of bimolecular reactions in which
contrary to the normal situation. This can be achieved a chemical down the sequence is also engaged in
by either back-activation or forward inhibition of the a preceding step. We shall refer to this method as
net reaction step flux as indicated by stoichiometric feedback in view of the fact that
the feedback requires stoichiometric utilization of
-Ai ^\act
-A2 -*-
(back activation) the feedback chemical. A second type is based on
inh f *\ (77) the inhibition or activation of specific catalysts (such
-
Ai —.-A2 -(forward inhibition)
as enzymes) by-products or substrates involved in

36 INDUSTRIAL AND ENGINEERING CHEMISTRY

the reaction sequence. That type will be referred to as the enzyme phosphofructokinase and its reactions as
nonstoichiometric since it is only necessary that catalytic being the source of oscillations. It is known that the
amounts of the feedback chemical be bound. Both reaction product (FDP) can activate the enzyme and
types of feedback occur in biological reactions as can hence the condition for back-activation is satisfied.
be seen in Figure 5. Mechanisms based on both types Regardless of the validity of this interpretation (77), the
are discussed in the applications below. mechanism is of theoretical interest and illustrates the
A general result which finds frequent applications to application of the preceding theorems.
sequential mechanisms can be derived and will be In abbreviated notation, the mechanism may be
referred to as the “two-variable sum rule.” stated as

If the of vx + vy
sum X + Y is a function
=
<t> k E
act
E
)
of y alone, then the sign of A is opposite to the product [GLU] F6P FDP (83)
1*2
of the signs of Yx and <fy (81) y 3

where the detailed chemistry and the rate laws are given
Proof: Since X + Y <£(y), then Xx + Yx 0
= =

in Table I.
and Xy + Yy <j>y. Substituting
=
these equations in The pertinent equations are
A XxYy XyYx one obtains A Yx<t>y and the
= —
= —

rule is established. If 0 is a function of x alone, then x =

vi —

v2(x, y) =
X(x, y)
A —Xy(j)x. Application of the sum rule derives
=

y =
Vi(x, y) —

vs(y) =
Y(x,y) (84)
from the fact that A* must be positive to satisfy the
Poincare Theorem 51. The associated flux diagram can be realized by inspection
of the reaction system to be of the form
As it will be assumed that antilimit cycles do not
occur for the mechanisms treated here, the condition
that B* (Xx* + Yy*) >0 will be necessary to obtain
=

oscillations (see Corollary 58). That condition usually

leads to a complicated inequality among the SCV (see
Appendix II). It should be noted that while the
dominance relations can determine the sign of X* or
Yy* separately, such relations can not determine the
signs of the sum (.Xx* + Yy*) since the derivatives are where each of the partial derivatives except Yy exhibits
with respect to different variables. However, in most
definite character over the entire positive quadrant;
chemical systems, the parameters enter in such a way as
Yy may be either self-activating or self-inhibitory. The
to allow independent scales between the variables. In
net self-character derives from the difference between
regard to the graphs of Figure 9, the z axis, may be the back-activation (Reaction Step 2) and the forward
stretched by proper choice of the SCV. In analytical
inhibition (Step 3).
terms, the flux functions X are of the form X(ax) so To achieve the appropriate oscillatory diagram (of the
that Xx aXx where x'
= 1
ax and a is a function of the
=

form 74a), it is necessary to have the self-activation of Y

SCV. Consequently, the derivatives of the flux func-
dominate at the singularity. The characteristic curves
tion with respect to one variable (say x) can have any
are both of the enzymatic form (Figure 9B). From the
relation (in magnitude) to the derivative of the flux
dominance relations (Table II), it is clear that the back-
function with respect to the other variable (y). Hence
activation can overcome the forward inhibition by
we may state
B* Rule of Thumb: If either X* or Yy* can be choosing rate constants such that Step 3 is nearly zero
order while Step 2 is nearly first order (as a function of
made positive, then B* X* + Y * can also be
=

made positive for some choice of the SCV. y) at the singularity. In accordance with the B* Rule
(82) of Thumb 82 the singularity can be made unstable
Mechanisms with special constraints between the [B* =
(Zx* + Yv*) >0].
variables, or special conditions on the parameters may As just illustrated, inspection of a mechanism based
well provide exceptions to the rule, but they are not on familiarity with the dominance relations provides
encountered in this study. a qualitative approach to the possible behavior of a

system. However, it is often useful to carry out a

A Back-Activation Oscillator simple analysis. Thus, one easily determines that
In the light of the preceding results, the mechanism Xx Xy
^2jx
of glycolysis can be examined as a basis for the experi-
mentally observed oscillations in the glycolytic inter- Yx =
02jx Yy =
V2\y ^3| y (86)
mediates. It provides several possibilities which can be where the subscripts (x, y) refer to partial derivatives
based on stoichiometric feedback as with ATP and
(77). Hence,
ADP or on nonstoichiometric feedback through the
action of the metabolites on the enzymes. Both experi- A =
XxYy -

XyYx =
.2,^3,, > o (87)
mental and theoretical results have directed attention to so that A is always positive, and consequently A* also.

VOL. 59 NO. 5 MAY 1967 37

Hence, the Poincare condition (Equation 51) is satis- Other mechanisms based on back-activation are easily
fied. From the conditions that examined. For example, if the last step is replaced by
a first-order reaction as
B* =
X* + Yy* =
( —

V2]x + V2]v
-

V2]v)* > 0 (88)

it is realized that k
£\ac'i
[G] -iX-'Y
J

< (93)
^3]y V2\y (89) 1 2 3

is required for oscillations as was evident by inspection.

then in view of the dominance relations, it is clear that
The rotational condition,
the self-character of Y will always be inhibitory since
\XyYx\ > \XZYV\ (90) the first-order Reaction 3 will always dominate the single
back-activation. Consequently, the system can at most
leads to the requirement that
exhibit damped oscillations (similar to Figure 12Z?), for
Vl\v > \vz\v
—

Vz\v\ (91) any settings of the SCV.

If the back-activation is made stronger as in
which is also satisfied when condition 89 is fulfilled.
(act)2
That such inequalities can actually be satisfied for some k E \ k
values of rate constants and input flux (v\) is clear from [G]
--
X —
Y —
(94)
2 3
the characteristic curves. One additional remark 1

concerns the existence of the singularity. From the then once again the system can exhibit sustained oscilla-
tions since the quadratic region of the double back-
dynamical characteristics (Figure 9) it is clear that
activation can dominate the first-order inhibition of
vi < Vmz and < Vm2 (92)
Step 3 so that Yy can again be net-activational.
in order that the intersection exist and consequently
that there be a singularity in the positive quadrant. Forward Inhibition Oscillators
If either of these inequalities is violated, reaction splitting
The preceding section has given examples of one type
(75) occurs and one or the other or both of the variables of oscillatory diagram (Equation 74a). To obtain the
(x ory) will increase indefinitely. second type (74b),
Because of the experimental and theoretical interest
in this particular mechanism, a more detailed mathe-
matical analysis is given in Appendix II. Additional
conditions on the values of the SCV and an analytical
proof that a limit cycle can exist are included. How-
ever, the preceding analysis has already made it evident
that the system can exhibit a variety of kinetic behavior
patterns according to the values of the SCV and in a
qualitative sense, the interrelationship between the
kinetics and the values of the SCV is evident from the for a sequential system, requires some form of forward
dominance relations. inhibition. That such mechanisms are necessarily more
Figure 12 shows computer solutions for different involved than those based on back-activation is easily
settings of the SCV. In Figure \2A the back-activation realized. For example, consider
has been largely removed and the system behaves as a
simple sequence of enzymatically catalyzed reactions
exhibiting a stable singularity of nodal point character.
[G]~-X^Y—
1 2 3
(96)
For Figure 12B, the back-activation is present but Y where the second step can be taken as any form of
(FDP) has net self-inhibition due to the third step being forward inhibition—e.g., Table I,-Cases 7 and 8. Such
more in the first-order region relative to the back-activa- a mechanism can have the appropriate self-coupling and
tion; consequently, the system exhibits only damped inhibitional cross-coupling of X to Y. However, there
oscillations. In Figure 12, C and D, conditions for sus- is no way for y to activate X (i.e., Xy 0); conse- —

tained, limit cycle, oscillations are obtained. The quently, the product of the cross-coupling strength
oscillations are strongly nonlinear when 5* is large and (Xv Yx) is zero and the system can not oscillate. The
positive (as in D) and becomes more nearly sinusoidal actual behavior of this mechanism (96) is readily ob-
and-of smaller amplitude when B* approaches zero as in tained from the shape of the dynamical characteristic
C. Figure 12D shows two different sets of initial condi- and the fact that the motion of X is independent of Y.
tions in the phase plane and the associated approach to There are two singularities in the positive quadrant of
the limit cycle. Note that the direction of rotation the phase plane, one being stable and the other un-
is clockwise and the slope of the major axis is negative stable, as schematically illustrated by a one-dimensional
in accord with Equation 65. The rotational direction phase plane.
and the slope of the major axis as actually seen in all 1st singularity 2nd singularity

computer solutions must be inverted for comparison to 0—^ (97)

Equation 65, because the positive axis is directed to the
left.

38 INDUSTRIAL AND ENGINEERING CHEMISTRY

For initial values of x less than x2, the system * moves which can be written as

toward the first (stable) singularity (x xh y y i). = =

(inh)2
While for x > x2, x continues to increase indefinitely and k E
f~^E
y — 0. [Gl^X-Y- (103)
There are several methods by which y can be made lj 2
|
3

to activate x, although they do not always lead to an

oscillatory mechanism. For example, consider the
general mechanism Even though the mechanisms given in this section in-
inh volve back-activation, they should not be confused
P3 with the mechanism of the previous section (83). If
[GJ-X-Y- (98) the forward inhibition in step 2 is removed, the system
will not oscillate even though the back-activation to
where v\ and v± are monotonic, but otherwise arbitrary Step 1 is present. The conclusion is evident from the
net flux diagram which cannot then exhibit opposite
functions, vz may be of the form given by Table I (Cases
7, 8) and v2(y) may be just a simple first-order reaction cross-coupling. In comparison with the previous mech-
anism (101), the reader may easily verify that the
or some complicated monotonic function arising from
reduced equations. The phase plane typically divides production of 2Y at Step 2 not only provides a nontrivial
into two regions of definite character as stationary state but also makes the sum rule inapplicable
since S =
v2(x) a3(y). The final step (3) must be of
—

an enzymatic type for two reasons. First, the existence

of the singularity requires v\ vz; if these are both =

first-order reactions then y* 0 or k\g kz- The = =

former solution gives trivial results and the latter is

physically unrealistic. When the third step is enzy-
matic, realizable values of the singularities exist. The
second reason derives from the rotational requirement
that |Xy Yx| > \Xx Yy\. From the net flux equations
x =
Vi 00 —

v2(x)

y —
2 v2{x) —

v\(y) —

z>3(y) (104)
and each region has its own singularity. The net
fluxes given by, one finds that

x =
vi —

vz(x) + v2(x) [A^T^j is to |AAEf/| as \vi\y\ is to |p3|V| (105)

y =
vz(x) —

v2{y)
—

Vi(y) (100) Hence, to satisfy the rotational requirement, v\ must

dominate vz. If v\ is made first order, then vz must be
sum toA=x + y ^i =
v%(y) (y) < 0, application
—
=

capable of lower order, such as occurs for an enzymatic

<j>

of the sum rule shows that A < 0 when the forward reaction.
inhibition is active—i.e., region II, where Yx < 0. In
Computer solutions for this mechanism are shown
that case the singularity has saddlepoint character and in Figure 13 for conditions yielding sustained oscilla-
the system cannot oscillate by the Poincare Theorem. tions. In accordance with Equation 66, the rotational
In Region I, the system is stable and has nodal point direction is counterclockwise; opposite to that for the
character (see Equation 63). back-activation oscillator (83), and the slope of the
As another example, consider the mechanism
major axis is negative. In A, nearly sinusoidal oscilla-
tions are obtained, while highly nonlinear oscillations
(101) are shown in B. The inflection in Y during the rising
portion should be noted; by proper adjustment of the
SCV, the duration of the inflection can be prolonged.
The other forms of inhibition (Table I, Case 8) may
H also be used.
where the step is required in order to provide a non- A mechanism having the same oscillatory diagrams,
trivial stationary state. While the net flux diagram but based on enzymatic reactions alone is easily con-
again possesses the appropriate oscillatory character, a structed as
similar application of the sum rule proves that the
inh(ES)
mechanism can not oscillate.
Mechanisms which can oscillate are easily con-
£i
*
fl Y
Ej

y
-
[G] (106)
structed, as for example,
(act) or (act)2
(inh)2—
\ E E where the activation may be single or double, either
[G]^x4y- (102) case yielding sustained oscillations for appropriate
ranges of the SCV.

VOL. 59 NO. 5 MAY 1967 39

 Inh.’(ES)
Step 1 of Mechanism 106 leads to essentially the same
-*• Y
E,
results since the other reaction steps are enzymatic and
[G]-^X
-

similar in both mechanisms. Computer solutions for

sustained oscillations when the double back-activation is
present are shown in Figure 15.
The double periodicity in y (Figure 15A) provides an
interesting feature. As can be seen in the associated
phase plane, the double periodicity is associated with a
crossing in the trajectory. In addition, the rotational
direction changes from that of the overall trajectory
(counterclockwise) to the opposite (clockwise). Such a
rotation is not, in principle, possible for this particular,
two variable, mechanism. While the phase plane has
regions of multiple character, y is always activational
on x, and the appropriate cross-coupling (see Theorem

66) for the clockwise rotation can never be achieved.

The clue to the behavior derives from the well known
0 TIME-- 17 sec. 10 0 fact that a two-variable system can never cross itself
in the phase plane. For the two-variable autonomous
Figure 13. Computer solutions for forward inhibition mechanism
equations which satisfy a Lipschitz condition, crossing
Values of SCV was chosen to give: A—nearly sinusoidal oscillations; in the phase plane would violate the necessary unique-
B—highly nonlinear oscillations. Induction occurs during rising portion ness of the solutions to the differential equations. Con-
of Y(t) curve

The mechanism is: sequently, additional independent variables must exist

which, in this example, must be one or several of the
6
enzymatic intermediates. Values of the rate constants
7

[GLU] + Y -> X EX++ EX+ + X

2
7 which provide the doubly periodic kinetics are given
X + E -> EX + Y 8 in the figure. Additional computer studies have
Y + El -> EiY
3
EX -> E + Y 9 Inh(ES)
E\Y -> E,
4
EX+ z=± EX + X
5

and the appropriate constants are:

A. Ki 0.06, K2
=
0.9, Kz 0.2, if4 =
0.04, Kb 0.9, K6 = = = =

0.02, K7 0.9, Ks =
0.9, K9 0.017. Scales: x,y
= 10. = =

All enzymes 1. T 14 sec.

=
[GLU] 1.0 = =

B. Same as A except for Ks =

0.3, K9 =
0.01, K9 = 0.0148. Scales:
x =
20, y 10
=

This particular mechanism possesses some interesting

features. The back-activation of Y on Step 1 only
affects the cross-coupling terms and does not enter into
the self-coupling terms since Y is not utilized. Con-
sequently, increasing the strength of that activation
from zero up to second order will increase the rotational
character directly. It is easily recognized that the
phase plane for this system has multiple character.
In addition, there can be two singularities in the positive
Figure 14. Computer solutions yielding sustained oscillations for the
quadrant for the single activation and up to four for mechanism indicated below
double activation. Depending on the values of the
SCV the singularities may be unstable or even of saddle- Case A—nearly sinusoidal oscillations for the mechanism indicated. Case
B—highly nonlinear oscillations. The mechanism is:
point character. Unlike many of the mechanisms
considered in this paper, an additional singularity exists [G] +E1~* E,G E2X E2 + Y
at the origin (x 0, y 0). For this mechanism and
= =

also Mechanism 103, that singularity may be stable. EiG Ei + X X + E2X E2+X
70
Computer solutions for the single back-activation are Y + Ei+^± El
shown in Figure 14 for values of the SCV giving sus- Y + Ez -> EzY
tained oscillations. The results are quite comparable 72

to the previous mechanism (103) as would be expected X + e2 —> e2x EzY -> Ez

from the dominance relations. The first-order charac- and the constants are K\ =
0.11, K2 —

0.3, Kz =
0.9, =
0.3,
ter of Reaction Step 1 in Mechanism 103 and the zero- K7 = 0.5, Ks = 0.3, K9 = 0.9, Ki0 =
0.03, Kn =
0.9, Ki2 =
0.034.
Scales: x, y =
10. All enzymes = 1. T 8.5 =
sec. [GLU] =
1.0.
to first-order character of the enzymatic activation in Case B is the same as A except for K2=
0.1

40 INDUSTRIAL AND ENGINEERING CHEMISTRY

verified that such kinetics only occur when the rate
constants for the back-activation of y on the first enzyme
are made small. Such values lead to a characteristic
time for approach to the steady state which is compa-
rable to those for substrate (x, y) changes, with the result
that the two-variable reduced equations are not appli-
cable.
While a complete treatment would require a study
of the three-variable problem, the behavior is qualita-
tively clear. The low rate constants introduce a phase
shift or a lag between the values of y and the time at
which the enzyme realizes the activation. Thus, just
preceding the dip in y, x is falling and y is rising. How-
ever, x falls too far since the first step is not yet activated
and consequently, y begins to drop. During the same
time, the activation of the enzyme is taking place and
becomes effective causing increased flux in the first
step and a rise of x, which in turn causes y to begin its
increase. If the values of the activational rate con-
stants are increased, the double periodicity disappears
and the reduced, two-variable equations are applicable
as shown in Figure 15, B and C. More general com-
ments on the applicability of the reduced equations are
given in Appendix III. Figure 15. Computer solutions for mechanism indicated above with
Figure 16 shows computer solutions for values of the double activation present
SCV which provide a stable singularity at the origin. Case A—very low activational rate constants cause dip in y{t) kinetics and
Depending on their initial values, the variables may crossing in the phase plane. Far right figure is an expanded view of crossing
shown in middle figure. Cases B and C show solutions for larger values of
either exhibit sustained oscillations or approach the the activational rate constants and loss of dip. The mechanism is:
origin as shown in Figure 16A. Net flux through the 1 8
system ceases at that point even though the primary sub- [G] +EX-+ EiG E2X -> E2 + Y
strate G is still present. The behavior is perhaps clearer 2 9
E4G -» Ei + X X + E2X^±E2+X
in Figure 16i?, where G is allowed to change. If G 70
3
were fixed the oscillations would be sustained, but as G Y + El+^±El 77

becomes too low, the oscillatory conditions are lost and 4 Y + E,-+ E3Y

the variables decay into the stable singularity at the 5 12

r + es+^ez E3Y £3
origin. A more thorough examination of the phase plane 6

and its structure for these mechanisms is given in Appen-

dix II.
x + e2 Z e2x
=
0.89, K2 0.05, Kz 0.9, K4
=
0.2, K5 0.02, K8
= = =

Back-Activation—Back-Inhibition Oscillator 0.01, K, 0.9, K8 =

0.77, K9 0.9, K10 =
0.03, Kn 0.9. = = =

Scales: x 20, y 10.

= All enzymes =7. = T 14 sec. =

The third type of oscillatory diagram (Equation 74d) [G] 7.0=

can be obtained for a sequential mechanism as follows B. Same as A except for K\ 0.9, K& 0.7, Ki 0.53,
=
0.9, = = =

inh(ES) or inh2(ER)
K8 = 0.2, K12 = 0.03. Scales: x —

70, T =
8.5 sec.

J I C. Same as A except for K\ 0.9, K5 — =

0.9, Kq =
0.9, K8 =
0.3,
E\ E2 |
E 3 Ki0 =
0.02, K12 0.03. T = = 8.5 sec.
-
Y -

[G^X
X X
(107)

which can be written as

(inh)2(ES) Similar previous example, the strength of the

to the
{-1
Ei E2 Ez
cross-coupling be increased independent of the self-
can
|
coupling through the back-inhibition of y on the first
^X Y
- —
[G] (108)
reaction step. Computer solutions for this mechanism
act(S)
are shown in Figure 17. Both single and double back-
and has the following net flux diagram inhibition can yield sustained oscillations, the more
nonlinear forms being obtained with the stronger in-
hibition.
The direction of the rotation is in accord with Theorem
65 and the major axis should have positive slope. Such
is the case in Figure \1B for which the major axis is well
defined. For Figure 17, A and C, it is difficult to define

VOL. 59 NO. 5 MAY 1967 41

 Inh(ES) Systems with Conservation Condition
One final example, closely related to the glycolytic
system, will serve to illustrate several points. The
mechanism can be written as
E E E k

[G] t^FDP^PGAz-*; ATP ADP (112)

/m l2\ l3\
ATP/ yADP ADP ATP ADP ATP

inh(ES) act(ES)

the first step resembles the conversion of F6P to FDP by

the enzyme phosphofructokinase which can be inhibited
by ATP and activated by ADP under appropriate
conditions. Reaction 4 can be taken as an ATPase
reaction which is necessary to keep ATP from increasing
indefinitely. In this case it should be noted that ATP
plus ADP is conserved, that is, y + 2 constant. Be- =

cause of this conservation condition, the apparent three-

Figure 16. Computer solutions for the mechanism above variable (*, y, z) problem reduces to a two-variable
Case A—values of G and initial values of X and Y set so that variables problem.
approach stable singularity at origin as shown in the associated phase plane As one choice, take FDP (*) and ATP (y) as the basic
Case B—two examples for which the primary substrate G is allowed to variables. Since increasing ATP necessitates a de-
change. As G decreases to low, oscillatory conditions are lost and variables crease in ADP, the activation due to ADP becomes an
approach the origin. Note also the nearly stepwise decay of G and rise of
product P. Mechanism and constants are the same as in Figure 15 except effective inhibition due to ATP and simply adds addi-
as follows: tional strength to the inhibition due to ATP directly.
Case D. Same as A except for K\ =
0.7, T =
8.5 sec. In effect, the first reaction step may be replaced by
Case E. Same as A except for K\ = 1.0. Scales: G =
1000
Case F. Same as A except for K\ =
0.9, K2 0.06, K5
=
0.9, K§ = = [G]AfDP (113)
0.9, K7 =
0.9, K10 =
0.025, Ku 0.035. Scales:
=

GLU =
500, P 500 =
ATP / ADP

>u
inh2
a major axis because of the three-cornered or triangular
The following diagram is easily obtained
shape of the trajectory and consequently, it is not possible
to determine the appropriate quadrant for the phase
angle. The trajectories in Figures 17A and 17C appear
even more triangular when the phase plane is normalized.

(114)
Forward Activation-Forward Inhibition Oscillator
An example of the fourth type of oscillatory diagram
(Equation 74c), is given by the following mechanism
act(ES) or act2(ES)
FDP activates the production of ATP through reaction
k Steps 2 and 3, while ATP inhibits the production of
[G] (HO) FDP through Reaction Step 1. Self-inhibition of
1 A
FDP derives from Reaction 2 while self-activation of
inh(ES)
ATP derives from reaction 1. For the net self-character
with the net flux diagram of ATP to be activational, the self-activation due to
Reaction 1 must be of higher order than the self-inhibi-
tion due to Reactions 2, 3, and 4. Reactions 2 and 3
are self-inhibitory in ATP since an increase in ATP
(111)
causes a decrease in ADP and a consequent decrease
in the rate of Reactions 2 and 3.
If the system is analyzed with FDP and ADP as the
basic variables, the net flux diagram appears as
It clear from the dominance relations that the
is
oscillatory diagram can be obtained for appropriate
values of the SCV.
Computer solutions for this mechanism are given in v= constant (115)
Figure 18. The rotation is counterclockwise and the
major axis has positive slope in accord with Theorem
65.

42 NDUSTRIAL AND ENGINEERING CHEMISTRY

and the reasoning is similar. The strengths of the is given as (see Appendix III, Figure 19):
various coupling terms are identical with the previous
diagrams, as may be easily verified, and the oscillatory
ATP ES + FDP
state requires that ADP be self-activational.
It may be noted that the rotational direction and the [G] + ES E + ADP
h
slope of the major axis are exactly opposite for the two ATP + ES ES+ + ADP (116)
diagrams (see Equation 65). This is expected since the
two variables (y and z) must have opposite wave forms and has the rate law
(phase difference 7r). =

Computer solutions for sustained oscillations are G«y

Vi = —

(117)
shown in Figure 19. That “square wave” kinetics can
be obtained for appropriate selection of the SCV is y + Km + K
easily realized. The mechanism for Reaction Step 1
where Vm K<ige, Km k^g/ki and K =
kz/k±. Setting = =

z =
y where a a is the
—

conservation constant, then

v\{y) has a graphical shape similar to Figure 97), with a
Inh(ES) or lnh.*(ES)
maximum at

1-Act. (S) ym =
(118)
K a
+
10
1
4 ~Km

However, the function v\ goes to zero at y a instead =

1. of aty 00 as shown in the

graph, effectively transform-
X

ing the point at infinity to the pointy a. Such trans- =

TO
0 formation typically provides a higher order of dominance.
For this system, one finds that

(Kmz-1 y2Ka)
-

^rv ~
—

(119)
I. [(y + Km)z + y2K] z

with respect to the variable y and with 2 being taken as

z =
a —

y. Consequ
0

TIME -
4.2 stc.
[l as y -> 0 (; a)

10
10 at y =
ym
1rv.
-

\ (120)
a
f
1
c
1 00 as y a (z —>
0)
\
z
X

0
and can be made 1

singularity is moved to the

point y — a. It is also clear
that the motion in the x, y phase plane is restricted to the
region 0 < y < a, the line y a forming an upper =

Figure 17. Computer solutions for indicated mechanism for several

settings of SCV boundary.
A similar example which provides unusual values of
Note the nearly triangular shapes of phase plane trajectories for Cases A
and C. For such shapes it is difficult to determine appropriate direction of the dominance magnitude is given by the mechanism
the major axis. In Case B, the direction of the origin axis is evident. The
mechanism is:
X + E L EX
X u Y
X + El EiX + X
-*
X + e2
7
e2x
1
EX 44 E + y
2 -> ini /
[GLU] + EiX^Ei + X
E2 + Y z
/
w
1

Z + E ft E+ +
ki
W (121)
Y + EiX ExX+
4 Y A- Ez- EzY
which has the rate law
io
Y -b EiX+ E1X++ EzY-^Ez
6 Vmxf
where x' =
x/Km (122)
x' + 1 + (z/w)K
A. Ki =
0.9, K2 =
0.2, Kz =
0.9, 0.1, Kb =
0, K$ = =
0,
K =
0.9, Ks =
0.05, =
0.9, K\0 0.03. Scales: x
= =
20, If Z and W are conserved, such that z
y =
10, [GLU] =
1.0. All enzymes 1, T 8.5 sec. = =
a w, = —

then
the graphs of v\ vs. z are monotonically rising but we may
B. Same as A except for K2 = 0.9, Kz =
0.9, K& 0.02, Ks= —

0.12,
K10 = 0.13. Scales: x, y —

1.0, T 4.25 sec.

= have positive (for K > 1 + x') or negative (for K <
C. Same as A except for K2 0.13, Kz = =
0.9, K$ =
0.02, K% =
1 + xr) curvature. The dominance equation with
0.016, ifio 0.01. Scales: x, y
= =
10, T =
8.5 sec. respect to co (with z a w) is = —

VOL 59 NO. 5 MAY 1967 43

 1 for w -* 0 all cases studied. Based on such studies, the following
aK
1* =
K Oscillatory Rule of Thumb may be formulated.
(1 + xf) w + {a
—

w)K as w
i Tv Oscillatory Rule of Thumb: When the following
conditions are satisfied for two variable chemical
(123)
systems, then there exist values of the SCV for which
and again the dominance can be given any value (as sustained oscillations can be obtained. The condi-
w a) by proper choice of the SCV. tions are (1) that singularities exist for finite values of
the variables (*, y); (2) that the variables can satisfy
Summary of Conditions for Oscillatory Reactions the oscillatory net flux diagram (Equation 53) in
In regard to mechanisms which can be reduced to the region of the singularity as determined by the
two-variable problems, the preceding computer studies dominance relations; and (3) that A can be made
have demonstrated a very important point. While the positive at the singularity (124)
theory has been based on necessary conditions, those There are certainly exceptions to this rule in terms of
conditions have been proved to have “^-sufficiency” in mathematical functions and probably exceptions even
Act(ES) or Act2(ES) when limited to chemically realizable functions. How-

G
k
--X
rv--Y l ever, the rule appears to hold for most types of mech-
anisms, because of the flexibility provided by the manner
1
V 2
,nh
Inh.(ES) 1 in which the SCV enter the rate laws.
It must be realized that many complex mechanisms
cannot be reduced to two-variable systems. In that
case, three-variable or even higher order diagrams must
be studied. The possible behavior patterns for such
systems is much more complicated, being related in
part to the roots of higher order polynomials. A pre-
TIME 8.5 sec.
liminary investigation of the three-variable problem,
based on the same type of arguments employed pre-
10 viously seems to indicate the following conclusions.
The basic equations for the three-variable problem
have the form

Vx
=
y, 2)
»v
=
Y(x,y, z) (125)
TIME 4.25 sec. 10
vz
=
Z{x,y, z)

As in the two-variable problem, the character of the

10
interactions is defined by the signs of the partial deriva-
tives. In this case there are nine, given as Xx, Xv, Xz,
Yx, etc. For example, the simple triangle scheme com-
posed of pseudo-first-order reactions

TIME 4.25 sec.

Figure 18.
settings of the SCV
Computer solutions for the mechanism indicated for several
—~AOc_
The mechanism is: has a net flux diagram represented as

[G] ^X Y + E2 e2y

X + El -> EyX E2Y -* E2

E\X ~^El + Y Y + E2Y EiY* + X

4
Y -> X
E2Y* E2Y** + X
10

A. Ki =
0.008, K2 0.9, X3 0.013, K4
=
0.9, Kb 0.05,= = =

X6 =
0.9, X7 0.07, K8 0, X9
=
0, Xi0 0.014. Scales:
= = =

x, y =
5, T 8.5 =
sec., [<3] 1.0, all enzymes 1 = =

B. Same as A except for K\ =

0.02, X3 =
0.026, Kb 0.2, X7
= =

0.2, K8 0.9, K9 = =
0.9 Kio =
0.02. Scales: x,y 10, =
where the self-coupling terms are all inhibitory and the
T =
4.25 sec.
cross-coupling terms all activational.
C. Same as A except for K\ =
0.04, X3 0.06, Kb = =
0.3, X7 =
With regard to oscillations the preliminary study in-
0.15, Ks =
0.9, K, = 0.9 Kio 0.05. Scales:
=
x, y =
10,
T =
4.25 sec. dicates that such a diagram can at most exhibit damped

44 INDUSTRIAL AND ENGINEERING CHEMISTRY

 The conditions given in the suggestion (128) are weak
and more stringent conditions can certainly be formu-
lated. However, it appears likely that the suggestion is
applicable to a system involving any number of varia-
bles, although considerably more study will be required
to prove it. With regard to reaction mechanisms, the
tentative conclusion is that systems which exhibit sus-
tained oscillations must have feedback contrary to the
simple reaction schemes (such as pseudo-first-order
systems).

RELATED ASPECTS
Once an oscillatory reaction has been established a
Figure 19. Computer solutions for the indicated mechanism. Note •variety of related oscillatory phenomena can be realized.
the square wave character of the kinetics Phenomena introduced by changing SCV and by projec-
The mechanism is: tion of oscillations have considerable interest in regard to
1 the analysis of experimental data.
6
ATP + E —> ES + FDP ADP + E2S -> E2 + ATP
2
7
The Control Stability Plane—Effects of the SCV
[G] + ES -*E + ADP PGA + F3 -» EsS
3 As already indicated, the kinetic behavior of the
8
ATP + ES^t ES++ + ADP
4 ADP + E*S - F3 + ATP dynamical variables x, y depends on the values of the
5 9 SCV which determine the structure of the x, y phase
ATP ADP
FDP + E2 -» E2S + PGA plane. Such dependence can be understood through
Ki =
0.9, K2 0.9, K3=
0.9, K4 0.8, Kb
=
0.02, KG
= = = the use of a control-stability plane. As an example,
0.9, K7 0.14, K8
=
0.9, K9 0.02. Ste/w: FF>F
= =
50, =
consider the mechanism
^4 TP, ADP 7, PGA =
0.7 X 5000, F=
77 [G] = =

7.0. ^477 enzymes —

1
k E^\Ct E

[GLU ] —

F6P^ FDP —
(130)
1 / 2\t
* y 3
oscillations, a result consistent with that already found. [ATP] ADP

However, the conclusions appears to remain valid

which is identical to Mechanism 83, except that the rate
even if all reaction steps in the Scheme 126 are replaced
constant for the enzymatic reaction is now pseudo-first
by monotonic, but otherwise arbitrary, rate laws which
order and depends on the value of [ATP].
have the same character. The fact that damped
The control-stability plane is plotted as a function of
oscillations can be obtained for such diagrams, but not
the SCV as shown in Figure 20/4. For different, but
for the similar two-variable diagram (Equation 62)
fixed values of the SCV (in this case [GLU] and [ATP]
can be attributed to the possibility of achieving a phase
shift of 7r around the loop. The condition that such were chosen), the kinetic behavior of the dynamical
variables x, y is examined and the point in the control-
oscillations be damped appears to derive from the fact
that all self-coupling terms are inhibitional. stability plane is appropriately labeled with C for sus-
tained oscillations, D for damped oscillations, N for
For a system to exhibit sustained oscillations, the pre-
unstable situations in which one or the other variables
liminary study suggests the following:
x, y approaches infinity, and S for stable, nonoscillatory
Suggested Condition for Sustained Oscillations: For a responses. The regions of the SCV giving rise to the
system to exhibit sustained oscillations, it is necessary various types of behavior can then be mapped out.
that (a) some of the self-coupling terms be of opposite More subtle features of the behavior, such as the ampli-
character and (b) that some of the cross-coupling tude and frequency of the oscillations or damping factors
terms be of opposite character (128) can also be indicated by contour lines, as shown in
For example, sustained oscillations can be obtained Figure 21.
for a diagram of the form If the SCV are no longer fixed but allowed to change
slowly (relative to the characteristic times or period of
the dynamical variables), the dynamical variables can
be treated as passing through a sequence of stationary
states and the behavior can be determined from the
control-stability plane. Changes in the SCV may arise
either through the natural course of the given reaction,
by side reactions, or by externally imposed variations.
If the SCV are slowly varying, then the behavior of the
dynamical variables may be determined by examining
the trajectory of the SCV through the control-stability
plane.

VOL. 59 NO. 5 MAY 1967 45

 CONTROL STABILITY PLANE would require a multidimensional control-stability
r\M space which included the rate constants as well.
E,
[Glu]—>
F6P^p«^FDP-^ Stimulated Oscillations
1 2 UTP] [ADP] [ADP] [ATP]

Systems which do not exhibit sustained oscillations

under stationary-state conditions, can, nevertheless,
B produce kinetic behavior similar to that of the previous
figures including both the increasing and decreasing
oscillations. However, for such systems, it is necessary
that the SCV change rapidly (i.e., with a characteristic
0 TIME- 17 sec.
time comparable to that for the dynamical variables)
[ATP]-0.4
NO OSC.
and move along an appropriate path in the control-
DAMPED OSC.
CONTINUOUS OSC.
stability plane.
As an example, consider the mechanism

0 TIME- 17 sec.
[Glu.]- 0.2
for which the control-stability plane as a function of
Figure 20. Effects of changing SCV for indicated mechanisms GLU and G has regions of damped oscillations but no
A—control stability plane showing regions of damped and sustained oscilla-
tions in variables (F6P, FDP) as a function of the control variables (GLU
regions of sustained oscillations. The damped oscilla-
and ATP). Cases B, C, and D illustrate kinetics when control variables tory regions refer to the fact that if the dynamical
change slowly and behavior can be correlated with passage through regions variables (*, y) are perturbed they will exhibit damped
of control stability plane as indicated by arrows in plane A; correspondence
is B to arrow 7, C to arrow 2, and D to arrow 3. The mechanism is: oscillations as they return to the stable (nonoscillatory)
stationary state. If the dynamical variables can, in
4
effect, be continuously perturbed as when the SCV
1

[GLU] -> F6P FDP + E* z=±E

(say, G) are allowed to change rapidly, then the oscilla-
5
2 6
F6P + E -> ES + FDP FDP + EiEiS
3 7
[ATP] + ES E -f [ADP] [ADP] + E,S -> Ei + [ATP] A

B
1 2

Constants: Kx =
0.05, K2 0.9, K3
= =
0.9, K4 =
0.9, Kb =
0.6,
K6 =
0.9, K>i =
0.2. Scales: [GLU] [ATP], [ADP] =
constant.
All enzymes =
7, F6P, FDP 10 =

Figure 20B, C, D, provides computer examples of

such analysis, for Mechanism 130 with GLU and ATP 0.5 1.0
taken as the changing SCV. The lines of motion are ATP
NO OSC.
indicated in Figure 20A and can be easily correlated DAMPED OSC.
with the kinetics. In general, it may be noted that if.the CONTINUOUS OSC.

SCV change very slowly, passage through the damped 1.0 1.0

oscillatory region does not give rise to oscillations in the

dynamical variables but only to slow changes correspond-
ing to the changes in the position of the stable singularity.
As the SCV pass through the region of continuous
oscillations, the dynamical variables will exhibit damped
or undamped oscillations with changing frequency 0.5 1.0
according to the associated amplitudes and frequencies
—
ATP
AMPLITUDE (PEAK TO PEAK) OF F6P
of the corresponding stationary state oscillations. Fi- IN CONCENTRATION UNITS

nally, Prigogine’s theorem that rotations cannot occur C

D
around the equilibrium state is applicable if the SCV
approach the complete equilibrium state for the system. Figure 21. Examples of quantitative computer study of oscillatory
features as a function of control variables {GLU, ATP) for the mech-
Thus, in the computer examples, as GLU — 0, all anism given in Figure 20
oscillations cease and the phase plane trajectory be-
comes a direct approach into the equilibrium point as A—regions of continuous and damped oscillations. B—contour lines
indicating stationary state level of F6P. Erratic appearance of these lines
can be seen in Figure 22. The structure of the control- in oscillatory region is due to taking the level as the mean value of the oscilla-
tions since the singularity cannot be determined. C—variation of period
stability plane for two of the SCV depends, of course, of sustained oscillations. D—contour lines for amplitude of these oscilla-
on the values of the other SCV. A complete picture tions

46 INDUSTRIAL AND ENGINEERING CHEMISTRY

tory response may be prolonged until the SCV pass out SCV kinetics must contain frequencies comparable to
of the damped oscillatory region in the control-stability the characteristic frequencies of the dynamical variables
plane. A precise treatment of the problems which if the stimulated oscillations are to arise (see Figure 23).
have time-varying SCV requires the study of higher Such stimulated oscillatory response patterns can be
order systems involving three or more dynamical varia- distinguished from the potentially sustained oscillatory
bles, as evidenced by the crossing which occurs in the response. The simulated response will disappear as the
phase plane trajectory of the two-variable approxima- SCV become more stabilized, while sustained oscilla-
tion (see Figure 22). The treatment and discussion tions will become more prolonged. However, it should
given in this and the preceding section consider that the be noted that the distinction between these two types of
SCV act as perturbations on the primary dynamical behavior does not necessarily reflect any major changes
behavior of the two-variable approximation. In many in the mechanism. The mechanisn can exhibit either
of these cases, the two-variable equations can be treated type of behavior according to the values of the SCV
as autonomous equations with time-dependent SCV. which may or may not be experimentally accessible—
Computer studies have indicated that the detailed e.g., Mechanism 83.
nature of the SCV kinetics, such as the curvature are Experimental studies of the cellular extracts have
also important. Thus, for some mechanisms a rapid shown more prolonged oscillations when the glucose
first-order decay in the SCV may be sufficient, but for was morenearly stabilized. However, true stationary-
others, a second-order decay may be needed to provide state conditions have been difficult to obtain experi-
sustained stimulation. Consequently, such oscillatory mentally, largely because of the problems of stabilizing
behavior may be dependent on appropriate side reac- the levels of ATP, as well as other substrates and prod-
tions for the SCV. A high degree of curvature as ucts of the reaction.
occurs upon stepwise change in the SCV may produce
oscillations when the damping constant is very large Projection of Oscillations
(and positive). The damped oscillations which occur The specific collection of reactions which provide
for the pseudo-first-order reactions (Figure SB) require the appropriate feedback for the basic dynamical varia-
such sharp changes. While the stimulation used in bles can reasonably be referred to as the source of the
Figure SB is applied to the dynamical variables, similar oscillations. If the basic variables began to oscillate,
results can be obtained if stepwise changes are made in their oscillations will be projected into other chemicals
the SCV. Effectively, the Fourier Expansion of the through reactions connected to the source. As a simple
example, consider the following scheme

* \
; ;

[GLU] —
G6P F6P
/WV/VW
—
FDP
/WV /WV
—
I
GAP —
DAP —

(132)
where the source of the oscillation is defined by the
dashed box with F6P and FDP taken as the basic dy-
namical variables (denoted by the squiggle). Oscilla-
tions in these basic variables will be projected outward
from the source both in the direction of the net flux
and against the direction of flux (through reversible
reactions). In the above example, G6P will oscillate
through its connection by reversible reaction to F6P;
it will furthermore reflect (see below for modification)
the oscillatory form of F6P. Similarly, the GAP and
DAP will oscillate but will reflect the oscillatory form of
FDP.
In general, the projection of oscillations will introduce
phase shift, distortion, and filtering. The projection of
small amplitude sinusoidal oscillations is easily calculated
and can be used as a first approximation for large ampli-
tude oscillations. For reactions of the form
X^Y-^ (133)
where X is the projector considered to be oscillating
Figure 22. Kinetics and phase plane trajectory for the mechanism of sinusoidally with frequency co and amplitude a according
Figure 20 when GLU is allowed to change naturally to
Note the loss of the oscillatory (rotational) behavior as the system approaches
equilibrium (GLU — 0). The mechanism is the same as in Figure 20 v =
x + a sin oot (134)
except for K\ =
0.047, K2 =
0.6, K% =
0.9, K± =
0.4, K$ =
0.65,
0.03
Ks 0.55, Kn where x is the (steady-state) level, then chemical Y
= =
mean

VOL. 59 NO. 5 MAY 1967 47

will have the projected oscillation given by STIMULATED OSCILLATIONS

y =
y + ft sin (cot + 0) (135) 10

A
where ft is the amplitude and 0 the angle of phase shift.
It is usually convenient to express amplitude changes in S

terms of relative (or per cent) values as 0

FDP

0 TIME 8.5 sec.

P
-

and P' (136)

x y
100
Conversion back to absolute values is given by

ft yft^
=
(137)
a x a
TIME— 17 sec.
The ratio of the relative amplitudes and the phase shift
will be given by the equations
Figure 23. Example of stimulated oscillations for the mechanism
indicated
(138)
a
A—only damped oscillations can occur when control variable G is main-
tained constant. B and C—stimulated oscillations which can arise when
G changes rapidly or has regions of sharp curvature. Note loss of such
oscillations (in C compared with B) as changes in G become slower or the
curvature is diminished. The mechanism is:
tan 0 = - -

(139)
7
1 6
where and y depend on the specific rate laws and are
m [GLU] -> F6P [G] + FDP ->

given in Table III. 2 7

Reactions 3, 4, and 5 in the table behave as simple

FOP + E -> ES S -F Ei -> EaS
3 8
linear filters. Small and large amplitude sinusoidal ES -* E + FDP ErS ->E + G
oscillations are projected as sinusoidal oscillations with 4
the associated phase shift and amplitude change. Non- FDP + E+ E + [G]
5
linear oscillations will be filtered, such that low-fre-
quency (Fourier) components (co < 7) will be more
A. Ki 0.006, K2
=
0.9, =
K3 0.9, KA
=
0.14, Kb 0.5, K6 = = =

strongly emphasized in the projected wave form. Con- 0.2, K 0.9, -Tv8 = =
0.02. Scales: F6P, FDP 10, [5], =

sequently, the projected wave form will be more nearly [G] constant, T
= =
8.5 sec.
B. Same as A. Scales: S, G 100, T 17 sec.
sinusoidal. In contrast, the bimolecular reaction (Case
— —

C. Same as A except for K% —

0.9. Scales: S, G 1000, T 8.5 = =

10 in Table III) provides a peak frequency pass filter,

TABLE III

Reaction V Y Comments Reaction M Y Comments

vi(x) x v% General Case E E0 1 yf

-f- Vm0 1 Both Michaelis-Menton
X-^ ^012/ (8) X-v Y_^
TA\y Ujx and v*\y are partial 1 + *' Kmo (1 +/)2 enzymes y' =
y/Km0
v°(y) derivatives evaluated at
Y —m (9) -v Y 1 kx vl considered constant;
x x and y
=
y —

General Case
Y + X .

7r/2 < 0 < x—i.e., the

vt. X v\x vl considered constant. Y wave form is inverted
(2) v\x
Vy
and v\y are partial de- with respect to X
y v\v
rivatives evaluated at wave form
(10) Yv Let Z = Z -F 7 sin +
ki ki (oot 5)
X/
X _V Y_* 1 h Not restricted to small
amplitudes Z with respect to X
kl k2
X Y-> 1 k2 + 1 Not restricted to small 1

k-\ amplitudes
kjc

[G]—> Y^=*
k kf X
k (Upstream projection)
N- (£)•+(£)• +
(£) 1 +
k'
kjc
k' ky
X Max of 7/a at u =
\/k' kx
E k 1 Michaelis-Menton Z with respect to Y
X_v Y_, k
enzyme
1 + *' xr =
x/ Km
k E o
Vm° k'
(7) X-, Y-, 1 +>' Define 5=0 —

7 then: tan y =--, 7/(3 =

LTn
Km* (1 + y02

48 INDUSTRIAL AND ENGINEERING CHEMISTRY

eliminating both low and high frequencies and passing
frequencies centered at co (kf k x)1/2. With such
=

reactions, higher order harmonics can be obtained from

the original nonsinusoidal wave forms. Such reactions
can also be cascaded to provide very sharp filtering and A

the conversion of nonlinear wave forms into sinusoidal PROJECTION

OF
shapes.
OSCILLATIONS
Excluding cases 3, 4, and 5 all reactions in the table
introduce nonlinear projection, as can be seen by the a+ b-c—
concentration dependence of the quantities fx and y. I
The formulas given above (138 and 139) are restricted
to small amplitude sinusoidal oscillations, which are
projected as nearly sinusoidal forms. The projection
of large amplitude sinusoidal oscillations produce non-
linear wave forms. Figures 24 and 25 illustrate typical
results for enzymatic and bimolecular reactions which
are easily interpreted. In the enzymatic case (Figure
25) the clipping of the top of the sinusoidal wave form
arises because the rate of the enzymatic reactions be-
comes nearly zero order in the concentration (for x >
Km).
In general, the effect of projecting oscillations can be
Figure 24. Projection of sinusoidal oscillations in a bimolecular
derived by comparison of the characteristic times (rr) reaction. Note the nonlinearities introduced in the projection of large
for a given reaction with the period of the oscillations amplitude sinusoidal oscillations in B
(t0) (for each Fourier component). For relatively fast The mechanism is:
reactions (tt <<C t0) the oscillation is projected with small 1 2 3
phase shift and amplitude change while for rT r0, the -* A + B C ->

magnitude of the phase shift approaches 90° (neglecting A. Kx =

0.3, K2 =
0.9, K3 =
0.39. Scales: A, C =
7. All en-
inversions as in Case 9, Table III) and the amplitude zymes =
1, T = 0.035 sec.

decreases. As demonstrated elsewhere (75), for the B. Kx =

0.03, K2 =
0.9, K3 = 0.04. Scale: T = 2 sec.

nonlinear reactions, the characteristic time (rr) depends

on the amplitude of the changes and, consequently, can
become comparable to or greater than the period (r0)
causing amplitude dependent distortion.
Specific formulas for projection into sequences of
reversible reaction can be derived but are considerably
more involved. Nonetheless, some general results are
apparent. Define a simple reaction sequence as of the
form

A0 ^ Ax A2 A3 ^ A, ^ (140)
Figure 25. Projection of sinusoidal oscillations for reactions in-
where the individual reactions may be of first-order dicated
character or enzymatically catalyzed. Taking the A—oscillations in x rise above the Km for the enzyme giving rise to clipping
original projector as A0, the phase shift between any in projected wave form. In this case, xit) < Km for the enzyme, and the
sinusoidal oscillations are projected without distortion. The mechanism is:
reaction pair (A* to A*+i) must be less than —90° in
accordance with the above remarks concerning the X + E-> EX
1

characteristic times. Consequently, following sub-

2
strates lag behind preceding ones and rotation is clock- EX E + Y
wise in the phase plane constructed with At on the ver- 3
tical axis and A*+i on the horizontal axis. The phase
shift is accumulative and, for sufficient steps, may exceed A. Kx =
0.9, K2 =
0.1, K3 =
0.18. Scales: y =
10, T = 8.5 sec.
All enzymes =
7.
any particular value; for example, between Ai and
B. Same as A except for K2 0.9, K3 0.4
may be greater than 180°. If the reactions
= —

A*+3, |&copy;|

are pseudo-first order, the ratio of relative amplitudes is

less than 1 as one proceeds down the sequence (a\+1 <
a'i). If enzymatic reactions are involved, the ratio of
relative amplitudes may be greater than 1 (e.g., Cases
7 and 8, Table III) as seen in Figure 26 and distortion
of the wave form can also arise.

VOL. 59 NO. 5 MAY 1967 49

 Detection of the Source

Given a system of chemicals with oscillatory kinetics,

the problem of determining the basic variables which act
as the source of oscillations has obvious importance.
Biologically, these variables provide a major center of
dynamical control; and chemically, the existence of
appropriate feedback reactions is implicated. However,
in view of the projection phenomena, the localization
of the source of oscillations is not trivial. In general,
neither the wave forms nor the phase relations between
different variables provides any clues as to the basic
Figure 26. Projection of sinusoidal oscillations for the reactions in- variables. Some information in regard to the basic
dicated. Note that relative amplitude of y is greater than for x in reactions must be known or assumed. If all reaction
Case A. Distortion of the projected oscillation can be seen in B
pathways are known, the potential basic variables of the
The mechanism is: oscillator may be deduced by a search for the appropriate
1
feedback to satisfy the oscillatory diagram. Subsequent
X-* Y
2
analysis or computer studies can then provide informa-
Y + E —> EY tion as to the values of the SGV.
3 With less information in regard to reaction pathways,
EY->E + (P) the following conclusions can be applied to partially
A. Kl =
0.2, K2 =
0.9, K3 = 0.07. Scales: all =
/, T = 1 sec. deduce the basic variables. Since oscillations are pro-
B. Same as A except for =
0.06 jected outward from the source, there will be a crossover
(9) in the direction of phase shift as the variables in a
sequence are analyzed. For example, consider a simple
PROJECTION OUTWARDS FROM SOURCE sequence of reactions containing a source of oscillations,
-------Act U*t— which for argument is located at A4, A5 but presumed
i
I XMU J
cl i.
unknown to start with.
[Glu]-M56P^fF6P-- '
FDP j-^P
"““SOURCE
[zlo Ax A4 # A5 :
A6

Source (141)
Since the oscillations are projected outward from the
source, and since for simple reactions, there is a phase
lag in the direction of propagation, the remark is evident
(see Figure 27 for an example). Phase plane plots with
Az- on the vertical axis and Ai+1 on the horizontal axis
will exhibit counterclockwise rotation for i 1, 2, 3 =

and will exhibit clockwise rotation for i 5, 6, 7, =

. .
(n .

1). If the magnitude of the phase shifts, given a

—

plus sign for counterclockwise rotation, is plotted in the

order of the sequence as shown in Figure 28, then a
crossover point (change of sign) occurs at the beginning
or the end of the source depending on the rotational
Figure 27. Computer example of projection of oscillations outward
from the source.Note that G6P is advanced (in time) with respect to
F6P and that P is advanced with respect to FDP
The mechanism is:
6
[GLU] -* G6P FDP + Ei* ^ Ei
7

G6P F6P 5
3 FDP + E2 -» E2S + P
4 9
F6P + Ei-* EiS + FDP E2S —> E2
5 10
EiS -> Ei P ->

A. Ki =
0.03, K2 0.13, K, 1.11, Ki
=
0.9, K5 0.45, Kg = = = =

0.7, K7 0.5, K8 =
0.9, '9 0.065, Kio
=
0.23. Scales: - = Figure 28. Phase shift cross-over plot
Substrates 10, T 2.131 :. [GLU]
=
1.0. All enzymes
=
7 = =
Phase shift between adjacent chemicals is plotted as a function of location
of chemicals in sequence (Equation 141). A cross-over point (change of
sign) occurs at beginning (dashed line) or end (solid line) of source of
oscillations

50 INDUSTRIAL AND ENGINEERING CHEMISTRY

 direction of the basic variables (A4, A5) (see Equation
65). It is important to use adjacent variables in the
sequence since accumulated phase shift could otherwise
lead to a misinterpretation of the data.
It may be noted that the phase shift crossover only
determines one of the basic oscillatory variables. For
the simple mechanisms, the other variable is adjacent to
it, either one unit up or down depending on the rota-
tional direction of the basic oscillator. When the
feedback reactions which comprise the source extend
over several substrates (see Equation 79), distinguishing
between the other oscillatory variable and the projected Figure 30. Filtering
variables becomes more difficult. In general, additional High frequency components generated by mixing of oscillations in a bi-
knowledge is required in regard to the reaction scheme. molecular reaction can be filtered out from the low-frequency envelope by
subsequent reactions as shown by D in figure. The mechanism is:
Further complications are encountered when coupled
reactions or other means of feedback connect the sub- A+B -* C
strates in the sequence as in the glycolytic example 3\
D
(Figure 5). Under such circumstances oscillations pro- and the constants are: K\ = 0.6, K2 =
0.8, K3 =
0.005, K\ =
0.0014.
jected from the source by one pathway are mixed with Scales: T = 8.5 sec., D, C = 1
oscillations projected by another pathway. For in-
stance 3PGA mixes oscillations projected from the source (within experimental accuracy) without their sum being
through GAP and through ADP. Such mixing intro- conserved.
duces additional perturbations on the wave forms.
Oscillations can also be projected into the kinetics of Beat Frequencies
primary substrates and final products giving rise to The phenomena of beat frequencies provide a simple
nearly stepwise decay or appearance as illustrated by means for obtaining lower order harmonics and may be of
GLU in Figure 16. Finally, it may be noted that
interest in connection with biological clocks having daily
whenever two variables are involved in a conservation
and yearly rhythms. A bimolecular reaction or its
loop, they must have exactly 180° phase shift and will
exhibit a straight line of slope —1 in the phase plane equivalent provides a chemical method for beating the
frequencies from two different sources of oscillations
(normalized) regardless of the wave form. The con- as in the mechanism
verse is not true; two variables may give straight lines

A + B —^
C —^
(142)
0)1 0)2
1 0 |i Wl>7Wl SIMPLE AHODULATKMt

IEAT FREQUENCY EFFECTS A linear analysis of the reaction shows that the kinetics
A+l—C —
of the intermediate (G) relates to the sum of the reactant
", ", kinetics and will contain terms of the form

c*)i + o)2\ (on o)2\

~

. .

--—If and cos (143)

“j =
5/6 <*>, leafs
0 TIME 2.13 set. (------Jt
1.0 1,0
..iW'l.C F involving both the sum and the differences of the fre-
z
B 1
.<

\
H

/ V / quencies. Computer solutions for this reaction are

S
S./WWVW
/wwvww 1
y y • given in Figure 29. Extremely low frequencies are
generated when the basic frequencies (coi and o)2) are
nearly equal. As shown in Figure 30, the low frequency
TIME 2.13 sec. 0 TIME - 2.13 sec. component can be filtered out by subsequent reactions
with appropriate rate constants. Whether the low
Figure 29. Generation of beat frequencies bv mixing of oscillations
in a bimolecular reaction frequencies of biological rhythms are generated through
beating or from a single source involving very slow
Cases A, B, and C correspond to different relations (as indicated) among reactions is certainly not clear at the moment.
the projector frequencies. Since the intermediates (C) contain sums and Another example of the production of multiply
difference frequencies, very low harmonics (the envelope) can be generated
when coi ^ 002 as in Case B. Note that only chemical C is shown in that periodic wave forms is given by a modification of the
case. The mechanism is: back-activation mechanism (Equation 83)
1 2
A + B C —>
Oact
E \ E k
A. K\ 0.9, K2 0.07, 001 9.6 cycles jsec., 2.45 cycles jsec. F6P FDP G
[GLU]
= = = = — —

Scale: 1,T 2.13 sec. All enzymes

C02 — —
(144)
I_I
—
—
1

B. Same as A except for, coi = 3.5 cycles j sec., co2 = 4.4 cycles /sec.
C. Same as A except for K2 =
0.9. Scale: 5. Computer solutions exhibiting a double periodicity are

VOL. 59 NO. 5 MAY 1967 51

 of dominance relations relating to the relative character
of a single reaction step. By such analysis, Bierman’s
suggestion (4) concerning the importance of autocata-
t lytic steps and reactions at a wall are substantiated in a
more general form. Fulfillment of the appropriate net
flux diagram and several simple quantitative conditions
(see Equation 124) appears sufficient to guarantee that
the system will exhibit sustained oscillations for some
values of the structural control variables, at least for
mechanisms based on commonly encountered reactions.
t Previous beliefs (/) that oscillating biochemical reac-
tions could probably not take place in homogeneous
systems and that they required coupling to diffusion
and absorption processes seem now to be unfounded.
The mechanisms discussed previously and their general
Figure 31. Computer solutions for mechanism indicated. The
agreement with the experimentally observed oscilla-
tions in extracts, leave little doubt that such reactions
doubly periodic kinetics shown can be obtained for appropriate values of
the SCV can occur in a single homogeneous phase. Reactions at
The mechanism is:
the wall are based on the Langmuir adsorption isotherm
which has a rate law dynamically equivalent to an
1 6
[GLU] -> F6P FDP + Ex - EiS enzyme reaction. It is interesting to note that on a
2 7 microscopic level, the enzymes (macromolecules) may
F6P + E-* ES eys -*ex + g easily be considered as 4walls” relative to the substrate
4

3 8 (small molecules). Although enzymatically catalyzed

ES -> E + FDP G ->
reactions can occur in a macroscopically homogeneous
4
FDP + E+ E + G phase, it might be argued that it is microscopically
composed of heterogeneous phases, although the physical
5

A. K\ significance and value of the word phase” is largely lost

4
0.025, K2
4
=
0.45, Kb 0.45, iC4
=
0.25, Kb
=
0.6, Kb
= = =

0.45, iC7 0.05, K%

=
0.037. Scales: F6P
=
50, FDP = =

5,G 70, T
=
8.5 sec. [GLU] =
7.0 = by doing so. In the dynamical sense, the distinctions
between heterogeneous and homogeneous reactions is
irrelevant. It is only necessary that the dynamical
shown in Figure 31. The kinetics can be crudely variables, whatever they represent, satisfy the appro-
explained by treating G as a perturbation on the basic priate rate laws.
oscillatory variables (F6P and FDP). The oscillations For complex systems of reactions, the nature of the
in FDP are phase shifted at G and then remixed at the theoretical analysis leads directly to the concept of a
source through the back-inhibition. The nonlinear “source of oscillations” as that subset of reaction path-
mixing of G at the source gives rise to the double perio- ways and chemical variables which possess the appro-
dicity. Such multiply periodic wave forms have already priate feedback. Oscillations generated at the source
been observed experimentally (as in Figure 7), although are projected into connected reactions and can give
the exact reaction responsible for such kinetics is not yet rise to a variety of interesting phenomena. Filtering
clear. and more general forms of wave shaping, generation of
multiply periodic forms, and the production of sub-
CONCLUSIONS AND SUMMARY harmonics can be readily obtained from simple reactions.
In short, virtually all the dynamical characteristics
Chemical
associated with modern electronics can be generated
Only a few years ago the theory of oscillating reactions through chemical reactions. By proper selection of
was restricted to a few specific mechanisms, involving reactions it would be possible in principle to construct a
plausible but experimentally unverified reaction schemes. chemical radio, although the relatively low frequencies
In contrast, the general treatment given here provides a developed would make it impractical in the audio
plethora of specific mechanisms easily constructed in range. However, such dynamics may play a role in
terms of known reactions. And conversely, given reac- biological functions involving transmission of informa-
tion schemes can be analyzed for the possibility of tion and control.
oscillating kinetics. The essential requirement for such Although the potential kinetic behavior may be deter-
oscillations is the existence of feedback in the reaction mined from knowledge of the reaction scheme, the
pathways. For systems reducible to two variables the problem of deducing the reaction scheme from the
feedback must be able to provide a net flux diagram with observed kinetics is complicated. Neither wave shapes
cross-coupling and self-coupling of opposite character nor phase relations provide any direct indication of what
in regard to the activational and inhibitory behavior of the basic reactions or pathways might be. In general,
the variables. The establishment of net activation or all that may be said is that if a perturbation of a single
inhibition can be determined by inspection and the use chemical in any chemical system leads to any form of

52 INDUSTRIAL AND ENGINEERING CHEMISTRY

oscillatory response (including overshoot-undershoot), techniques. If assays of the concentrations are taken at
then the system must contain closed reaction pathways time intervals greater than the period of the oscillations,
which provide feedback. Such feedback pathways may the resultant data will appear to form a curve with some
be stoichiometric or nonstoichiometric; and if the system interesting bumps but not periodic oscillations. In
exhibits undamped oscillations in the transition from some cases, a disbelief in oscillating reactions or a strong

one stationary state to another or sustained oscillations desire to obtain smooth curves has caused evident
under stationary-state conditions, then some of the periodicities to be dismissed as experimental error.
reaction pathways provide self-activation and cross- The fact that convincing evidence was first obtained by
coupling of opposite character. Further deductions continuous recording techniques such as spectropho-
require some presumptions or a priori knowledge of the tometry and fluorometry is easily understood.
nature of the reactions involved. Thus, if the reactions In the past two years, a considerable number of
are sequential and the order of the reactions is known, oscillating reactions have been discovered in various
then the source of oscillations can be partly deduced aspects of cellular chemistry (2), ranging from single
from the phase-shift-crossover theorem, and the direction enzyme reactions (23) to the more complex reactions
of rotation will indicate the specific nature of the feed- of mitochondria. One example of the latter is shown
back pathways involved. in Figure 32. Upon addition of oxygen to a valino-
Once the oscillations are established, they provide a mycin-supplemented mitochondrial preparation, oscilla-
convenient tool for analyzing other reactions in the tions are observed in the concentrations of H+ and K+.
system. For reaction sequences exhibiting oscillations, There are concomitant changes in light scattering
phase shift and wave form changes can provide informa- (not shown on the graph) usually attributed to changes
tion as to the size of rate constants and the nature of the in the mitochondrial volume. So far as the theory
reaction steps (whether first order or second order, or presented here is concerned, it should be noted that the
enzymatic). volume may be taken as one of the basic dynamical
The situation becomes particularly interesting in the variables and treated in the same manner as the chemical
case of cellular chemistry, such as glycolysis. The a variables. In all these cases, it remains to determine
priori information derives from studies of separate enzyme the appropriate feedback pathways which provide the
and metabolic reactions. Presuming that such basic source of the oscillations.
reactions are still valid in the extract and the cell, it Although it has not been experimentally possible to
becomes possible to deduce the control situation and the obtain indefinitely sustained oscillations, it seems likely
existence of additional feedback pathways by careful that the oscillations could be maintained if the values
analysis of the kinetics. Such a study is now being of the SCV could be kept constant (see Figure 4). How-
carried out for the heart muscle extracts and will be ever, the disappearance of the primary substrates, the
reported in detail elsewhere, but preliminary results buildup of final products and the deterioration of the
indicate that the source of oscillations can be attributed enzymes are somewhat difficult to overcome experi-
to the action of F6P and FDP on the enzyme phospho- mentally.
fructokinase (similar to the Mechanism 83) and that the The lack of sustained oscillations in the mitochondria
observed oscillations in DPNH are due to projection. and intact cell experiments can be interpreted several
ways. Since the oscillations are only initiated by some
Biological form of stimulation (addition of valinomycin for the
In regard to biological phenomena, the results ob- mitochondria and anoxia for the intact cells), they may
tained here provide a basis for some interesting ques- be derived from mechanisms which can never exhibit
tions and speculations. It is evident that oscillating sustained oscillations or from mechanisms which could
reactions could provide a physical chemical basis for the but do not have appropriate values of the SCV. Since
variety of gross periodic phenomena such as evidenced the experiments involve large populations of cells (or
by biological “clocks.55 Stability of period for such a mitochondria), there is also the possibility that individual
clock is a natural property of the associated limit cycles. cells are continuously oscillating, but with different
Enzymatic reactions at key places may maintain the and randomly distributed phases so that the population
necessary stationary-state condition in spite of fluctua-
tions in the environmental variables (76). The very
low frequencies of biological clocks can be derived from
relatively slow reactions (such as protein synthesis) or
through the development of beat frequencies. How-
ever, many more experiments will be required to deter-
mine if chemical reactions alone provide the source of
such periodicities.
The simplicity of the feedback pathways required for
oscillating reactions and the known occurrence of such
pathways in a large number of cellular reactions make Figure 32. Experimentally observed oscillations in K+ and H+
it curious that oscillatory kinetics have not been ob- concentrations of respiring mitochondria in the presence of valinomycin,
served more often. In part, this is due to experimental as obtained by Chance

VOL. 59 NO. 5 MAY 1967 53

 average would be nonoscillatory. The stimulus could APPENDIX I

provide a moment of phase of synchronization which The proofs of the Theorems 53 and 54 given in the
would disappear in time because of minor differences
text are developed below. They depend on the two
among the individual cells. However, microfluorom- well known theorems from nonlinear dynamics, which
etry of single cells (77) indicates that they are not con- can be found elsewhere (70). We will limit our basic
tinuously oscillating, although the experiments may not
study to the two-variable interactions so that the basic
be conclusive. Intracellular compartmentation and
differences in individual mitochondria and/or enzymes equations take the form
introduce the possibility that local regions of a single cell vx =
X(x,y)
continuously oscillate but again go unobserved because
of asynchronism. Such possibilities seem unlikely, and vy
=
Y{x,y)
there is no experimental evidence to support them. where Xx, Xy, Yx, and Yy represent the various partial
The physiological significance of the observed oscilla- derivatives.
tory kinetics is not clear since they thus far have only The two theorems of interest here are
been obtained under relatively abnormal conditions. Poincare Theorem: If C is a closed trajectory, then it
Neglecting some of the subtleties mentioned above, the must contain at least one singular point the index of
direct interpretation of the existing facts is as follows: which is +1. This requires that
The observation of oscillatory kinetics in cells, cell ex-
A* =
(XxYv
-

XvYx)* > 0
tracts, and related systems under abnormal physiological
conditions is a manifestation of the appropriate feed- If C is closed trajectory then the
Bendixson Theorem: a
back pathways which exist under normal physiological
function J =
Xx + Yy musteither be identically zero
conditions but with the values of the SCV such that in the
or must change sign region (R) enclosed by C.
stable, nonoscillatory states are obtained. The inter- of definite character. A region of definite
Regions
pretation has a close dynamical analogy in electronics. character is defined as a region throughout which the
A feedback circuit designed for amplification can become
an oscillator, when the circuit parameters are perturbed
partial derivatives (Xx, Xyi YX} Yy) do not change.
The specific signs are conveniently designated in
or modified, and thereby ceases to fulfill its design pur-
matrix type form—e.g.,
pose. The “purpose” of feedback in cellular reactions is
certainly not clear, but it appears at this time, that read as the signs of
oscillatory reactions observed under abnormal physio- ^
logical conditions may represent a breakdown in the
desired functioning of such pathways. Special Systems. A system is said to be special when
the boundaries of regions with definite, but different
Theoretical Methods
character impose special nonarbitrary relationships
An important theoretical aspect of this paper lies in the between the various partial derivatives. Otherwise,
techniques utilized for the analysis of complex reactions. the system is said to be nonspecial. For example, the
The various reduction theorems and approximations system
can be readily applied to kinetic analysis of any type
of reaction scheme. The analysis of complex kinetic
patterns can be likened to the analysis of optical spectra
or the many particle problem. Given some basic in-
formation such as atomic structure or the force laws
(reaction mechanism), approximations such as LG AO
are made to reduce the complexity of the problem
(usually to some form of two-variable interactions).
The resultant predictions are compared to the experi-
mental results and the approximations are improved is general (nonspecial) while the system

by treating other variables as perturbations on the basic

solutions. If the approximations are known to be good,
then various deductions can be made as to the nature of
the interactions. The essence of the problem is to select
the best approximation to start with since perturbation
theory becomes too involved to be carried very far. In
physics such a problem leads to distinctions between is special since the junction of the two regions, where
strong and weak interactions. In the analysis of kinet- two signs change at once, requires a boundary along
ics, it leads to distinctions between fast and slow reac- which both Xy(x, y) and Yy(x, y) are simultaneously
tions (in terms of characteristic times) and weak and zero in contrast to the nonspecial system. Such special
strong control. The final result of such analysis is to relationships do not generally exist for arbitrarily chosen
provide a knowledge of the important variables and an functions and to avoid unnecessary rigor it is convenient
understanding of the behavior in terms of those variables. to dispose of them. Such special cases should be treated

54 INDUSTRIAL AND ENGINEERING CHEMISTRY

 By the Poincare Theorem, the singularity must have
an index of +1, and thus
A =
XxYy XyYx > 0
-

at the singularity.
Because XxYy < 0 because of the opposite signs, it
follows that

XyYx < 0
at the singularity—i.e., Xy and Yx must also have
opposite signs at the singularity, and because of the
definite character, everywhere within the region. This
Figure 33. Turning points of a closed cycle
proves Part a of the theorem. Part (b) follows immedi-
ately from Part (a) and the condition A > 0.
Theorem II: For the general system, if C is a simple,
smooth, closed trajectory then it must (a) encompass a
region (.R) which contains subregions such that Xx and
Yy have opposite signs; and (b) the trajectory C must
pass through regions for which Xy and Yx have opposite
signs.
Proof: The proof of (a) follows directly from the
Bendixson Theorem utilized as in the proof of Theorem
I. The proof of (b) follows.
If a cycle is a simple smooth curve then it must possess
at least four “turning points” (as indicated in Figure 33)
Figure 34. Time domain for trajectory of Figure 33
which can be ordered such that
Point 1: vx > 0; vy
=
0; vy < 0

separately or, as we shall do here, incorporate them into 2: vx

=
0; vy < 0; vx < 0
the theorems as limiting forms of the nonspecial cases.
3: vx < 0; vy
=
0; v > 0
Theorem I: If a closed trajectory exists in a region of y

definite character which contains one and only one 4: vx

=
0; vy > 0; vx > 0
elementary singular point, then:
in the time domain. A simple system would appear as in
(a) The region must have a character of one of the
following forms Figure 34.
That the cycle must contain at least four turning

(!!),(: x x X)
points is considered self-evident. It amounts to a two-
dimensional version of Rolle’s Theorem. From the
basic equations
i.e., Xx and Yy must have opposite signs and Xy and Yx
must have opposite signs as well. vx =
X(x,y)
(b) In addition, the magnitude of the product of the Vy
=
Y(x,y)
cross derivatives must be greater than the magnitude of
the product of the self-derivatives—i.e., it follows that
vx =
vxXx + vvXy
|XyYx\ > |XxYy\
Vy VX YX + VyYy
Proof: If the region contains a closed trajectory, then
by the Bendixson Theorem, Applying the conditions of point to the above equations 1

j
we conclude that Yx < 0 and from point 2 that Xy > 0.
=
X, + Yy Points 3 and 4 simply reiterate the previous results.
must change sign within the region. For general systems it follows immediately that the cycle
Hence, passes through regions for which Yx and Xy have opposite
signs—e.g., for general systems, the transition from
Xx + Yy
= 0

for some values of * and y within the region. Because

the system is general, the possibility that Xx and Yy (- ~) (+ +)
are simultaneously zero is eliminated. Thus, Xx and implies the existence of a region with the character
Yy must have definite and opposite signs at these zeros of
J. Since the region is of definite character, these signs
must prevail throughout the region, and hence also at (+') (-+) -

the singularity. This completes the proof of Theorem II.

VOL. 59 NO. 5 MAY 1967 55

APPENDIX II—ANALYTICAL ANALYSIS OF Henceforth, all discussion will be in the barred (-)
OSCILLATORY REACTIONS variables, although the bar will no longer be noted.
Two of the mechanisms described in the text (83, Graphical Aspects. A rough sketch of the rate laws
106) are more fully analyzed in this appendix. While (Equations 150) is given in Figure 35. Note that the
computer studies are a virtual necessity to determine the asymptote of ^3 is p while the maximum possible value of
exact nature of the solutions and even provide a proof V2 is one for x9 y — 00, and that /3 and 8 may be greater or
that a limit cycle does exist, such studies do not provide less than one. The singularities (x y 0) are deter- = =

a very clear understanding of the dependence of the mined by the intersection of the curves as follows: the
solutions on the various parameters. In addition, it is intersection of v\ 8 with gives the value of y*; the
=

possible to miss some interesting properties of the solu- value of v is determined to make V2 (**, y*) 8 (as —

tions if certain values of the parameters are overlooked indicated in Figure 35). It is obvious that 0 > 5 and
or cannot be readily realized on the computer. The 8 < 1 if the singularity is to exist for positive values of

techniques utilized in the text provide a rapid, but v and y.

very loose appraisal of the potential properties of a To a large extent, the possible character of the singu-
mechanism. Although the analysis given in the appen- larity and the regions of definite character can also be
dix is considerably more involved, it does provide some realized from the slopes of the curves. Such simple
quantitative details, although the emphasis is still on the graphical considerations will facilitate understanding of
possible kinetic behavior. More sophisticated mathe- the analysis which follows.
metical techniques can be applied to obtain more Singularities. The exact values of the singularities
quantitative relationships although such effort seems are readily obtained from the Equations 150 to be

warranted in view of both the chemical and mathemat-

*
* &(y* + «)
ical approximations used to develop the basic equations. y = =
(152)
J3
—

§ va -

$)

BACK-ACTIVATION OSCILLATOR and the conditions

Basic Equations. The mechanism given as 13 > 8 and 5 < 1 (153)

are necessary for the singularity to exist in the positive

[G]
—^
*
x
rTY
—^
*
—^
(145)
quadrant. If the Conditions 153 are violated, then
1 2 3
either x or y or both will simply increase indefinitely in
time. Hereafter, we shall assume that these conditions
are satisfied.

satisfies the rate laws (see Tables I and II) Regions of Interest. The breakdown of the phase
plane into the regions of definite character is easily
a x y af y realized. The functions XX9 Yx and Xv have a fixed sign
Vl =
k = =
(146)
V2
7T—j-7—[—
y(x + b) + c y + b over the entire positive quadrant as can be seen from
and has the basic equations
x =
v\ —

V2 ~

X(x, y)
(147)
y =
V2
—

v%
=
Y(x,y)
Transformation of the Variables. The study of
these equations is facilitated by the following trans-
formation
x _ y a
* y t (148)
i'1
= = =

T T’ —
Z

which yields the following equations aP \S

1
x = —

[n —

r
y v Vo (149)
—
—

where

y x Py Figure 35. Upper sketch shows rate laws for back-activation mech-
v\ =
8 V2 =
Vz (150) anism
y (1 + x) + a 1 + y
Several values of x are indicated for the function V2(x, y). Asymptotes for
and the various curves are indicated. The value of y* is determined by inter-
section of lines a A and v%(y); the value of x is then chosen to make
=

c a k vfx, y) intersect at the same point as indicated by x*. The lower sketch
shows function Z(x, y) and function Z'(y) used to determine regions of
« =
777
bb
0 = ~

a
{ =

P
5
a
(151)
definite character Yy

56 INDUSTRIAL AND ENGINEERING CHEMISTRY

the monotonic character of the functions X(x, y) and
Y(x, y). The quantity Yy can be of either sign and
requires further analysis. That function can be ex-
pressed as

Y Yy = Z ~
Z'
13

x/a/3 1 a
where Z =
Z'
^ and a =

(y/a' + 1)
,5
(1 + ^)2
,
V+x
(154)
Both Z and Zr have the same general form as a func-
tion ofy and are sketched in Figure 35B. There can be
at most one intersection of Z and Z', although there
may be none depending on the values of a, 0, and x.
Simple examination shows that
if x/a0 < 1 then Z lies below Z' for all y if a' < 1;
that is, v > a 1 —

Z lies above Z' fory > y' > 0 if

a > 1 ; i.e., x < a 1 —

where y' is some positive value depending on a, p, and y.

If x/afi > 1 then Z lies above Z' for all y if af > 1;

j
i.e., x < a
l 1 —

Z lies below Z' for y > y' > 0 if

j
{ a' < 1; i.e., x > a 1 —

The result gives rise to three possible cases, which are

Figure 36. Sketches of definite character and lines of interest according
sketched in Figure 36, accordingly; to three possible cases

a —

1 Values of XT, Xy, and Yx are definite throughout entire plane. Quantity
-

> 0 > 0 a) Yy may take either sign. Cross-hatch regions indicate positive values.
a
Appropriate asymptotes are indicated on the right-hand side. Some typical
trajectories are also shown
a —
1
0 > -

> 0 (2)
a

which must be satisfied if oscillations are to occur. It is

a0 > 0 > a —
1 (3)
easily verified that there exist values of the parameters
Critical features of the regions, such as asymptotes and which can fulfill Condition 158.
starting points are indicated in the figure. Finally, it may be noted that A XxYy XyYx > 0= —

Evaluating Yy at the singularity yields over the entire positive quadrant and that the rotational
^
condition can also be satisfied as proved in the text
Yy* > 0 for < ~-1 (156)
/3 (see Equation 91).
CL
The Phase Plane and Its Trajectories. Consider-
1 able insight into the possible motion can be obtained
Yv* < 0 for /3 > (157)
with relatively little effort by the well known technique
a
of plotting the direction vectors along special lines in the
Thus, the oscillatory conditions (for which Yy* > 0
phase plane. The direction vectors have an x axis
can only be fulfilled in Case 1 (Equation 155) for which
component proportional to vx and a y axis component
P < (a l)/a. At most, only damped oscillations may
—

proportional to vv. As the name implies, the vectors

occur in the other cases. The requirement that Yy* > 0 indicate the direction and magnitude of motion at that
is a necessary, but weak, condition. The stronger
point in the phase plane. The simplest lines of interest
requirement is that B* X* + Fy* > 0, which yields =
are the axis (* 0 and y =
0) and the lines of x
=
0 =

the more involved condition:

and y = 0. Possible shapes of these lines are sketched
8 1 in Figure 36 subject to the condition 8 < 1 and P > 8.
= ’
X
W+ (0 -

8) a] Jp The line x of hyperbolic shape with the vertical

~
0 is
-
1
axis (y =
0) asymptote and the line [*
as one 8/ =

(1 ~

8Y p + af (P
-

8Y > o (1
—

8) ], as the other asymptote. The line y 0 is also =

*}} hyperbolic and always has x ap aty 0. However,

= =

(158) there are three cases corresponding to Equation 155.

VOL 59 NO. 5 MAY 1967 57

 projection of x =
0 asymptote. Segment 2 is projected
with a slope of —

1/f from the intersection of segment 1

and the line y y*. Segment 3 can be chosen arbi-
=

trarilyclose to the *-axis. It is clear that trajectories

cannot actually reach the axis since for small y, the equa-
tions become x 8 and y
=
y(x/a (3). As y —>
= —

0 so also does y; the * directed motion becomes the

dominant feature and as soon as x > a(3, the y motion
becomes directed toward increasing values of y. Since
segment 3 is not identical with the x axis, it must even-
tually intersect the line x 0; segment 4 is projected
=

from that point of intersection. Segment 5 has a slope of

(
—

1/f) and begins at the intersection of segment 3 and

the line y =
y*. At the point where segment 5 inter-
sects the line y 0, segment 6 is projected and the
=

boundary is closed.
It is clear that the direction vectors cut into (or are
tangent to) the boundary at all points on it except for a
small region at the intersection of segments 2 and 3;
y* y—
the size of that region approaches 0 as segment 3 is
Figure 37. Phase plane showing limit cycle (dark solid line) for moved closer to the line (y 0). Consequently, vir-
=

back-activation oscillator. Dashed lines represent typical trajectories. tually all trajectories which start within the boundary
Light solid lines indicate boundary used for proof of limit cycle existence are thereafter confined within the boundary. By cor-
ollary (Equation 58), a limit cycle must exist if B* > 0.
The same definition of a boundary can be applied to
In Cases 1 and 2, the line x (3/(1 (3) is an asymptote
= —

the other cases (2 and 3 of Equation 155) to demonstrate

and the curvature is positive (Case 2) or negative (Case that the trajectories cannot go off to infinity, the
1). For Case 3, the liney I/O? =
1) is the asymptote.
—

only exception being a trajectory which starts on the

The relationship between the lines and the regions has line y =
0. It is also clear that trajectories which start
only been subjected to the conditions that /?/(1 (3) > outside the boundary and in the neighborhood of y
—

0 =

8/(1 8) since (3 > 8 by Equation 153; and that in Cases will approach very large values of x (in all cases) and
—

2 and 3, the singularity cannot occur with the region of theny (in Cases 1 and 2) before being trapped within the
Yy > 0, while it must in Case 1.
boundary due to the x 0 asymptote (segment 1).
=

Along the axis one obtains While the preceding analysis proves that a finite
(3y
boundary cycle exists, it does not provide a specific
for v =
0 x =
8/£ j —
~

~—;— cycle, since segment 2 is projected to an unknown dis-

1 + y
tance from the y 0 line. However, a quantitative
=

y = 0 x —

<5/f y =
0 (159) boundary can be easily obtained.
From the basic Equations 147, the equations for the
trajectories can be written as
For this particular set of equations, the line y y* =

is also of interest since the direction vectors have a slope

of (—T/f) at all points along it. The nature of a few
typical trajectories is sketched. In Cases 2 and 3 the where y on the right-hand side should be considered as a
singularity may be a stable node or focus. function of x, y; that is, y V2(x\ y) =
v%(y). Recall —

that the slope of the direction vectors is given by tan

Proof of Limit Cycle
0 =
x/y. The equations for the boundary line in the
The case of most interest here is shown in Figure 37, region of segment 2 must be such that
for which it is possible to have B* > 0 giving rise to an
unstable focus. A proof that a limit cycle exists when 0 > tan 0B —

7^ > tan 0 (161)

2?* > 0 is based on Theorem 57. There are several yB
ways to construct a boundary cycle. The first analysis to ensure that the direction vectors will cut into the
will be just a proof that the trajectories cannot go off to
boundary line. The subscript B refers to the equations
infinity. Such a proof is sufficient to guarantee the for the boundary line. Any approximation which
existence of a limit cycle, but gives no information as makes Equation 160 integrable and which satisfies
to the boundaries of the limit cycle. The second type of
Equation 161 can be used to generate segment 2 of the
analysis provides quantitative values of the boundary
boundary line. Taking
cycles.
In the first case, the line labeled “outer boundary
cycle” is constructed as follows. Segment 1 is just a
-"sW (162)

58 INDUSTRIAL AND ENGINEERING CHEMISTRY

on the right-hand side of Equation 160 will fulfill the FORWARD INHIBITION OSCILLATOR
the stated requirements since x is always greater than
Basic Equations. The mechanisms
8/(1 —8) in the region of interest. Integration of Equa-
tion 160 with Equation 162 substituted for y(x, y) on
the right-hand side, yields an equation of the form
^_mnh(ES)
V2 + C I E E
*i) B In (163)
A(x2
[GKXVT (168)
— —

_yi + C_
Y Y
where
has rate laws given as

A = 1 f*P (1
~ ~

S\
ay
a8{ (1-8) j v\ —

ky v2 =
Vz =
(169)
b + cx + dx2 y + b'

B-(-?-) Wl-S)-1] 1 V ' J

c -
“W -

J--8 and satisfies the equations

\p 8/ -

0 -

8
X =
Vi —

v2
(164)
y = 2 v2 —

v\ —

vz (170)
The indexes 1 and 2 refer to two points on the resultant Transformed Variables. It is convenient to trans-
boundary line sketched at segment 2' in Figure 37. form to new variables given by
Take point 1 as having the coordinates x\ 5/(1 5), = —

yi y* 8/(P
=
8). The point 2 should be determined
= —

by the intersection of Equation 163 with the line of y =

y =
y/bf x
*‘Vt
\j——
x t =
kt (171)

0, leading to rather involved arithmetic. Such compli- and to define the transformed rate laws by
cations can be avoided by recognizing thaty2 will usually
be small and that *2 can be approximated by x2 aP- = fix ay
V\ =
y v2 Vz =
(172)
Solving Equation 163 fory2 and taking y2 + c c yields ~
1 + fx + x2 1 + y
lny2 = —Ar + 2? In C' + lnyi (165) Thus the new equations of motion are

where x =
5[ih —

v2] =
X(x, y)
(1
~

S)
-
5\2 aP (1 -

8) /’ = 2 v2 —

V! —

Vz =
Y(x, y) (173)
_f (aP =

a8\ 1-5 J ap (1
-

5)
-
5
where

yi =
(166) b
^ ^ =

b'^. 7
kb' b’ kVbd y/bd
and B is given by Equation 164.
(174)
Thus, for a 4, P 1/2, f =
10, 5 V4 one finds = = =

that y2 e~27 while for a 12, (3 V2, f 1,5 V4, — = = =

Henceforth, all discussion will be in the bar (-) varia-
theny2 e~10. The remainder of the specific boundary
^
bles although the bar will no longer be written.
can be obtained by projecting this value of y around Graphical Analysis. The analytical discussion is
the boundary cycle according to the previous definitions considerably facilitated by simple graphical analysis.
of the segments. Upper bounds on the possible values The rate laws are graphed in Figure 38. The v and y
of the v and y excursions of the limit cycle can then be scales are not the same and cannot be compared directly.
obtained.
More restrictive boundaries can be obtained by
iteration of the differential Equation 160 and by using v,(y)
similar techniques for the other segments of the boundary
cycle.
Frequency Approximations. In accord with Equa-
tion 66, a first approximation to the frequency is given by:

VA* = (1
-

«)(/3 -

5) X
Vf/3
=
x 1
1
(167) y* y —

V:1 + (/3/5 -

1)

However, the formula is restricted to small amplitude Figure 38. Graphical sketch of rate laws for forward inhibition
oscillator
oscillations. Note that the <3 given by Equation 167
is in the barred variables. Both x and y are plotted along abscissa and their scales are not necessarily
Conversion to original the same, y* and v* are determined by intersection of v\(y) and Vz(y).
variables is given by co aco/b'. =
The x* are determined by projecting v* into vFx)

VOL. 59 NO. 5 MAY 1967 59

 Line Equations Properties
a y —
oo for y — oo 8 for y
x 0 y =
-y -

=
1 + y y 8
for y y for y 0
$ =
8y
—

(at + 1 )y 0 x =0 a + 1

for x — oo
2 fix
= 0 Max at x = 1 —
=
—8/2 always
1 + f* + *2
fix for $
!02 x — 0

8fix as for y but

X
1 + fx + x2 — 8 fix for x 0

x = 0 y =
Vi(x) See Figure 38 and apply graphical analysis to determine shape
of line

y)i=o =
y
—

v3(y) Changes direction as y passes y*

y —
0 y + vz(y) = 2 v2(x)

x)y=0 8/2 v3 (y)] Changes direction as y passes y*

—

[y
—

The curves retain the shapes shown for any nontrivial quadrant, it is necessary that
values of the parameters and the maximum in the
a > 1 (178)
v2{x) curve occurs for x 1. The occurrence of the =

singularity requires that x y 0 and consequently = —

b> 2 (179)
that vi vz v2
=
v*. For the singularity to exist,
= =

the former (178) being the condition that v\ and

it is clear that the intersection of v\ and v% must provide a
value of v* which lies below v2m. When that condition possess an intersection other than the origin and the
latter (179) being the condition that the intersection
is fulfilled there will be two possible values of x labeled
lie below v2m.
x-* and x+*. The origin (x y 0) is also a singu- = =

Domains. The values of the partial derivatives Xx,

larity. The breakdown of the phase plane into regions
Yx, Xv, and Yy are easily derived. However, it is clear
of definite character can also be realized from the graph.
from Equation 173 and the shape of the curves (Figure
In terms of the parameters, the graphical features have
38) that the quantities Yy and Xy have fixed signs over
the following values:

fi
Initial slope of vz(y) = ol (175)
2 + f
v> m
= a

Singularities. Solving the equations x y 0 = =

yields three possible singularities S+, S~) according to

(y* =
a —

S± =
\ *_fe±V/62 —
4 and S0 =

(X± 2

(176)

where
Figure 39. Sketch of regions of definite character for forward in-
hibition oscillator
b = -

i 077) Note change of character for x > or < / and that B can be > or <0 for
a —

1
x > /, but must be <0 for x < 7. Note also the three singularities at the
For the singularities to be real and exist in the positive heavy dots. Cross-hatched regions indicate where A < 0.

60 INDUSTRIAL AND ENGINEERING CHEMISTRY

the entire positive quadrant while Xx and Yx change
signs as * passes through the value x 1. =

The values of A and B are easily shown to satisfy

A =
XxYy -

XyYx =
-8(v2,x)[l
-

0*|,] (180)
B =
Xx+ Yy =
-8v21x
-

1 -

vz\y (181)
where v2\x ^ dv2/dx, and
1 —

x2) o:
- and a'' =
<182)
[i + (TT#
the sign of A changes with the sign of V2\x (he., x 1) =

or when v^y 1, which requires y

= 's/a 1. The = —

condition for B to change sign is more complicated, but

it is easily demonstrated that the curve of B 0 always —

lies above the line x 1, a result which is obvious =from

the basic equations. The singularity (6+) may or may
not be included in the region of B > 0, depending on
the values of the parameters. Again, it is easily realized
that the rotational condition (lA^T^ > I^TyJ) can be
satisfied at the singularity S+, but not at So or S-. The
properties of the singularities can be summarized as:
s A B Comments limit
Figure 40. Phase plane showing cycle (dark solid line) for
So <0 <0 Nodal point, always forward inhibition oscillator
S- >0 <0 Saddle point, always Dashed lines represent typical trajectories. Line of x =
0, y =
0 is also
indicated
S+ >0 ^0 May be a stable or unstable
focus depending on values of
parameters
A breakdown of the various domains is shown in Figure point moves very slowly in time as it passes that singu-
39. with the values of x(t) and y(t) being nearly con-
larity
The condition for the singularity, S+, to be unstable stant in the time domain (kinetics).
and hence the possibility of a limit cycle is that For different values of the parameters a smaller limit
cycle may be obtained being close to the singularity,
CL + 1 8(cx. —

l)2 S+, and the oscillations will be more nearly sinusoidal.

B+* =--—
+ ---
.

(1
-

X-2) > 0 (183)

In such case the frequency will approach the value given
a (3

by:
where the x_ is defined in terms of b by Equation 176. 8(cl —

1)^
Phase Plane Structure and the Trajectories. The & =
A+* = —-—

(1
-

x_2) (184)
ap
complete phase plane structure and the general shape
of the trajectories can be readily established as shown in
The inappropriateness of such frequency formula for
Figure 40. The lines of interest and the magnitude of
the rate vectors are listed on the previous page. large trajectories is particularly clear in this example.
The frequency is greatly changed for trajectories which
The slope of the line y 0 at the origin equals
the saddlepoint singularity (£-), but the
—

pass near
(a + l)/2 /3 while that for x 0 is given by 1//3; hence =

values of co2 *
A+ do not in any way reflect the proper-
=

dx dx
the condition that is that a > 1 which is ^ ties of that singularity. Note that Equation 184 gives
dyx= 0 dy y' = o the value of do (in the barred variables). Conversion
again the requirement that the singularities S± exist. to the unbarred variables is given by co kd). =

The dashed lines in Figure 40 illustrate some typical

trajectories, while the heavy line is representative of a
possible limit cycle similar to that actually obtained
APPENDIX III—ANALOG COMPUTER
(see Figure 14). The kinetic curves can be deduced
from the trajectories by noting that short direction SOLUTIONS
vectors imply slow movement in time. The reason for The Johnson Foundation Electronic Analog Com-
the long delay in each cycle (see Figure 14) is also clear. puter, Mark II (74), is designed for the study of chemical
For the large limit cycles, the trajectory passes close to reactions based on the law of mass action. The reduced
the saddlepoint singularity (S-); as in the neighborhood equations utilized in the theoretical analysis cannot be
of any singularity, the rates approach zero and the phase directly programmed on the computer, and conse-

VOL. 59 NO. 5 MAY 1967 61

quently the complete set of chemical reactions must be a
x
= '

as' a (185)
-

solved. Rate constants and computer scale factors are a

generally chosen to provide large (greater than 10)

ratios between the characteristic times for enzymatic where a is the
concentration of chemical A, x is the dis-
intermediates and the basic variables (x, y). Neverthe- tance in centimeters from 0 to the value of the curve, d
is the full-scale distance (normally 5 cm.), as is the scale
less, the ratios are not infinite and the reduced equations
do not hold exactly. Some small effects due to the factor of the reacting chemical and a is the arbitrarily
chosen concentration unit—e.g., mmoles, mmoles, etc.
“many variables” are still present. A more important
source of differences between the computer solutions and A fraction put before a scale factor gives initial level of a
the true solutions of the reduced equations is due to the chemical as the fraction of full scale.
fact that the steady-state equations are only a good The conversion of computer time to chemical time is
approximation to the correct reduced equations which given by
are special limit lines. The exact nature of the correct
reduced equations is discussed elsewhere {15). How- t = a (10)3 (computer sec.)-1 T (186)
ever, the differences in fundamental forms is relatively
minor in the cases discussed and would not greatly affect where a is the time-scale factor (any value may be used)
the conclusions, although the computer solutions would and has units of chemical time and T is measured in
differ in minor respects. A careful comparison of these computer seconds.
effects have been studied on a digital computer and Full scale chemical values and the time-scale factor
are reported elsewhere {15). {a) may be chosen in accordance with chemical reaction
data. When this is done, rate constants of mono-
Analog computer studies of complex reactions are k
subject to additional errors which relate to the interaction molecular reactions {A —*) are computed by using
of computer errors and the differential equations under
study. For stable reactions or in the neighborhood of ,
K
stable singularities, the effect of computer errors k = -

(187)
(in asa
stability and accuracy) is usually reduced. On the
other hand, for reactions with unstable singularities
where K is the computer rate constant, as is the computer
the computer errors are generally enhanced and may
scale factor of the reacting chemical, and a is the time-
cause drastic changes in the solutions obtained at
scale factor. k
different times. Such errors also cause shifts in deter-
For bimolecular reactions (A + B —x),
mining the boundaries between regions of stable oscilla-
tions and damped or undamped oscillations. K
The computer solutions presented here have all been k = —-

(188)
asbsa<7
repeated more than once with exceedingly small differ-
ences, although there may be particularly sensitive It should be noted that when a chemical is denoted by
rate constants which require adjustment on different [A] =constant (3 (0 < (5 < 1.0), the reaction becomes
=

occasions (typically a sensitive K of 0.032 may change to order and the effective rate constant is found
pseudo-first
0.037). While such error limitations do not particularly by K* (3K.
=

affect the theory and may be considered mostly an

annoyance, the reader who wishes to study or reproduce REFERENCES
these results should be aware of such problems. (1) Bak, T., “Contributions to the Theory of Chemical Kinetics,” Fr. Baggs Kgl.
Regardless of the minor annoyances associated with Hofbogtrykkeri, K0benhaven, 1959.
(2) Bennister, T. T., Biochem. Biophys. Acta 109, 97 (1965).
analog computers, their usefulness in the study of com- (3) Betz, A., Chance, B., Arch. Biochem. Biophys. 109, 585 (1965).

plex reactions cannot be denied. The wide variety of (4) Bierman, A., Bull. Math. Biophys. 16, 203 (1954).
(5) Bray, W. C., J. Am. Chem. Soc. 43, 1262 (1921).
kinetic behavior for such mechanisms and the complexity (6) Chance, B., Estabrook, R., Ghosh, A., P.N.A.S. 51, 1244 (1964).
of the dependence of such behavior on the rate constants (7) Chance, B., Ghosh, A., Higgins, J., Maitra, P. K., Ann. N. Y. Acad. Sci. 115,
1010 (1964).
frequently requires a complete exploration of the kinet- (8) Chance, B., Hess, B., Betz, A., Biochem. Biophys. Res. Comm. 16, 182 (1964).
(9) Chance, B., Holmes, W., Higgins, J., Connelly, C. M., Nature 182, 1190 (1958).
ics over several thousand settings of the SGV.
(10) Davis, H. T., “Introduction to Non-Linear Differential and Integral Equa-
The location of an oscillatory region, for example, can tions,” Dover, New York, 1962.
(11) Frenkel, R., Arch. Biochem. Biophys. 115, 112 (1966).
be viewed as “fishing” in the multidimensional SGV (12) Frisch, L., Ed., “Symposium on Quantitative Biology,” The Biological Labora-
tory, Cold Springs Harbor, L. I., N. Y., 1961.
space. The fish can be “played in” as in going from (13) Hearon, J. Z., Bull. Math. Biophys. 15, 121 (1953).
damped oscillatory solutions to sustained solutions by the (14) Higgins, J., “Computers in Biomedical Research,”*Chap. 4, Vol. II, Academic
Press, New York, 1965.
adjustment of SGV to achieve less damping. The rela- (15) Higgins, J., “Control of Energy Metabolism,” Chap. 1, Academic Press,
New York, 1965.
tively high speed of analog computers (compared to (16) Higgins, J., Ph.D. thesis, University of Pennsylvania, 1959.
digital), the ease of changing SGV and the immediate (17) Higgins, J., P.N.A.S. 51, 989 (1964).
visual display of the solution make such studies practical. (18) Lotka, A. J., J. Am. Chem. Soc. 42, 1595 (1920).
(19) Minorsky, N., “Non-Linear Oscillations,” Van Nostrand, Princeton, 1962.
The settings of SGV for the computer solutions in the (20) Prigogine, I., “Thermodynamics of Irreversible Processes,” Wiley, New York,
1955.
text are given below. While the values of the computer (21) Rice, F. O., Reiff, O. M., Phys. Chem. 31, 1352 (1927).
parameters are given, they can be converted to chemical (22) Volterra, V., “Theorie Mathematique de la Lutte la Vie,” Gautiers-Villars,
Paris, 1931.
parameters by the following formula: (23) Yamazaki, I., Yohota, K., Nakajima, R., Biophys. Res. Comm. 21, 582 (1965).

62 INDUSTRIAL AND ENGINEERING CHEMISTRY