Non-Standard Finite Difference Schemes for

Michaelis-Menten type Reaction Diffusion Equations

Michael Chapwanya1 , Jean M-S Lubuma1∗and Ronald E Mickens2
1
Department of Mathematics and Applied Mathematics, University of Pretoria
Pretoria 0002, South Africa, (m.chapwanya@up.ac.za, jean.lubuma@up.ac.za)
2
Department of Physics, Clark Atlanta University
Atlanta GA 30314, USA, (rohrs@math.gatech.edu)

Abstract
We compare and investigate the performance of the exact scheme of the Michaelis-Menten
(M-M) ordinary differential equation with several new non-standard finite difference (NSFD)
schemes that we construct by using Mickens’ rules. Furthermore, the exact scheme of the M-M
equation is used to design several dynamically consistent NSFD schemes for related reaction-
diffusion equations, advection-reaction equations and advection-reaction-diffusion equations.
Numerical simulations that support the theory and demonstrate computationally the power
of NSFD schemes are presented.
AMS Subject Classification (2010): 65L12; 65L99; 65M06; 65M99
Keywords: nonstandard finite difference method, exact scheme, Lambert W function, Michaelis-
Menten equation, advection-reaction-diffusion equations

1 Introduction
The Michaelis-Menten (M-M) ordinary differential equation

du u
=− ,
dt 1+u
is of great interest in applications. It models the enzyme kinetics in pharmacology and was solved
in [1] and [2] with a singular perturbation technique where u is one of the leading order terms
of the expansion of the outer solution. The right-hand side of the M-M equation also arises in
predator-prey and infectious disease models as a functional response in specific extended forms
ranging from the Michaelis-Menten-type, the ratio-dependent-type, the Beddington-De Angelis-
type, to the general Holling type II considered in the Rosenzweig-Mac Arthur model (see, for
instance, [3] and [4]).
The M-M equation is simple in mathematical structure. However, its solutions cannot be
found explicitly. Thus, there has been a surge in the construction of numerical methods that
can provide useful and reliable information on this differential model. The particular type of
information and properties on which many researchers have focused are the linear stability of
the the hyperbolic critical point ũ = 0 and the positivity of the solutions. Nonstandard Finite
Difference (NSFD) schemes, which are elementary stable and preserve the positivity of the exact
solutions have been extensively investigated for general dynamical systems including the M-M
equation and the combustion model (see the books [5], [6] and [7] and the references therein).
The point of departure of the current work is the exact scheme of the M-M equation, which was
designed only recently by Mickens [8]. The exact scheme is based on a general existence theorem
in [5, p. 71] that is applicable in this case, thanks to the so-called Lambert W function which per-
mits us to write the solution of the M-M equation in a compact form [9]. Furthermore, given the
specific nature of the M-M functional response, we construct three new NSFD schemes, which are
∗ Corresponding author: jean.lubuma@up.ac.za

1
all topologically dynamically consistent in the sense of [10]. The analysis is taken one step further
by designing various NSFD schemes for the major related reaction-partial differential equations of
which the M-M model is the limit equation in the space independent case. This is done by using
Mickens’ rules [5]. One focus of the work is to provide the details of the numerical solutions for
the M-M equation by comparing and investigating the performance of the exact scheme, the exact
scheme-related methods and other NSFD schemes. The focus on detailed theoretical and compu-
tational proofs distinguishes our work from previous work in the literature, where the emphasis is
just on plugging the M-M equation into one of the standard ODE solvers. In this regard, we can
mention the relatively recent work [11] (and the references therein), where numerically integrated
solutions of the M-M equation are considered, without details.
The rest of the paper is organized as follows. In the next section, we recall the definition of
the Lambert W function, the main tool we use in this work. The M-M equation is discussed at
length in Sect. 3 where we study its exact scheme and compare the performance with three new
NSFD schemes. NSFD schemes for the M-M conservative oscillator or Hamiltonian system are
investigated in Sect. 4. Sects. 5, 6 and 7 are devoted to the analysis of NSFD schemes for the
M-M reaction-diffusion equation, the M-M advection-reaction equation and the M-M advection-
reaction-diffusion equation, respectively. Numerical simulations that illustrate the power of all our
NSFD schemes are presented in Sect. 8. Sect 9 provides concluding remarks pertaining to how the
new NSFD schemes fit in the literature and how the study can be extended.

2 The Lambert W Function

Special functions play such an important role in science and engineering that there are many
handbooks on them in the literature and most of these are user friendly (e.g. [12, 13]). In this
work, we will extensively use the Lambert W function, also called the Omega function or the
product log, given by the relation

z = W (z) exp(W (z)), (1)

as the multivalued inverse of the non-injective complex-valued function

w 7→ w exp(w).

The canonical reference for the Lambert W function is [9] where it is comprehensibly studied. In
the numerical simulations, we will use the Matlab built-in function “lambertw(*)”. Since we will be
dealing with the dynamics of systems that are defined on R+ ,we restrict ourselves to nonnegative
real arguments z ≥ 0 and thus are interested in the (principal) branch satisfying W (z) ≥ 0, which
is a single-valued function.
Often special functions are characterized by differential equations. In the case under consider-
ation, implicit differentiation of (1) shows that W satisfies the differential equation
dW W (z)
= . (2)
dz z(1 + W (z))

3 The Michaelis-Menten ODE

The Michaelis-Menten (M-M) ordinary differential equation (ODE) is
du au
=− , a > 0, b ≥ 0, (3)
dt 1 + bu
with initial condition
u(0) = u0 . (4)
Equation (3) arises in pharmacology models of the basic enzyme reaction. The theory of enzyme
kinetics and of the M-M equation, which comes from a singularly perturbed system of ODE’s, can
be found in [14]. The decay equation
du
= −au, a > 0, (5)
dt

2
is the linearized form of (3) about its unique.fixed-point ũ = 0. In what follows, it is more
instructive to look at (5) as the comparison equation associated with (3) (see [15]). Considered
with the same initial condition in (4), the equations (5) and (3) have exact solutions

u(t) = exp(−at)u0 , (6)

and
1 ( )
u(t) = W (b exp(−at)u0 exp(bu0 ) , (7)
b
respectively. The closed form expression (7), based of the Lambert W function, for the solution
of the M-M equation is established in [9]. From these exact solutions, we deduce the following
intrinsic properties of the M-M model:

Theorem 1 For a given initial value u0 ≥ 0, we have

0 ≤ u(t) ≤ u(t) ≤ u0 , ∀t ≥ 0, (8)

lim u(t) = u(t), ∀t ≥ 0, (9)

b→0
and
u(t) ↓ 0 as t → ∞, (10)
where the latter notation means that the function u(t) decreases monotonically to the hyperbolic
fixed-point 0 as t → +∞.

Remark 2 In view of the sign of the right hand side of Eq. (3), the hyperbolic fixed point ũ = 0 is
locally asymptotically stable. However, Theorem 1 says more than this local property. In particular
the differential equation (3) defines a dynamical system on the positive interval [0 ∞), with the
fixed-point ũ = 0 being globally asymptotically stable.
W (z)
The property (9) is obtained by using (1) where lim = 1. Notice that the relation
z→0 z
u(t) ≤ u(t) in (8) is an illustration of a well-known monotonicity result (see Theorem 8.XI in [16]).
Thus for 0 < u0 < ∞, all trajectories are bounded and decrease monotonically to zero and u ≡ 0
is locally asymptotically stable fixed-point.
Let us consider a sequence {tk = k∆t}k≥0 of equally-spaced time points where the parameter
∆t > 0 is the step size. We denote by uk an approximation to the solution u at the point t = tk .
The design of an exact scheme for the M-M model has been considered only recently, namely
in Mickens [8]. Setting from (7), uk = u(tk ), Mickens proposed for the M-M model, along the lines
of [5], the exact scheme that reads as follows:
1 ( )
uk+1 = W (b exp(−a∆t)uk exp(buk ) . (11)
b
In order to write (11) as a non-standard finite difference (NSFD) scheme, we observe that the exact
scheme
uk+1 = exp(−a∆t)uk , (12)
for the decay equation is equivalent to
uk+1 − uk
= −auk+1 , (13)
ϕ1 (∆t)
or
uk+1 − uk
= −auk , (14)
ϕ2 (∆t)
where
exp(a∆t) − 1 1 − exp(−a∆t)
ϕ1 (∆t) = and ϕ2 (∆t) = . (15)
a a
In view of the property (9), that must be replicated by the numerical methods, the equivalent
form ( )
uk+1 − uk W (b exp(−a∆t)uk exp(buk ) − buk
= , (16)
ϕ2 (∆t) bϕ2 (∆t)

3
which is indeed a NSFD method, is obtained for the exact scheme (11) (see [8]). With the formu-
lation (16), it is readily seen from the mean-value theorem and (2) that
( ) ( ) ( )
W (b exp(−a∆t)uk exp(buk ) − buk W (b exp(−a∆t)uk exp(buk ) − W buk exp(buk )
=
bϕ2 (∆t) bϕ2 (∆t)
−au(tk )
≈ .
1 + bu(tk )
Exact schemes that are effective in applications do not exist in many cases. Therefore, using
Mickens’ rules [5], we also consider the following NSFD schemes for the M-M equation:

uk+1 − uk auk+1
=− ; (17)
ϕ1 (∆t) 1 + buk

uk+1 − uk auk+1
=− ; (18)
ϕ1 (∆t) 1 + buk+1
uk+1 − uk auk
=− . (19)
ϕ2 (∆t) 1 + buk
By analogy with the continuous model, we denote the sequence (uk ) in (17)–(19) by (uk ) when
b = 0. Thus, each uk is a discrete solution of the decay equation (5). Despite its apparent implicit
structure, the NSFD scheme (18), where we are interested in positive discrete solutions, admits
the equivalent explicit form in (21) below.
Our aim is to investigate and compare the performance of the NSFD schemes (17), (18) and
(19) with the exact scheme (16). In particular, we are interested in determining if the properties
in Theorem 1 hold.

Theorem 3 The NSFD schemes (17)–(19) are dynamically consistent with all the properties stated
in Theorem 1. More precisely, for u0 ≥ 0 and any ∆t > 0, we have

0 ≤ uk ≤ uk ≤ u0 , ∀k ≥ 0,

lim uk = uk , ∀k ≥ 0
b→0

and
uk ↓ 0 as k → ∞.

Proof. We prove the theorem for the NSFD scheme (18), the situation being much easier for
the other two schemes (17) and (19). Equation (18), with the explicit expression of the denominator
function in (15), is quadratic in uk+1 in the sense that

b(uk+1 )2 + (ea∆t − buk )uk+1 − uk = 0. (20)

We assume that uk > 0. Then, the algebraic equation (20) has a negative and a positive roots.
The required positive root or solution of (18) is
 √
 buk − ea∆t + (buk − ea∆t )2 + 4buk
u k+1
= if b > 0 (21)
 k a∆t 2b
u e if b = 0.

From (18) and the definition of ϕ1 in (15), we have

uk
e−a∆t uk ≤ uk+1 = ≤ uk , (22)
1 + (ea∆t − 1)(1 + buk+1 )−1
which, for k = 0, yields
0 ≤ u1 = e−a∆t u0 ≤ u1 ≤ u0 .
Proceeding by mathematical induction on k, we obtain the first claim in the theorem. The other
two claims follow from (21) and the decreasing property of the sequence (uk ) in (22).

4
Remark 4 For most of the schemes studied below, we will primarily be interested in the preser-
vation of the positivity of solutions, which is the more relevant property from the practical point of
view. However, similarly to Remark 2 for the continuous case, Theorem 3 states much more than
the local property of elementary stability of all the NSFD schemes (17)–(19), which simply means
that these schemes have ũ = 0 only as a fixed-point and this fixed-point is locally asymptotically
stable as for the continuous model (see, e.g. [17] and [5]). In particular, ũ = 0 is globally asymp-
totically stable. Furthermore, since each one of the NSFD schemes (17)–(19) can be written in the
form uk+1 = G(∆t; uk ) where in each case the function G satisfies the conditions

dG(∆t; u)
G(0; u) = u and > 0,
du
we have in fact for any value of ∆t the replication of all topological dynamical properties, a concept
introduced in [10], where it is referred to as topological dynamic consistency. On the contrary, the
standard finite difference scheme

uk+1 − uk auk
=− , (23)
∆t 1 + buk
is known to be elementary unstable (see, e.g., [5]). It is clear that this standard scheme fails to
have positive solutions and to satisfy all the above properties for arbitrary values of ∆t. On the
other hand, the NSFD schemes (17)–(19) are not constructed in a random way. For instance, if
instead of (17), we consider the NSFD scheme

uk+1 − uk auk
=− ,
ϕ1 (∆t) 1 + buk+1

then, on writing it in the equivalent form similar to (20), it easy to check that there exists no
positive solution uk+1 when uk > 0.

Remark 5 While there is actually no need for all three NSFD schemes, each is, however, a valid
scheme in its own and of themselves. Nevertheless, we opted to display the three schemes for the
purposes of completeness, specifically that concerns on systematic methodologies of constructing
NSFD schemes are often raised in the literature. Among other things, the schemes show the
natural way of constructing the denominator functions ϕi (∆t).

4 The M-M Hamiltonian system

The steady-state of the reaction diffusion equations to be considered in this work is governed by
the second-order differential equation

d2 u au
2
− = 0, a > 0, b ≥ 0, (24)
dx 1 + bu
which involves the Michaelis-Menten reaction term. The trivial steady-state ũ = 0 is found to be
unstable, since the hyperbolic equilibrium (0, 0) of the equivalent autonomous system

du1
= u2
dx
du2 au1
= , (25)
dx 1 + bu1
where
u1 = u and u2 = u′ ,
is a saddle-point. Equation (24) is also equivalent to the conservation law
( )2
′ 1 du
H ≡ H[u (x), u(x)] := + K(u) = constant. (26)
2 dx

5
Here, H is the Hamiltonian of the system and
 ( )
 bu − ln(1 + bu)
 −a if b > 0
K(u) = b2 . (27)

 −au2
if b = 0
2
With xm = m∆x (m ∈ Z, ∆x > 0) being the space node points and um an approximation of u at
the point x = xm , the linear equation

d2 u
− au = 0, (28)
dx2
has the exact scheme (see [5])
um+1 − 2um + um−1
− aum = 0, (29)
ψ(∆x)2

where
2 ρ∆x √
ψ(∆x) = sinh( ), ρ = a. (30)
ρ 2
We follow the methodology in [18, 17] to consider, for (24), the NSFD scheme
( )
um+1 − 2um + um−1 b(um+1 − um−1 ) − ln(1 + bum+1 ) + ln(1 + bum−1 )
−a = 0, (31)
ψ(∆x)2 b2 (um+1 − um−1 )

where, by the mean-value theorem, we take

b(um+1 − um−1 ) − [ln(1 + bum+1 ) − ln(1 + bum−1 )] um

= if um+1 = um−1 or b = 0. (32)
b2 (um+1 − um−1 ) 1 + bum

Note that
b(um+1 − um−1 ) − ln(1 + bum+1 ) + ln(1 + bum−1 )
≥ 0,
b2 (um+1 − um−1 )
by again the mean-value theorem, since

d[bu − ln(1 + bu)] b2 u

= ≥ 0 for u ≥ 0.
du 1 + bu
The underlying point of the NSFD scheme (31) is its equivalence to a discrete analogue of the
conservation law (26). More precisely, we have the relation
( )2 ( )2
1 um+1 − um 1 um − um−1
+ K∆x (um ) = + K∆x (um−1 ), (33)
2 ψ(∆x) 2 ψ(∆x)
where K∆x (um ), with an initial guess K∆x (u0 ), is derived from the identity

K(um+1 ) − K(um−1 )
K∆x (um ) = K∆x (um−1 ) + .
2
For instance if K∆x (u0 ) = 0, we have
 m
 ∑ K(ui−1 ) − K(ui−1 ) K(um+1 ) − K(u0 )

 for m ≥ 1
 2
=
2
K∆x (um ) = i=1 (34)

 |m|
∑ K(um−1+i ) − K(um+1+i )

 if m < 0.
i=1 2

We will not consider the numerical implementation of the NSFD scheme (31) and of its equivalent
formulation (33), since this is done, for instance, in [17, 19] for similar discrete conservative systems,

6
where Mickens’ rule on the nonlocal approximation of nonlinear terms is essential. Furthermore,
the NSFD scheme (31) is equivalent to the system of difference equations
 u
1,m+1 − u1,m

 = u2,m+1
ψ(∆x) ( ) (35)
 u2,m+1 − u2,m = a b(u1,m+1 − u1,m−1 ) − [ln(1 + bu1,m+1 ) − ln(1 + bu1,m−1 )] ,

ψ(∆x) b2 (u1,m+1 − u1,m−1 )

which, as a discrete analogue of the system (25), is an elementary stable NSFD scheme because
the scheme reduces to the exact scheme (29) when b = 0. Since our primary focus is not on the
replication of the conservation law (26), we shall also consider the NSFD scheme
um+1 − 2um + um−1 um
−a = 0,
ψ(∆x)2 1 + bum

which, though not preserving this property because of the violation of Mickens’ rule on the nonlocal
approximation of nonlinear terms, corresponds to an elementary stable NSFD scheme of the form
similar to (35).

5 The M-M PDE

The reaction diffusion equation with the M-M term is

∂u au ∂2u
=− + . (36)
∂t 1 + bu ∂x2
Its comparison equation or its linearized equation about the equilibrium solution ũ = 0 is the
classical linear parabolic equation
∂u ∂2u
= −au + . (37)
∂t ∂x2
We consider the same initial function u0 (x) for the problems (36) and (37):

u(0, x) = u0 (x) = u(0, x). (38)

au au
The real-valued function u → satisfies ≤ au on R+ and has bounded derivative
1 + bu 1 + bu
a
u 7→ on R+ . On the other hand, the Sturm-Liouville problem defined by the operator
(1 + bu)2
in the right-hand side of (37), with boundary conditions ϕ(0) = ϕ(1) = 0, has eigenvalues and
associated (L2 -orthonormal) eigenfunctions

λn = −a − n2 π 2 and ϕn (x) = 2 sin(nπx), n = 1, 2, · · · ,

respectively. Thus, the standard existence, uniqueness, comparison and stability results (see for
example [20, 21]) apply to the problem (36), as specified in the next theorem.

Theorem 6 For a given bounded and continuous function u0 (x), the initial value problems (36),
(37) and (38) admit unique solutions u(t, x) and u(t, x) such that

0 ≤ u0 (x) =⇒ 0 ≤ u(t, x) ≤ u(t, x) ≤ M := sup u0 (x). (39)

x

The equilibrium solution ũ = 0 of (36) and (38) is L∞ asymptotically stable. Furthermore, if the
initial-value problem is coupled with the Dirichlet boundary condition u(t, 0) = u(t, 1) = 0 , then
the equilibrium solution ũ = 0 is L2 asymptotically stable in the precise manner that the kinetic
energy ∫
1 1
E(u(t)) := |u(t, x)|2 dx, (40)
2 0
decreases monotonically to 0 as t → ∞.

7
 Our aim is to consider for (36) various NSFD schemes and to demonstrate theoretically and
computationally that they are dynamically consistent with the property (39). The stability prop-
erties will be demonstrated computationally. The general strategy, which strictly speaking makes
sense when the so-called “lumped parameter assumption”[21] is valid, consists in suitably combin-
ing the space-independent NSFD schemes in Sect. 3 with the time-independent schemes in Sect.
4.
Firstly, we deal with the schemes (17), (19), (4) and (31) to obtain the following explicit NSFD
schemes, respectively:

m − um
uk+1 k
auk+1 uk − 2ukm + ukm−1
=− m
k
+ m+1 , (41)
ϕ1 (∆t) 1 + bum ψ(∆x)2
and
( )
m − um
uk+1 k
uk+1 b(ukm+1 − ukm−1 ) − ln(1 + bukm+1 ) + ln(1 + bukm−1 )
= −a m
ϕ1 (∆t) ukm b2 (ukm+1 − ukm−1 )
ukm+1 − 2ukm + ukm−1
+ , (42)
ψ(∆x)2

m − um
uk+1 k
aukm ukm+1 − 2ukm + ukm−1
=− + , (43)
ϕ2 (∆t) 1 + bukm ψ(∆x)2
and
( )
m − um
uk+1 k b(ukm+1 − ukm−1 ) − ln(1 + bukm+1 ) + ln(1 + bukm−1 )
= −a
ϕ2 (∆t) b2 (ukm+1 − ukm−1 )
ukm+1 − 2ukm + ukm−1
+ . (44)
ψ(∆x)2
Secondly, we are interested in combining the exact scheme (16) with (4) to obtain the explicit
NSFD scheme
( )
m − um
uk+1 k W (b exp(−a∆t)ukm exp(bukm ) − bukm uk − 2ukm + ukm−1
= + m+1 . (45)
ϕ2 (∆t) bϕ2 (∆t) ψ(∆x)2
In terms of the Lambert W -function, the following explicit NSFD scheme, which is tailored from
the scheme (42) and (45), preserves the conservation law in the stationary case:
{ [ ( )] }
m − um
uk+1 k
uk+1 1 − ln 1 + W (b exp(−a∆t)ukm+1 exp(bukm+1 ) + ln(1 + bukm−1 )
= −a k m
+ [ ( ) ]
ϕ2 (∆t) um b b W (b exp(−a∆t)ukm+1 exp(bukm+1 ) − bukm−1
ukm+1 − 2ukm + ukm−1
+ . (46)
ψ(∆x)2
Notice that both NSFD schemes (45) and (46) can, in view of the mean-value theorem, reduce to:

m − um
uk+1 k
uk+1 uk − 2ukm + ukm−1
= −a m k + m+1 . (47)
ϕ2 (∆t) 1 + bum ψ(∆x)2
As observed first by Mickens [22], the next result confirms the relevance of a functional relation
between step sizes in order for an explicit NSFD scheme to preserve the positivity and boundedness
property.
Theorem 7 Assume that ψ(∆x) is chosen such that the functional relation
ϕi (∆t) 1
2
= (48)
ψ(∆x) 2
occurs between the step sizes. Then, we have the following positivity and boundedness property for
the explicit NSFD schemes (41) and (42), (43) and (44), (45) and (46):
0 ≤ u0m ≤ M =⇒ 0 ≤ ukm ≤ ukm ≤ M, ∀k ≥ 0, ∀m ∈ Z, and S k ↓ 0 as k → ∞
Here, ukm denotes the discrete solution for the equation (37), resulting on setting b = 0 in each one
of the above-mentioned NSFD schemes and S k := supm ukm .

8
 Proof. Observe first that (48) is similar to a typical condition of Lax-Richtmyer stability of
the involved explicit finite difference schemes for the linear equation (37) (see e.g. [23]). This
condition allows us to re-write each one of the above NSFD schemes in the compact form

ukm+1 + ukm−1
uk+1
m =
2[1 + g(b)]
um+1 + ukm−1
k
uk+1
m =
2[1 + g(0)]

for some function g satisfying

g(0) ≥ g(b) ≥ 0.
This implies that (S k ) and (S k ) are nonnegative monotonic decreasing sequences such that

S k ↓ 0 as k → ∞

and by mathematical induction on k, we have

0 ≤ ukm ≤ ukm ≤ M.

Remark 8 We will not investigate the combination of (18) with (4) and (31), which leads to the
following implicit NSFD schemes that are not easy to solve in uk+1
m :

m − um
uk+1 k
auk+1 ukm+1 − 2ukm + ukm−1
=− m
+ . (49)
ϕ1 (∆t) 1 + buk+1
m ψ(∆x)2

and
( )
m+1 − um−1 ) − ln(1 + bum+1 ) + ln(1 + bum−1 )
b(uk+1 k+1 k+1 k+1
m − um
uk+1 k
= −a
m+1 − um−1 )
b2 (uk+1
ϕ1 (∆t) k+1

ukm+1 − 2ukm + ukm−1

+ . (50)
ψ(∆x)2

As far as the positivity of discrete solutions is concerned, it is possible to achieve it by consid-

ering implicit NSFD schemes in which (∆x)2 is used in place of ψ(∆x)2 and the condition (48) is
not needed. The precise result reads as follows:

Theorem 9 Assume that Dirichlet boundary conditions are prescribed for the problem (36), namely

u(t, −p) = α and u(t, p) = β,

(with p > 0, α ≥ 0 and β ≥ 0) so that a finite number of nodes xm = m∆x, m = 0, ±1, · · · , ±N ,

are considered where N ≥ 2 is an integer and ∆x = p/N. Then the positivity of the solutions of
(36) and of the comparison equation (37) is preserved by all the implicit NSFD schemes obtained
from (41), (42), (43), (44), (45) and (46) by replacing

ukm+1 − 2ukm + ukm−1 m+1 − 2um + um−1

uk+1 k+1 k+1
with . (51)
ψ(∆x)2 (∆x)2

Proof. To be specific, we focus only on the following implicit NSFD scheme obtained from
(41), although the analysis below applies to all of them:

m+1 − 2um + um−1

uk+1 k+1
m − um
k+1
uk+1 k
auk+1
=− m
+ . (52)
ϕ1 (∆t) 1 + bukm (∆x)2

Consider the column vector

[ k+1 T
]
−N +1 · · · um · · · uN −1
U k+1 = uk+1 k+1
.

9
Then the method (52) is equivalent to solving at each step the algebraic linear system

Ab U k+1 = F k (53)

where
 aϕ1 (∆t) 
1 + 2c + 1+buk
−c 0··· 0
 1−N

 −c 1 + 2c + aϕ1 (∆t)
−c · · · 0 
 1+buk 
 2−N

 0 −c 1 + 2c + aϕ1 (∆t)
··· 0 
Ab = 

1+buk 3−N 

 .. .. .. .. 
 . . . . 
 
 −c 
aϕ1 (∆t)
0 0 0··· − c 1 + 2c + 1+bu k
N −1

ϕ1 (∆t)
c=
(∆x)2
and [ ]T
F k = uk−N +1 + αc, uk−N +2 · · · , ukN −2 , ukN −1 + βc .
The tridiagonal matrix Ab is an M -matrix, being a strictly diagonally dominant matrix with
positive diagonal and non-positive off diagonal [25] . Thus

U k ≥ 0 =⇒ U k+1 = A−1
b F ≥ 0.
k

Furthermore, since Ab ≤ A0 , we have [26]

0 ≤ U 1 = A−1 −1 0
0 F ≤ Ab F = U ,
0 1

which by mathematical induction yields

0 ≤ U k ≤ U k.

6 The M-M advection reaction equation

The advection equation with a M-M reaction term is
∂u ∂u au
+d =− , d > 0. (54)
∂t ∂x 1 + bu
Considering separately the space and the time independent forms of (54), which are both of the
type (3), it makes sense to combine suitably the NSFD schemes presented in Section 3. We obtain
the numerical schemes

m − um
uk+1 k uk − ukm−1 auk+1
+d m = − m
,
ϕ1 (∆t) dϕ1 (∆x/d) 1 + bukm−1
m − um
uk+1 k uk − ukm−1 aukm−1
+d m = − ,
ϕ2 (∆t) dϕ2 (∆x/d) 1 + bukm−1
( )
m − um
uk+1 k uk − ukm−1 W (b exp(−a∆t)ukm−1 exp(bukm−1 ) − bukm−1
+d m = , (55)
ϕ2 (∆t) dϕ2 (∆x/d) bϕ2 (∆t)
which, on requiring the functional relation

ϕi (∆t) = ϕi (d−1 ∆x) i.e. ∆x = d∆t, i = 1, 2, (56)

reduce to the following NSFD schemes:

m − um−1
uk+1 k
auk+1
=− m
, (57)
ϕ1 (∆t) 1 + bukm−1

10
 m − um−1
uk+1 k
aukm−1
=− , (58)
ϕ2 (∆t) 1 + bukm−1
and ( )
m − um−1
uk+1 k W (b exp(−a∆t)ukm−1 exp(bukm−1 ) − bukm−1
= . (59)
ϕ2 (∆t) bϕ2 (∆t)
Observe that the schemes (57), (58) and (59) depend only on uk+1
m and ukm−1 , in accordance with
a general result obtained by Mickens [24] for the PDE

∂u ∂u
+d = f (u).
∂t ∂x
The reason for this is that the solutions should only move along the characteristics and, in fact
along the characteristics, the advection-reaction equation (54) becomes, for a fixed x ∈ R, the M-M
ODE:
dU aU
=− , (60)
dt 1 + bU
where
t → t, x → x + dt, U (t) := u(t, x + dt). (61)
Note that Eq. (54) is just Eq. (3) with a linear advection term. It may be compared to

∂u ∂u
+d = −au
∂t ∂x
to examine the existence, boundedness and other solution properties. Also, in view of the connec-
tion with the M-M ODE (60), the properties of the solutions of (54) along the characteristics are
similar to those stated in Theorem 1. In the next theorem, we state how some of these properties
are preserved by the NSFD schemes (57), (58) and (59) along the lines of Theorems 3 and 7.

Theorem 10 Under the relation (56), we have

0 ≤ u0m ≤ M =⇒ 0 ≤ ukm ≤ M, ∀k ≥ 0, ∀m ∈ Z, and S k ↓ 0 as k → ∞.

Proof. For a fixed space grid point xm−k , we have xm−k = xm − dtk by (56). The assumption
on the initial values in the theorem implies that 0 ≤ u0m−k ≤ M which, for the function U
associated with x = xm−k and given by (60)-(61), means that 0 ≤ U 0 ≤ M. It follows from
Theorem 3 and (61) that 0 ≤ U k ≤ U k−1 ≤ U 0 ≤ M or 0 ≤ ukm ≤ uk−1m−1 ≤ um−k ≤ M . Thus
0

0≤S ≤Sk k−1

≤ S ≤ M and S ↓ 0.
0 k

7 The M-M advection reaction-diffusion equation

The advection-diffusion equation with a M-M reaction term is

∂u ∂u au ∂2u
+d =− + (62)
∂t ∂x 1 + bu ∂x2
and the physical solutions should have the positivity property (39). In order to construct a scheme
that preserves the property (39), it seems natural to combine the approach in Sections 5 and 6.
However, this would require the imposition of conditions (48) and (56), which taken simultaneously
are incompatible.
First, let us avoid the condition (48) and keep condition (56). Using (51), one possible NSFD
scheme is
( )
m − um−1
uk+1 k W (b exp(−a∆t)ukm−1 exp(bukm−1 ) − bukm−1 uk+1 − 2uk+1
m + um−1
k+1
= + m+1 . (63)
ϕ2 (∆t) bϕ2 (∆t) (∆x)2

Analogously to Theorem 9, if boundary conditions are specified, the NSFD scheme (63) reduces
to a workable formulation similar to (53), with an M matrix such that the positivity property is
preserved.

11
 Secondly, we want to avoid both conditions (48) and (56). To this end, we follow an idea in
[27, 28]. Given xm = m∆x, we introduce

x̄m := xm − d∆t

the backtrack point of xm , which is not in general on the grid. For any xm , there exists an integer
m1 such that
xm1 −1 < x̄m ≤ xm1 .
Using the linear Lagrange interpolation at the points (xm1 −1 , ukm1 −1 ) and (xm1 , ukm1 ), let us con-
sider the following point on this line:
x̄m − xm1 k x̄m − xm1 −1 k
ūkm := um1 −1 + um1 .
−∆x ∆x

Notice that ūkm ≥ 0 if ukm1 −1 and ukm1 are nonnegative. One possible NSFD scheme for (62) is
( )
m − ūm
uk+1 k W (b exp(−a∆t)ūkm exp(būkm ) − būkm uk+1 − 2uk+1 k+1
m + um−1
= + m+1 , (64)
ϕ2 (∆t) bϕ2 (∆t) (∆x)2

which enjoys the same dynamic consistency properties as the scheme (63).
Finally, the explicit NSFD scheme
( )
m − ūm
uk+1 k W (b exp(−a∆t)ūkm exp(būkm ) − būkm uk − 2ūkm + ukm1 −1
= + m1 +1 , (65)
ϕ2 (∆t) bϕ2 (∆t) ψ(∆x)2

can be used. The left-hand sides of (64) and (65) approximate the left-hand side of (62), following
a rewriting similar to (55). In summary, we have the following result:

Theorem 11 With suitable boundary conditions, the NSFD schemes (63) and (64) preserve the
property (39). The same property is preserved by the NSFD scheme (65) under the relation (48).

8 Numerical simulations
In this section, we demonstrate computationally the theoretical results obtained in the previous
sections. The results confirm in particular that the NSFD schemes constructed by using Mickens’
classical rules are reliable and thus they can be used when exact schemes are not available. We
have purposely decided not comment too much on standard finite difference schemes, since their
poor performance has been extensively reported in the literature (see, for instance, [17], [27], [5],
[6] and [7]). In all the examples presented below, we take a = 0.1; the time step is ∆t = 2, which
is considered to be large for standard numerical schemes. In the simulation involving Lambert W
function, we use the Matlab built-in function “lambertw(*)”.
Figure 1 supports the theory in Theorem 3 and Remark 2. It is seen that the NSFD schemes
(17), (18) and (19) perform as efficiently as the exact scheme (11) or (16) regarding the preservation
of the dynamics of the system stated in Theorem 1 when b = 1 and b = 0, respectively. Among
other things, the same figure displays the global asymptotic stability of the equilibrium solution
ũ = 0 given in Remark 4. The contrast with the standard scheme (23) and the Matlab ODE45
solver is visualized in Figure 2. The apparent good results by the ODE45 solver are at the high
cost of very small time steps ∆t, which is a serious concern when one is interested in the long term
behavior of the solution, or for stiff/chaotic problems, [19].

12
 1
Eqn. 17, b = 1
Eqn. 17, b = 0
0.8 Eqn. 11, b = 1

0.6
u

0.4

0.2

0
0 10 20 30 40 50 60
t
1
Eqn. 18, b = 1
Eqn. 18, b = 0
0.8 Eqn. 11, b = 1

0.6
u

0.4

0.2

0
0 10 20 30 40 50 60
t
1
Eqn. 19, b = 1
Eqn. 19, b = 0
0.8 Eqn. 11, b = 1

0.6
u

0.4

0.2

0
0 10 20 30 40 50 60
t

Figure 1: NSFD schemes (17), (18) and (19) versus the exact scheme (11)

13
 1 1
Eqn. 23, b = 1 ode45, b = 1
Eqn. 23, b = 1 ode45, b = 0
0.8 Exact, b = 1
Eqn. 11, b = 1 0.8
Exact, b = 0
0.6
0.6
0.4 u
u
0.2 0.4

0
0.2
−0.2

−0.4 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
t t

(a). (b).

Figure 2: Standard scheme (23) in (a) and ODE45 Solver in (b).

14
 Figures 3, 6, 7 and 8 illustrate Theorem 7 for the explicit NSFD schemes (41), (42), (43) and
−x2
(44), respectively. The initial function is u0 (x)
√ =e ∈ R. The value of the space size ∆x is
, x√
obtained according to (48), i.e. ψ(∆x) = 2ϕi (∆t) = 2ϕi (2) , where i = 1, 2, while we take
b = 1 and b = 0. It is seen on the pictures that the NSFD schemes compare nicely with the exact
scheme-related method (45), as far as the replication of the properties in Theorem 6 is concerned.
This includes the stated L∞ -stability of the equilibrium solution ũ = 0, while the decay to 0 of the
kinetic energy is displayed on Figure 5 (a). The poor performance of the standard finite difference
method is shown on Figure 4, which exhibits negative solutions, and Figure 5 (b). Here and after,
the associated Figure (a) shows, in each case, a side-view profile corresponding to Figure (b) taken
at x = 0. The excellence performance of the NSFD schemes is further confirmed by the values of
the solutions tabulated in Tables 1, 2, 3 and 4.

1
Eqn. 41, b = 1
Eqn. 41, b = 0
0.8 Eqn. 45, b = 1
1

0.6
u u0.5

0.4

0.2 0
1
60
0.5 40
0 20
0 10 20 30 40 50 60 0 0
t x t

(a). (b).

Figure 3: NSFD scheme (41) versus the exact scheme-related method (45).

t x u u Exact
0 0 1.0000 1.0000 1.0000
10 0.1667 0.3611 0.5475 0.5537
20 0.3333 0.1264 0.2459 0.2504
30 0.5000 0.0430 0.0911 0.0929
40 0.6667 0.0142 0.0295 0.0300
50 0.8333 0.0045 0.0088 0.0089
60 1.0000 0.0014 0.0025 0.0010

Table 1: Comparative results corresponding to Figure 3

1
Std. scheme, b=1
Std. scheme, b=0
0.8 Eqn. 45, b=1
1
0.6
0.5
u
0.4 u
0
0.2
−0.5
0 2
60
0 40
20
0 10 20 30 40 50 60 −2 0
t x t

(a). (b).

Figure 4: Standard scheme corresponding to (43) for ϕi (∆t) = ∆t = 2, ψ(∆x) = ∆x = 2.

15
 0.35 0.5

0.3
0.4
0.25
Kinetic energy
Kinetic energy

0.3
0.2

0.15 0.2

0.1
0.1
0.05

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
t t

(a). (b).

Figure 5: Kinetic energy profiles for: (a) NSFD scheme (43) and, (b) the standard finite difference
scheme related to (43) for ϕi (∆t) = ∆t, ψ(∆x) = ∆x.

16
 1
Eqn. 42, b = 1
Eqn. 42, b = 0
0.8 Eqn. 45, b = 1
1

0.6
u u0.5

0.4

0.2 0
1
60
0.5 40
0 20
0 10 20 30 40 50 60 0 0
t x t

(a). (b).

Figure 6: NSFD scheme (42) versus the exact scheme-related method (45).

t x u u Exact
0 0 1.0000 1.0000 1.0000
10 0.1667 0.3611 0.5477 0.5537
20 0.3333 0.1264 0.2461 0.2504
30 0.5000 0.0430 0.0912 0.0929
40 0.6667 0.0142 0.0295 0.0300
50 0.8333 0.0045 0.0088 0.0089
60 1.0000 0.0014 0.0025 0.0010

Table 2: Comparative results corresponding to Figure 6

17
 1
Eqn. 43, b = 1
Eqn. 43, b = 0
0.8 Eqn. 45, b = 1
1

0.6
u u0.5

0.4

0.2 0
1
60
0.5 40
0 20
0 10 20 30 40 50 60 0 0
t x t

(a). (b).

Figure 7: NSFD scheme (43) versus the exact scheme-related method (45).

t x u u Exact
0 0 1.0000 1.0000 1.0000
10 0.1667 0.3608 0.5777 0.5537
20 0.3333 0.1263 0.2741 0.2504
30 0.5000 0.0429 0.1052 0.0929
40 0.6667 0.0142 0.0343 0.0300
50 0.8333 0.0045 0.0101 0.0089
60 1.0000 0.0010 0.0015 0.0010

Table 3: Comparative results corresponding to Figure 7

18
 1
Eqn. 44, b = 1
Eqn. 44, b = 0
0.8 Eqn. 45, b = 1
1

0.6
u u0.5

0.4

0.2 0
1
60
0.5 40
0 20
0 10 20 30 40 50 60 0 0
t x t

(a). (b).

Figure 8: NSFD scheme (44) versus the exact scheme-related method (45).

t x u u Exact
0 0 1.0000 1.0000 1.0000
10 0.1667 0.3608 0.5778 0.5537
20 0.3333 0.1263 0.2743 0.2504
30 0.5000 0.0429 0.1053 0.0929
40 0.6667 0.0142 0.0343 0.0300
50 0.8333 0.0045 0.0102 0.0089
60 1.0000 0.0010 0.0029 0.0010

Table 4: Comparative results corresponding to Figure 8

19
 Mike: we should add two figures: one on the decay of the kinetic energy and another one on
standard finite difference method.
Figure 9 illustrates Theorem 9 for the preservation of the positivity of solutions when the
implicit NSFD scheme (52) is used. From now on, the standard scheme is no longer plotted due
to its poor performance mentioned and observed above. We take p = 1 and consider homogenous
Dirichlet boundary conditions i.e. α = β = 0. The initial value is u0 (x) = | sin πx| for x ∈ [−1, 1]
and u0 (x) = 0 for |x| > 1. The space step size is ∆x = 0.1, which varies independently from
∆t = 2 in the sense that the requirement (48) is not met. Apart from the positivity of the discrete
solutions, Table 5 confirms the expected comparison between the solutions ukm for b = 1 and ukm
for b = 0.

1
Eqn. 52, b = 1
Eqn. 52, b = 0
0.8
1

0.6
u u0.5

0.4

0.2 0
1
60
0.5 40
0 20
0 10 20 30 40 50 60 0 0
t x t

(a). (b).

Figure 9: Positivity of the implicit NSFD scheme(52).

t x u u
0 0 3.090 × 10−1 3.090 × 10−1
10 0.1667 1.255 × 10−6 1.265 × 10−6
20 0.3333 6.344 × 10−12 6.398 × 10−12
30 0.5000 2.799 × 10−16 2.822 × 10−17
40 0.6667 1.135 × 10−22 1.144 × 10−22
50 0.8333 4.321 × 10−28 4.357 × 10−28
60 1.0000 1.559 × 10−33 1.572 × 10−33

Table 5: Comparative results corresponding to Figure 9

20
 Figures 10 and 11 illustrate the positivity and boundedness property in Theorem 10 for the
explicit NSFD schemes (57) and (58) with initial value u0 (x) = e−x , x ∈ R. With d = 0.1 and
2

b = 1, we take the space step size in accordance with (56), i.e. ∆x = 0.2. The results compare
nicely with the exact scheme-related method (59), as also shown on Tables 6 and 7.

1
Eqn. 57, b = 1
Eqn. 59, b = 1
0.8
1

0.6
u u0.5

0.4

0.2 0
1
60
0.5 40
0 20
0 10 20 30 40 50 60 0 0
t x t

(a). (b).

Figure 10: Positivity and boundedness of the NSFD scheme (57) versus the exact scheme-related
method (59).

t x u Exact
0 0 1.0000 1.0000
10 0.1667 0.5328 0.5389
20 0.3333 0.2565 0.2616
30 0.5000 0.1091 0.1118
40 0.6667 0.0429 0.0440
50 0.8333 0.0162 0.0166
60 1.0000 0.0060 0.0062

Table 6: Comparative results corresponding to Figure 10

21
 1
Eqn. 58, b = 1
Eqn. 59, b = 1
0.8
1

0.6
u u0.5

0.4

0.2 0
1
60
0.5 40
0 20
0 10 20 30 40 50 60 0 0
t x t

(a). (b).

Figure 11: Positivity and boundedness of the NSFD scheme (58) versus the exact scheme-related
method (59).

t x u Exact
0 0 1.0000 1.0000
10 0.1667 0.5623 0.5389
20 0.3333 0.2865 0.2616
30 0.5000 0.1275 0.1118
40 0.6667 0.0514 0.0440
50 0.8333 0.0197 0.0166
60 1.0000 0.0073 0.0062

Table 7: Comparative results corresponding to Figure 11

22
 Finally, Figure 12 illustrates the positivity of solutions in Theorem 11 for the scheme (65) for
the initial√condition u0 (x) = e−x , b = 1, d = 0.1 and ∆x obtained according to (48) i.e.
2

ψ(∆x) = 2ϕ2 (2).

1
Eqn. 65

0.8
1

0.6
u u0.5

0.4

0.2 0
1
60
0.5 40
0 20
0 10 20 30 40 50 60 0 0
t x t

(a). (b).

Figure 12: Positivity of the NSFD scheme (65).

23
9 Conclusion
This paper is motivated by the exact scheme of the Michaelis-Menten ordinary differential equation
designed recently in [8]. Using Mickens’ rules [5] and the specific form of the right-hand side of
the M-M equation, we introduced several new NSFD schemes for the M-M equation as well as for
associated reaction-diffusion-type equations. We demonstrated theoretically and computationally
that the new NSFD schemes are dynamically consistent with the continuous model. They all
perform as efficiently as the exact scheme and exact-related methods, confirming thus the power
of the nonstandard approach.
Our future interest is on the application of the results of this study to the design of NSFD
schemes for epidemiological models where the contact between the susceptible individuals S(t),
and the infected individuals I(t), in the total population N (t), is expressed by a nonlinear term of
S(t)I(t)
the form , which is similar to that of the M-M equation (see [29]). Preliminary work in
N (t)
this direction is done in [30]. Furthermore, we are investigating how the study could be extended
du
to the combustion model = u2 (1 − u) whose exact scheme can be deduced from [9] as
dt
{ [ ( )]}−1
uk+1 = 1 + W (uk − 1)−1 exp (uk − 1)−1 − ∆t .

Acknowledgments
The work was initiated when JM-SL was on sabbatical leave at Clark Atlanta University in Febru-
ary 2009. JM-SL is grateful to the South African National Research Foundation for funding the
visit as well as to R.E. Mickens and researchers in the Physics Department for making his visit the
most pleasant and productive. We are also sincerely grateful to the anonymous reviewers whose
comments have significantly improved this work.

References
[1] R.E. O’Malley, Jr., Singular perturbation methods for ordinary differential equations, Springer-
Verlag Applied Mathematical Sciences Series, New york,1991.

[2] L.A. Segel and M. Slemrod, The quasi-steady-state assumption: a case study in perturbation,
SIAM Review, 31 (1989), 446–477.

[3] D.T. Dimitrov and H.V. Kojouharov, Positive and elementary stable nonstandard numerical
methods with applications to predator-prey models, Journal of Computational and Applied
Mathematics, 189 (2006), 98–108.

[4] D.T. Dimitrov and H.V. Kojouharov, Nonstandard numerical methods for a class of predator-
prey models with predator interference, Proceedings of the 6th Mississippi State-UBA Confer-
ence on Differential Equations and Computational Simulations, 67-75, Electronic Journal of
Differential Equations Conference, 15, Southwest Texas State University, San Marco, 2007.

[5] R.E. Mickens, Nonstandard finite difference models of differential equations, World Scientific,
Singapore, 1994.

[6] R.E. Mickens (Ed.), Applications of nonstandard finite difference schemes, World Scientific,
Singapore, 2000.

[7] R.E. Mickens (Ed.), Advances in the applications of nonstandard finite difference schemes,
World Scientific, Singapore, 2005

[8] R.E. Mickens, An exact discretization of the Michaelis-Menten type equations with applica-
tions, Journal of Biological Dynamics, In press.

[9] R.M. Corless, G.H. Gonnet, D.E.G. Hare, D.J. Jeffrey and D.E. Knuth, On the Lambert W
function, Advances in Computational Mathematics, 5 (1996), 329–359.

24
[10] R. Anguelov, J. M.-S. Lubuma, M. Shillor, Topological dynamic consistency of nonstandard
finite difference schemes for dynamical systems, Journal of Difference Equations and Applica-
tions, 17 (2011), 1769–1791.
[11] S. Schnell and P.K. Maini, Enzyme kinetics at high enzyme concentration, Bulletin of Math-
ematical Biology, 62 (2000), 483–499.
[12] M. Abramowitz and I.A. Stegun (editors), Handbook of mathematical functions with formulas,
graphs and mathematical tables, Dover Publications, New York, 9th edition, 1972.
[13] W.H. Beyer (ed), CRC standard mathematical tables, 28th edition, 1987.
[14] J.D. Murray, Mathematical biology I: an introduction, Springer, New York, 2002.
[15] R.K. Miller, Nonlinear Volterra integral equations, Mathematics Lecture Note Series, W.A.
Benjamin, Inc., Philippines, 1971.
[16] W. Walter, Differential and integral inequalities, Springer-Verlag, Berlin, 1970.
[17] R. Anguelov and J.M-S. Lubuma, Contributions to the mathematics of the nonstandard finite
difference method and applications, Numerical Methods for Partial Differential Equations, 17
(2001), 518-543.
[18] R. Anguelov, P. Kama and J.M.-S. Lubuma, On non-standard finite difference models of
reaction-diffusion equations, Journal of Computational and Applied Mathematics, 175 (2005)
11-29.
[19] Y. Dumont and J.M.-S. Lubuma, Non-standard finite difference methods for vibro-impact
problems, Proceedings of the Royal Society of London Series A: Mathematical, Physical and
Engineering Sciences, 461A (2005), 1927-1950.
[20] J.D. Logan, Nonlinear differential equations, Wiley-Interscience, New York, 1994.
[21] J. Smoller, Shock waves and reaction diffusion equations, Springer-Verlag, 1983.
[22] R.E. Mickens, Relation between the time and space step-sizes in nonstandard finite-difference
schemes for the Fisher equation, Numerical Methods for Partial Differential Equations 13
(1997), 51-55..
[23] K.W. Morton and D.F. Mayers, Numerical solutions of partial differential equations, Cam-
bridge University Press, London, 1994.
[24] R. E. Mickens, Nonstandard finite difference scheme for scalar advection-reaction partial dif-
ferential equations, Neural, Parallel, and Scientific Computations, 12 (2004), 163 - 174.
[25] R.S. Varga, Matrix iterative analysis, Prentice-Hall, Englewood Cliffs, NJ, 1962.
[26] A. Berman and R. J. Plemmons, Nonnegative matrices in the mathematical sciences, Aca-
demic Press, New York, 1979.
[27] H.V. Kojouharov and B.M. Chen, Nonstandard methods for advection-diffusion-reaction equa-
tions, In: R.E. Mickens (Editor), Applications of nonstandard finite difference schemes, World
Scientific, Singapore, 2000, pp. 55-108.
[28] J.M-S Lubuma and K.C. Patidar, Solving singularly perturbed advection reaction equation
via non-standard finite difference methods, Mathematical Methods in the Applied Sciences, 30
(2007), 1627-1637.
[29] S.M. Garba, A.B Gumel and J.M.-S Lubuma, Dynamically-consistent nonstandard finite dif-
ference method for an epidemic model, Mathematical and Computer Modeling, 53 (2011),
131-150 (doi:10.1016/j.mcm.2010.07.026).
[30] M. Chapwanya, J.M.-S Lubuma and R.E. Mickens, From Enzyme Kinetics to Epi-
demiological Models with Michaelis-Menten Contact Rate: Design of Nonstandard Fi-
nite Difference Schemes, Computers and Mathematics with Applications, 2012, in press,
(doi:10.1016/j.camwa.2011.12.058).

25