EFFECTIVE KINETICS OF CHEMICAL REACTION

NETWORKS
arXiv:2402.11762v1 [q-bio.MN] 19 Feb 2024

TOMOHARU SUDA

Abstract. Chemical reaction network theory is a powerful framework

to describe and analyze chemical systems. While much about the con-
centration profile in an equilibrium state can be determined in terms
of the graph structure, the overall reaction’s time evolution depends on
the network’s kinetic rate function. In this article, we consider the prob-
lem of the effective kinetics of a chemical reaction network regarded as
a conversion system from the feeding species to products. We define
the notion of effective kinetics as a partial solution of a system of non-
autonomous ordinary differential equations determined from a chemical
reaction network. Examples of actual calculations include the Michaelis-
Menten mechanism, for which it is confirmed that our notion of effective
kinetics yields the classical formula. Further, we introduce the notion
of straight-line solutions of non-autonomous ordinary differential equa-
tions to formalize the situation where a well-defined reaction rate exists
and consider its relation with the quasi-stationary state approximation
used in microkinetics. Our considerations here give a unified framework
to formulate the reaction rate of chemical reaction networks.

1. Introduction
Chemical reaction network theory is a useful framework to describe chem-
ical systems [1, 2]. Once constituting reactions are specified, we may use
it to deduce the dynamical properties of the system. Remarkably, knowing
the topological properties of the network is often sufficient for determining
the long-time behavior, and detailed knowledge of the reaction speed is not
required [3]. Here, we note that long-time behavior studied in the chemical
reaction network theory is largely independent of time scale. This is be-
cause it is defined in terms of the limiting behavior of a dynamical system,
which is actually an infinite-time behavior and, therefore, invariant under
the rescaling of time.
The question of kinetics is different. Here, the main concern is finite-
time behavior, which is heavily dependent on the time scale and, therefore,
beyond the scope of usual dynamical systems theory. Further, obtaining a
mathematically well-defined notion of kinetics itself is not very straightfor-
ward. Here we recall the standard definition of the reaction rate, which is
given via the time differential of the extent of the reaction, which is in turn
defined only for a single reaction [4, 5]. Chemical reaction networks rep-
resent composite reactions, and therefore, it is not obvious how the notion
1
2 TOMOHARU SUDA

of reaction rate can be applied. In addition, we notice that there is much

ambiguity as to what reaction is elementary.
That said, the reaction rate for composite reactions has been considered
for various chemical reaction networks, including the classical example of
Michaelis-Menten kinetics [6, 7]. The derivation for such effective rate laws
usually involves the quasi-stationary state approximation (QSSA), where
some of the reaction processes are assumed to be in a stationary state [8,
9]. We notice that the principle working here is to regard a subnetwork
consisting of intermediate steps as a black box conversion system and find
the rate of conversion. The purpose of the present article is to propose a
mathematical framework that generalizes this procedure of abstraction. The
concept of kinetics can be defined as the conversion rate of one set of species
to another by a chemical system. This is experimentally observable and, in
the case of elementary reactions, coincides with the mathematical notion of
reaction rate. This is what we call effective kinetics here.
Naturally, our approach here is similar in spirit to the classical notion of
QSSA and the reduction schemes of chemical reaction networks, in partic-
ular, the recent theory of chemical reaction circuits [10, 11, 12]. The main
difference is that we do not suppose the existence of a steady state. In this
view, the kinetics of an effective reaction may depend not only on the con-
centration profile of input species but also on how they are supplied. This
is most prominent in the case of the delay effect, which has been considered
in connection with the signal processing theory and biochemical oscillators
[13, 14, 15]. When a system has a long chain of reactions, it takes some
time for an input to have an observable effect on the output. Even when
the reactant species are supplied so that the concentration is kept constant,
the overall reaction may not have a well-defined rate. For example, if the
system exhibits an oscillatory behavior, we cannot expect the existence of
such a rate. That said, as we will see later, our notion of effective kinetics
coincides with QSSA under some circumstances.
The rest of the article is organized as follows. First, in Section 2, we
provide definitions of basic concepts to fix notation. Section 3 contains the
definition of effective kinetics and calculation examples for simple systems,
including the Michaelis-Menten mechanism. Section 4 introduces the notion
of straight-line solutions of an ordinary differential equation, which is useful
in describing the problem of the existence of reaction rate. Finally, in Section
5, we consider how our framework relates to the classical QSSA. Section 6
includes concluding remarks.

2. Preliminaries
In this section, we introduce some basic terminology to fix notation.

Definition 2.1. Let A = {a1 , a2 , · · · , aN } be a finite set. The set of all

formal sums of the elements of A with natural number coefficients is denoted
 EFFECTIVE KINETICS OF CHEMICAL REACTION NETWORKS 3

by
NA := {n1 a1 + n2 a2 + · · · + nN aN | n1 , n2 , · · · nN ∈ N}.
The set of all formal sums of the elements of A with real number coefficients
is denoted by RA .
Definition 2.2 (Reaction Networks). A chemical reaction network is a
quadruple (S, R, s, t) such that
(1) S and R are finite sets.
(2) s and t are maps s, t : R → NS .
Here S is the set of species, and R is the set of reactions. Each reaction
ρ ∈ R has the form s(ρ) −−→ t(ρ). For each ρ ∈ R, s(ρ) and t(ρ) are called
complexes. By abuse of notation, we denote by s(ρ) the set of species α
with a non-zero coefficient in s(ρ). The case for t(ρ) is similar.
Kinetics of the chemical reaction networks are described using ordinary
differential equations (ODEs). An important distinction is that of au-
tonomous and non-autonomous equations. While most chemical reaction
networks are considered isolated systems, which are modeled by autonomous
equations, this is not the case for our situation of open chemical networks
where there can be interactions between the outside and inside of systems.
Definition 2.3 (Autonomous and non-autonomous ordinary differential
equations). Let f1 , · · · , fn be functions on Rn . An equation of the form
dxi
= fi (x1 , x2 , · · · , xn ) i = 1, 2, · · · , n
dt
is called an autonomous ordinary differential equation. If f1 , · · · , fn
are functions on R × Rn , an equation of the form
dxi
= fi (t, x1 , x2 , · · · , xn ) i = 1, 2, · · · , n
dt
is called a non-autonomous ordinary differential equation. The prob-
lem of finding functions xi (t) (i = 1, 2, · · · , n) which satisfy the given initial
condition xi (t0 ) = ai is called the initial value problem (IVP).
In what follows, C(X, Y ) denotes the set of all continuous maps from X
to Y where X and Y are assumed to be metric spaces.
Definition 2.4 (Kinetic reaction networks). A kinetic chemical reaction
network is a quintuple (S, R, s, t, K) such that
(1) (S, R, s, t) is a chemical reaction network.
(2) K is a map R → C(RS , R≥0 ).
The dynamics of a kinetic reaction network is given by the following ODE:

dx X
(1) = K(ρ)(x)(t(ρ) − s(ρ)),
dt
ρ∈R
4 TOMOHARU SUDA

where x ∈ RS denotes the concentration vector of species. Here we note

that equation (1) defines a dynamical system on RS≥0 .
Given an autonomous ODE, we may modify the equation so that its
response to external inputs is considered.
Definition 2.5. Let
dxi
= fi (x1 , x2 , · · · , xn ) i = 1, 2, · · · , n
dt
be an autonomous ODE. For an input function I : R → Rk , the induced
system is the non-autonomous ODE given by
dxi
= fi (x1 , x2 , · · · , xn−k , I1 (t), I2 (t), · · · , Ik (t)) i = 1, 2, · · · , n − k.
dt
We define the response R[I; a] : R → Rn of the system under the input
I by
R[I; a](t) := (x1 (t), x2 (t), · · · , xn (t)),
where R[I; a](t) is the solution for IVP of these equations under the initial
condition x(0) = a. We call an index i is terminal if Ii (t) 6= 0 for some t
and internal if Ii (t) = 0 for all t.

3. Effective kinetics of chemical reaction networks

In this section, we introduce the concept of effective kinetics and present
examples of calculations for some simple systems.
3.1. Definition of effective kinetics. Here, we consider a situation where
some species are supplied from outside of the system or concentration is
controlled. The dynamics of such systems are described as open chemical
reaction networks [16, 17].
Definition 3.1. Let (S, R, s, t, K) be a kinetic reaction network. The ki-
netics of an open reaction network with the terminal species Sb ⊂ S and
input Ib ∈ RSb is defined by
dxi X
= K(ρ)(xi , Ib (t))(ti (ρ) − si (ρ)),
dt
ρ∈R

where xi , ti and si are defined by projections to RSi , where Si = S\Sb .

Remark 3.2. It is more common to define open chemical networks using
input by addition. However, we do not take this view here because the main
interest is to find the conversion rate.
We define the effective kinetics by a ‘partially integrated’ solution.
Definition 3.3 (Effective kinetics). Let (S, R, s, t, K) be a kinetic reaction
network. The effective kinetics Es of species s ∈ S under the input Ib of
the terminal species Sb ⊂ S and the initial concentration a is defined by
X
Es [Ib ; a](t) := K(ρ)(R[Ib ; a])Ps (t(ρ) − s(ρ))
ρ∈R
 EFFECTIVE KINETICS OF CHEMICAL REACTION NETWORKS 5

where Ps : RS → R is defined by the projection to the subspace R{s} .

By definition, we have
dxs
= Es [Ib ; a](t).
dt
Definition 3.4 (Average effective kinetics). Let (S, R, s, t, K) be a kinetic
reaction network. The average effective kinetics of species s ∈ S under
the input Ib of the terminal species Sb ⊂ S and the initial concentration a
is defined to be
1 T
Z
lim Es [Ib ; a](t)dt.
T →∞ T 0

When the average effective kinetics exists, it makes sense to discuss the
rate of overall reaction without mentioning time. We note that the average
effective kinetics exists if there is a well-defined rate at which the effective
kinetics converge. Later, we will consider this situation in connection with
the quasi-stationary state approximation scheme.
Here, we note that the initial behavior does not affect the average effective
kinetics.
Lemma 3.5. Let (S, R, s, t, K) be a kinetic reaction network. For all T0 > 0,
we have
1 T 1 T0 +T
Z Z
lim Es [Ib ; a](t)dt = lim Es [Ib ; a](t)dt.
T →∞ T 0 T →∞ T T0

3.2. Examples of calculation for simple systems.

Example 1 (Michaelis-Menten). As a first example, we consider the Michaelis-
Menten mechanism of enzymatic reaction:
k+ l
E + S ←−→ ES −−→ E + P.
k–
The governing equation is given by
d[S]
= −k+ [E][S] + k− [ES]
dt
d[E]
= −k+ [E][S] + (k− + l)[ES]
dt
d[ES]
= k+ [E][S] − (k− + l)[ES]
dt
d[P ]
= l[ES].
dt
When we consider the case with input, we ignore the equation for [S] and
set [S](t) = s(t). Then, we obtain an effective kinetics of the form
R Z t Rr
−k+ 0t s(τ )dτ −(k− +l)t
EP [Ib ; a](t) = lk+ E0 e ek+ 0 s(τ )dτ +(k− +l)r s(r)dr.
0
Here, the nonlinearity of the MM kinetics is reflected in the non-additivity
of the response.
6 TOMOHARU SUDA

If we keep the concentration of S constant in the MM kinetics, this ex-

pression is simplified and we have
lk+ E0 s  
EP [Ib ; a](t) = 1 − e−(k+ s+k− +l)t ,
k+ s + k− + l
where s is the concentration of S. As t → ∞, the effective kinetics ap-
proaches to the classical form of the MM kinetics. This derivation is known
as reactant stationary assumption in the literature [21].
Example 2 (First-order network). For a reaction network consisting of
first-order reactions, equation (1) takes the form
dx
= Ax,
dt
where A is a constant matrix.
Let the controlled value be xb (t) ∈ RSb . We assume that the coordinates
are set up so that the controlled values take indices 1, 2, · · · , k. By splitting
matrix A into four blocks
 
Abb Abi
A= ,
Aib Aii
we see that the equation of the internal species is
dxi
= Aii xi + Aib xb (t).
dt
The solution to this equation is
Z t
xi (t) = etAii ai + etAii esAii Aib xb (s)ds.
0
The condition of consistency is
dxb
= Abi xi (t) + Abb xb (t),
dt
therefore the control has to be chosen depending on ai unless Abi = 0. In
the physical sense, this consistency condition gives the flux resulting from
the change of concentration.
Effective kinetics is given by
 Z t 
tAii tAii sAii
Es [Ib ; a](t) = Asi e ai + e e Aib xb (s)ds + Asb xb (s),
0
where As = (Asb Asi ).
For some systems, the average effective kinetics under constant concentra-
tion control becomes constant dependent on the concentration of the input
species. Phenomena of this kind can be observed, of course, if subnetworks
with stable equilibria exist. However, there is another pattern.
Example 3. Let us consider the average effective kinetics for the network
α k+
A −−→ B −
↽−
−⇀
−C
k–
 EFFECTIVE KINETICS OF CHEMICAL REACTION NETWORKS 7

If we control the concentration of A, the effective kinetics of C is given by

Z t
EC [Ia ; a](t) = αk+ e−(k+ +k− )(t−s) a(s)ds,
0
assuming [C](0) = [B](0) = 0. Thus the average effective kinetics of C under
constant concentration is
αk+ [A]
.
k+ + k−
Note that there is no subnetwork with equilibria. In this case, we can-
not apply the quasi-stationary state approximation. If we apply the quasi-
stationary state approximation to B, we obtain α[A] as the kinetics of C.

4. Straight-line solutions of non-autonomous ordinary

differential equations
In the rest of this paper, we consider the problem of the existence of a
well-defined effective reaction rate under the assumption of constant input
concentration. Such consideration of kinetics can be observed in the classical
pseudo-first order approximation or the derivation of the Michaelis-Menten
law by reactant stationary assumption. Here, we introduce the concept of
straight-line solutions as a mathematical tool to formulate the notion of
“well-defined reaction rate” precisely.
4.1. Straight-line solutions of non-autonomous ordinary differen-
tial equations. The simplest solution of an ordinary differential equation
is the equilibrium solution, which is defined by a zero of a vector field. The
straight-line solution, which we define as follows, can be regarded as the
next simplest kind of solution.
Definition 4.1 (Straight-line solutions). An ordinary differential equation
ẋ = F(t, x) admits a straight-line solution if there is a solution of form
x(t) = ta + c
where a and c are constant vectors.
In what follows, the coefficient a is called the rate of the straight-line
solution.
We note that the existence of a straight-line solution is equivalent to that
of a constant vector c with F(F(c)t + c) = F(c) for all t. In this sense,
straight-line solutions are algebraically determined from the system.
4.2. Affine systems. First, we consider straight-line solutions for affine
systems. Equations of this kind arise when we consider first-order networks
with constant input, where the equation of kinetics typically takes the form
ẋ = Ax + b.
For such systems, the notion of the straight-line solution characterizes the
effective kinetics completely.
8 TOMOHARU SUDA

Lemma 4.2. Let x(t) be a solution of an affine ordinary differential equa-

tion ẋ = Ax + b such that there exists the limit
lim ẋ(t) = a.
t→∞
Then a ∈ ker A. Consequently, there exists a straight-line solution of the
form ta + c, where a = Ac + b.
Proof. First, we observe that
x(t)
lim = a.
t→∞ t
This is established as follows. Let ǫ > 0 be arbitrary. Then, by the assump-
tion, there exists t0 > 0 such that
kẋ(t) − ak < ǫ
for all t > t0 . Therefore we have
1 t kx(0)k
Z
x(t)
−a ≤ ẋ(s)ds − a +
t t 0 t
Z t0 
1 t − t0
≤ kẋ(s) − ak ds + kx(0)k + ǫ
t 0 t
≤ 2ǫ
for sufficiently large t.
From this result, we obtain
Ax(t)
Aa = lim
t→∞ t
ẋ(t) − b
= lim
t→∞ t
= 0.
Finally, a − b ∈ im A follows from the closedness of the image. 
While we do not apply the results obtained here directly to chemical
reaction networks, they illustrate the basic characteristics of the concept.
Theorem 4.3. An affine ordinary differential equation ẋ = Ax + b ad-
mits a straight-line solution for all choices of b if and only if the algebraic
multiplicity of eigenvalue 0 coincides with the dimension of the kernel of
A. In addition, there is a non-constant straight-line solution if and only if
b 6∈ im A.
Proof. First we note that an affine ordinary differential equation ẋ = Ax+b
admits a straight-line solution for all choice of b if and only if Rn = im A +
ker A. Indeed, x(t) = ta + c is a solution of the equation if and only if
a = Ac + b and a ∈ ker A. Consequently, the existence of straight-line
solutions is equivalent to ker A ∩ im A = {0}, which is exactly the condition
for the algebraic multiplicity of eigenvalue 0 to coincide with the dimension
of the kernel of A.
 EFFECTIVE KINETICS OF CHEMICAL REACTION NETWORKS 9

To show the second statement, first let b = Ab1 + b2 , where b2 ∈ ker A

is non-zero. Then x(t) = tb2 − b1 is a non-constant straight-line solution.
Conversely, if b = Ab1 , then each straight-line solution ta + c satisfies

a = A(c + b1 ) ∈ ker A,

which implies a = 0. Therefore the straight-line solution is constant. 

Remark 4.4. If b 6∈ im A, there is no bounded solution. Indeed, the exis-

tence of eigenvalue 0 implies that there is an invariant function of the form
f (x) = w · x for the ODE ẋ = Ax, where wT is a left eigenvector of A with
eigenvalue 0. As is easily seen, we have f˙ = w · b for the case with input b.
Thus, the solutions are unbounded if w · b 6= 0.

Corollary 4.5. An affine ordinary differential equation ẋ = Ax + b admits

a straight-line solution for all choices of b if the algebraic multiplicity of 0 is
one. When b 6∈ im A, the non-constant straight-line solution is unique up to
the time-translation. In particular, the rate is uniquely determined. Further
in addition, if all eigenvalues of A has non-positive real parts, all solutions
x(t) satisfies
lim kẋ(t) − ak = 0,
t→∞

where a is the rate of the straight-line solution.

Proof. By Theorem 4.3, there exists a straight-line solution under these

assumptions. To show the uniqueness of the non-constant straight-line so-
lutions, let ta + c and ta′ + c′ be solutions with non-zero rates. Then we
have
A c′ − c = a′ − a ∈ ker A,

which implies a′ = a because we have ker A ∩ im A = {0}. As ker A is

spanned by a, we have c′ = c + ka for some k. Therefore, the non-constant
straight-line solution exists and is unique up to the time-translation.
For the last statement, we first note that im A is invariant under the
flow generated by A. By assumption, 0 is a globally asymptotically stable
equilibrium of the flow restricted to im A. Let ta + c be a straight-line
solution. Then, by direct calculations, we may show that every solution of
the equation has the form

x(t) = c + ta + etA y0
for some y0 . Then we have

kẋ(t) − ak = ketA Ay0 k,

which implies limt→∞ kẋ(t) − ak = 0. 

The rate in Corollary 4.5 can be calculated explicitly.

10 TOMOHARU SUDA

Theorem 4.6. Let A be a matrix with eigenvalue 0, which has algebraic

multiplicity one, and other eigenvalues have negative real parts. Then the
rate of the straight-line solution of ẋ = Ax + b is given by
1
adj(−A)b,
p′ (0)
where p(s) is the characteristic polynomial of A and adj(−A) is the adjugate
matrix of −A.
Proof. By the assumption, we have p′ (0) 6= 0 and there exists a polynomial
q such that p(s) = sq(s), where q(0) = p′ (0). Let x(t) be a solution with
initial value x0 . By Corollary 4.5, ẋ(t) is bounded. Therefore we have
kx(t)k ≤ M t + C for some positive constants M and C. This enables us
to consider the Laplace transform X(s) of x(t). Now we use the final value
theorem to calculate the rate a = limt→∞ ẋ(t).
By taking the Laplace transform, we obtain
b
(sI − A)X(s) = x0 + .
s
Therefore we have
 
b
p(s)X(s) = adj (sI − A) x0 + .
s
Noting that A adj (sI − A) = −p(s)I + s adj (sI − A), we obtain
   
b p(s)
s AX(s) + =A X(s) + b
s q(s)
 
1 b
= adj (sI − A) x0 + +b
q(s) s
 
1 b
= (−p(s)I + s adj (sI − A)) x0 + +b
q(s) s
1
= adj (sI − A))(sx0 + b) − sx0 .
q(s)
Therefore we have
1
lim ẋ(t) = adj(−A)b.
t→∞ p′ (0)

A graph-theoretical sufficient condition is given as follows.
Theorem 4.7. Let the weighted digraph associated with matrix A satisfy
the following conditions:
(1) The loop of each vertex has a negative weight.
(2) Other than loops, each edge has a non-negative weight.
(3) The digraph is irreducible.
(4) The in-degree of each vertex is zero.
 EFFECTIVE KINETICS OF CHEMICAL REACTION NETWORKS 11

Then, the affine ordinary differential equation ẋ = Ax+ b admits a straight-

line solution for all choices of b. Further, all solutions x(t) satisfies

lim kẋ(t) − ak = 0,
t→∞

where a is the rate of the straight-line solution.

Proof. Under these assumptions, the matrix A is Metzler and irreducible.

Therefore, there is a non-negative irreducible matrix B and α > 0 such that
A = −αI + B. Further, it has w = (1, 1, · · · , 1) as a left eigenvector with
eigenvalue 0, which implies that the Perron-Frobenius eigenvector of B is
w. As σ(A) = σ(B) − α, we conclude that the algebraic multiplicity of
eigenvalue 0 is one for A, and all eigenvalues of A have non-positive real
parts. Therefore, Corollary 4.5 is applicable. 

4.3. Systems with the skew-product structures. If a chemical reac-

tion network has a group of species produced irreversibly, then the system
of ordinary differential equations assumes a form called skew-product struc-
ture.

Definition 4.8 (Skew-product structure). An ordinary differential equa-

tion ẋ = F(x) has the skew-product structure if there is a splitting of
variables x = (p, q) such that the equation assumes the following form:
dp
= f (p)
dt
dq
= g(p, q)
dt
It is clear that the skew-product structure is dependent on the choice of
coordinates.
The next result gives conditions for an ODE with the skew-product struc-
ture to have a non-constant straight-line solution.

Lemma 4.9. Let an ordinary equation ẋ = F(x) have the skew-product

structure given by F = (f , g).
(1) If there is a straight-line solution, then the equation ṗ = f (p) has a
straight-line solution.
(2) If there is a non-constant straight-line solution ta + c, then either
Jp f (c1 ) or Jq g(c2 ) is degenerate, where c = (c1 , c2 ) and Jx G(c)
denotes the Jacobian matrix of G with respect to x evaluated at c.
(3) If there is a globally asymptotically stable equilibrium, then the only
straight-line solution is constant.

An obvious scenario for (2) in the last lemma to happen is that g is

independent of q. In this case, the existence of straight-line solutions can
be verified easily.
12 TOMOHARU SUDA

Theorem 4.10. Let an ordinary differential equation have the following

form:
dp
= f (p)
dt
dq
= g(p)
dt
Assume that there is a globally asymptotically stable equilibrium for dp
dt =
f (p). Then, there exists a straight-line solution and every solution x(t) =
(p(t), q(t)) satisfies
lim kq̇(t) − g(p∗ )k = 0,
t→∞
dp
where p∗ is the equilibrium of dt = f (p). The straight-line solution is unique
up to the time-translation.
Proof. It is obvious that
 
p∗
x∗ (t) =
tg(p∗ )
is a straight-line solution. The second assertion follows from
kq̇(t) − g(p∗ )k = kg(p(t)) − g(p∗ )k
and the global asymptotic stability of p∗ . The uniqueness of the straight-line
solution follows from Lemma 4.9. 

Moreover, we may calculate the time for the rate to stabilize if additional
assumptions are satisfied.
Corollary 4.11. Let an ordinary differential equation have the following
form:
dp
= f (p)
dt
dq
= g(p)
dt
Assume that there is a globally asymptotically stable equilibrium for dp dt =
f (p), p∗ is uniformly exponentially stable i.e. there exist λ, γ > 0 such that

kp(t) − p∗ k ≤ γe−λ(t−t0 ) kp(t0 ) − p∗ k,

and g is globally Lipschitz with Lipschitz constant L. Then, for all solution
x(t) = (p(t), q(t)), we have
kq̇(t) − g(p∗ )k ≤ Lγe−λ(t−t0 ) kp(t0 ) − p∗ k.
A more general case where the existence of the rate can be established
is given by the following result, which can be obtained immediately from
Theorem 4.3.
 EFFECTIVE KINETICS OF CHEMICAL REACTION NETWORKS 13

Corollary 4.12. Let an ordinary differential equation have the following

form:
dp
= f (p)
dt
dq
= Aq + g(p),
dt
where A is a real matrix. Assume that there is an equilibrium for dp dt = f (p).
Then, a straight-line solution exists if the algebraic multiplicity of eigenvalue
0 coincides with the dimension of the kernel of A.

5. Skew product systems and quasi-stationary state

approximations
In this section, we consider the effective reaction rates for reaction net-
works with the skew-product structure. Such a situation is rather common.
If a kinetic reaction network (S, R, s, t, K) has a species P ∈ S as an irre-
versible product, that is, it appears only as a product, then we may write
the equation of kinetics in a skew product form
dx
= f (x)
dt
d[P ]
= g(x).
dt
When we regard a chemical reaction network as a conversion system, some
species are contained only as reactants, and others appear only as products.
Definition 5.1. Let (S, R, s, t) be a chemical reaction network.
(1) A species F ∈ S is a feeding species if F ∈ s(ρ) ∪ t(ρ) implies
F ∈ s(ρ).
(2) A species P ∈ S is a product species if P ∈ s(ρ) ∪ t(ρ) implies
P ∈ t(ρ).
(3) A species I ∈ S is an intermediate species if there are reactions
ρ and ρ′ with I ∈ s(ρ) and I ∈ t(ρ′ ) .
In the classical procedure of quasi-stationary state approximation, a sub-
network of a chemical reaction network is assumed to be in a stationary state.
This amounts to considering the equilibrium solution for some species, and
the overall reaction rate is determined in terms of them. Here, we would
like to consider such an approximation scheme for the case of networks with
product species. With this in mind, we define the quasi-stationary state
approximation as follows.
Definition 5.2. An ordinary differential equation with the skew product
structure
dp
= f (p)
dt
dq
= g(q, p)
dt
14 TOMOHARU SUDA

admits quasi-stationary state approximation (QSSA) if there is p∗ such that

f (p∗ ) = 0.
Remark 5.3. It is usual for QSSA to assume slow-fast dichotomy of time
scales [22, 8, 23]. This distinction is not apparent in the treatment here, nor
do we assume the steady state’s stability. However, we may regard p as the
fast variable in mind.
An important example is an ordinary differential equation with the skew
product structure and an additive input term k
dp
= f (p) + k
dt
dq
= g(q, p)
dt
admits QSSA if there is p∗ such that
f (p∗ ) + k = 0.
If QSSA is applicable, the average effective kinetics of P is given by g(x0 ),
where f (x0 ) = 0.
Reaction networks with additive inputs may arise in the following way.
Definition 5.4. For a kinetic reaction network (S, R, s, t, K), F ⊂ S is
additive input if
(1) Each F ∈ F is a feeding species.
(2) For all ρ ∈ R, F ∩ s(ρ) 6= ∅ implies s(ρ) ⊂ F.
In particular, F is first-order input if F ∩ s(ρ) 6= ∅ implies s(ρ) = F for
some F ∈ F.
A product species P ∈ S is a first-order output if P ∈ t(ρ) implies
t(ρ) = P .
Then, it is easy to observe that the networks with additive inputs induce
a non-autonomous equation with an additive input term if we control the
concentration of input species. Conversely, equations of reaction networks
with additive inputs can be regarded as that of a reaction network where
first-order inputs are added.
The QSSA is applicable for a rather large class of networks with first-order
inputs and outputs.
Theorem 5.5. Let (S, R, s, t, K) be a mass-action reaction network and a
species P ∈ S be first-order output. If the network satisfies the following
conditions, then QSSA is applicable for the kinetics of P when the concen-
tration of feeding species is controlled:
(1) If a species F never appears as a product, then F is a first-order
input.
(2) If a species X never appears as a reactant, then X is a first-order
output.
 EFFECTIVE KINETICS OF CHEMICAL REACTION NETWORKS 15

α
(3) If reactions of the form 0 −−i→ Fi are added and the outputs are
substituted with zero complexes, the resulting network has deficiency
zero and is weakly reversible.
Proof. By assumptions, the equation of kinetics assumes the following form:
df
= −Kf
dt
dp
= Bx
dt
dx
= S̃r(x) + Kf ,
dt
where K is diagonal and invertible. The equation of kinetics of the network
in (3) is given by
df
= −Kf + α
dt
dx
= S̃r(x) + Kf .
dt
Since this modified network is weakly reversible and deficiency zero, it ad-
mits a positive equilibrium (f0 (α), x0 (α)) for all α. As α is arbitrary and K
is invertible, f0 is arbitrary. Therefore, for all f0 , there exists x0 such that
0 = S̃r(x0 ) + Kf0 ,
which implies that the skew product system
dp
= Bx
dt
dx
= S̃r(x) + Kf ,
dt
admits the QSSA. 
Corollary 5.6. Let (S, R, s, t, K) be a mass-action reaction network, a
species P ∈ S be first-order output, and a species F ∈ S be first-order input.
If the network satisfies the following conditions, then QSSA is applicable for
the kinetics of P under the controlled concentration of feeding species:
(1) If a species F never appears as a product, then F is a first-order
input.
(2) If a species X never appears as a reactant, then X is a first-order
output.
(3) Terminal strong-linkage classes consist of first-order outputs. For the
network obtained by reversing all reactions, terminal strong-linkage
classes consist of first-order inputs.
(4) If reactions of the form 0 −−→ Fi are added and the outputs are
substituted with zero complexes, the resulting network has deficiency
zero.
16 TOMOHARU SUDA

Proof. By assumption (3), every complex C ultimately reacts to a first-order

output. By considering the reversed network, we see that every complex C is
contained in a path starting from a first-order input to a first-order output.
Therefore, the modified network in (4) is weakly reversible. 
We recall that a reaction network (S, R, s, t, K) is conservative if the
cokernel of the stoichiometric matrix contains a positive vector. For con-
servative networks, we may calculate the result of QSSA using flux balance
analysis.
Lemma 5.7. Let (S, R, s, t, K) be a kinetic reaction network and P ∈ S
is an irreversible product of the network. If the network is conservative
with conserved charge µ = (µ0 , µP ), the result of QSSA under an additive
constant input k is given by
d[P ] µ0
= k,
dt µP
provided QSSA is applicable.
Proof. By assumptions, the kinetics of the network is governed by an equa-
tion of the form
dx
= S̃r(x) + k
dt
d[P ]
= g(x),
dt
where x is the vector of the concentration of species other than P , S̃ is the
corresponding stoichiometric matrix. We note that the total stoichiometric
matrix S of the network is
 
S̃
S= ,
uT
where u is a non-negative vector. If QSSA is applicable, there is concentra-
tion profile x0 such that
S̃r(x0 ) + k = 0.
By the definition of conserved charge, we have
µT Sr(x0 ) = µ̃T S̃r(x0 ) + µP uT r(x0 ) = 0.
Noting that g(x) = uT r(x), we have
d[P ] 1 µ̃T
= − µ̃T S̃r(x0 ) = k.
dt µP µP


By specializing this result to the case where additive inputs occur natu-
rally, we obtain the following result.
 EFFECTIVE KINETICS OF CHEMICAL REACTION NETWORKS 17

Theorem 5.8. Let (S, R, s, t, K) be a kinetic reaction network and P ∈ S

is an irreversible product of the network. If F = {F1 , F2 , · · · , Fk } is a set
of first-order inputs and the network obtained by deleting F and reactions
starting from F is conservative, the result of QSSA is given by first-order
reactions:
d[P ]
= R1 [F1 ] + R2 [F2 ] + · · · + Rk [Fk ],
dt
provided QSSA is applicable. In particular, if the total mass of the network
except {F1 , F2 , · · · , Fk } is conserved, the production rate of P coincides with
the intake rate of F by the network.
Proof. In this case, we may apply the result to the network obtained by
deleting F and reactions starting from F and input Af , where A is a matrix
denoting the intake rate and f is the concentration vector of the species in
F. 
Corollary 5.9. For a reaction network satisfying the assumptions of The-
orem 5.5, the result of QSSA is given by first-order reactions.
Thus, as far as the intake rate is first order, the effective reaction at
the steady state appears to be first order regardless of detailed reaction
mechanisms.
In the setting of the theorem, nonlinear coupling between species affects
the applicability of QSSA rather than the linearity of the effective rate.
k k l
Example 4. Let us consider a network F −−1→ X, F −−2→ Y, X + Y −−→ P.
The kinetics has a skew product structure:
d[X]
= −l[X][Y ] + k1 [F ]
dt
d[Y ]
= −l[X][Y ] + k2 [F ]
dt
d[P ]
= l[X][Y ].
dt
It is easy to observe that QSSA is not applicable if k1 6= k2 . If k1 = k2 , the
effective production rate of P is k1 [F ].

6. Summary and conclusions

In this article, we have considered the effective kinetics of a chemical re-
action network defined as a conversion rate of feeding species into product
species and studied its basic properties. As is observable from the com-
parison with QSSA, approximation or calculation of the rate of a reaction
mechanism can be carried out within the framework of effective kinetics
in a unified fashion. The discussion here mainly pertained to the prob-
lem of obtaining a well-defined reaction rate under the assumption that the
concentration of the feeding species is kept constant. While we barely men-
tioned it here, it appears to be interesting to consider also the response
18 TOMOHARU SUDA

of chemical reaction networks under varying inputs using the framework of

non-autonomous dynamical systems. In particular, the existence of the rate
will be closely related to that of an attractor, and in this way, we may obtain
a clearer picture of how the output of a reaction system stabilizes.

Acknowledgements
This work was supported by Kakenhi Grant-in-Aid for Scientific Research
(22H05153, 23K19021) and RIKEN Incentive Research Grant. The author
would like to express deep gratitude to Dr. Hideshi Ooka and Prof. Ryuhei
Nakamura for their support and fruitful discussion.

References
[1] Polly Y Yu and Gheorghe Craciun. Mathematical analysis of chemical reaction sys-
tems. Israel Journal of Chemistry, Vol. 58, No. 6-7, pp. 733–741, 2018.
[2] Martin Feinberg. Foundations of Chemical Reaction Network Theory, Vol. 202.
Springer, 2019.
[3] Martin Feinberg. Chemical reaction network structure and the stability of complex
isothermal reactors—i. the deficiency zero and deficiency one theorems. Chemical
engineering science, Vol. 42, No. 10, pp. 2229–2268, 1987.
[4] Alan D McNaught, Andrew Wilkinson, et al. Compendium of chemical terminology,
Vol. 1669. Blackwell Science Oxford, 1997.
[5] K. J. Laidler. A glossary of terms used in chemical kinetics, including reaction dy-
namics (iupac recommendations 1996). Pure and Applied Chemistry, Vol. 68, No. 1,
pp. 149–192, 1996.
[6] P.L. Houston. Chemical Kinetics and Reaction Dynamics. Dover Books on Chemistry.
Dover Publications, 2012.
[7] Athel Cornish-Bowden. Fundamentals of enzyme kinetics. John Wiley & Sons, 2013.
[8] Matthias Stiefenhofer. Quasi-steady-state approximation for chemical reaction net-
works. Journal of Mathematical Biology, Vol. 36, pp. 593–609, 1998.
[9] T Turanyi, AS Tomlin, and MJ Pilling. On the error of the quasi-steady-state ap-
proximation. The Journal of Physical Chemistry, Vol. 97, No. 1, pp. 163–172, 1993.
[10] Katalin M Hangos, György Lipták, and Gábor Szederkényi. Structural reduction of
crns with linear sub-crns. IFAC-PapersOnLine, Vol. 54, No. 14, pp. 149–154, 2021.
[11] Yuji Hirono, Takashi Okada, Hiroyasu Miyazaki, and Yoshimasa Hidaka. Structural
reduction of chemical reaction networks based on topology. Physical Review Research,
Vol. 3, No. 4, p. 043123, 2021.
[12] Francesco Avanzini, Nahuel Freitas, and Massimiliano Esposito. Circuit theory for
chemical reaction networks. arXiv preprint arXiv:2210.08035, 2022.
[13] Michael Samoilov, Adam Arkin, and John Ross. Signal processing by simple chemical
systems. The Journal of Physical Chemistry A, Vol. 106, No. 43, pp. 10205–10221,
2002.
[14] Josh Moles, Peter Banda, and Christof Teuscher. Delay line as a chemical reaction
network. Parallel Processing Letters, Vol. 25, No. 01, p. 1540002, 2015.
[15] Béla Novák and John J Tyson. Design principles of biochemical oscillators. Nature
reviews Molecular cell biology, Vol. 9, No. 12, pp. 981–991, 2008.
[16] John C Baez and Blake S Pollard. A compositional framework for reaction networks.
Reviews in Mathematical Physics, Vol. 29, No. 09, p. 1750028, 2017.
[17] F. Horn. The dynamics of open reaction systems. In Proceedings of a Symposium
in Applied Mathematics of the American Mathematical Society and the Society for
 EFFECTIVE KINETICS OF CHEMICAL REACTION NETWORKS 19

Industrial and Applied Mathematics, Vol. 8 of SIAM-AMS proceedings, pp. 125–137.

American Mathematical Society, 1974.
[18] Tomás Caraballo and Xiaoying Han. Applied nonautonomous and random dynamical
systems: applied dynamical systems. Springer, 2017.
[19] Peter Kloeden and Meihua Yang. An introduction to nonautonomous dynamical sys-
tems and their attractors, Vol. 21. World Scientific, 2020.
[20] Peter E Kloeden and Martin Rasmussen. Nonautonomous dynamical systems. No.
176. American Mathematical Soc., 2011.
[21] Santiago Schnell. Validity of the michaelis–menten equation–steady-state or reactant
stationary assumption: that is the question. The FEBS journal, Vol. 281, No. 2, pp.
464–472, 2014.
[22] Lee A. Segel and Marshall Slemrod. The quasi-steady-state assumption: A case study
in perturbation. SIAM Review, Vol. 31, No. 3, pp. 446–477, 1989.
[23] Martin Wechselberger. Geometric singular perturbation theory beyond the standard
form, Vol. 6. Springer, 2020.

RIKEN Center for Sustainable Resource Science, RIKEN, Wako, Japan

Email address: tomoharu.suda@riken.jp