Proceedings of the 15th International Conference

on Computational and Mathematical Methods

in Science and Engineering, CMMSE 2015
3–7July, 2015.

A convergent and dynamically consistent finite-difference

method to approximate the positive and bounded solutions
of the classical Burgers-Fisher equation

Ana E. González1 and Jorge E. Macı́as-Dı́az1

1 Departamento de Matemáticas y Fı́sica, Universidad Autónoma de Aguascalientes

emails: [email protected], [email protected]

Abstract

In this work, we investigate a spatially two-dimensional advection-diffusion-reaction

equation that generalizes the Burgers’ and the Fisher’s equations both analytically and
numerically. The partial differential equation of interest is a nonlinear model for which
the existence and the uniqueness of positive and bounded solutions are analytically
established here. At the same time, we propose an exact finite-difference discretization
for the Burgers-Fisher model of interest and show that, as the continuous counterpart,
the method proposed is capable of preserving the positivity and the boundedness of
the numerical approximations, as well as the temporal and spatial monotonicity of the
discrete initial-boundary conditions. Moreover, we prove that the method is convergent
with first order in time and second order in space.
Key words: Burgers-Fisher equation, existence and uniqueness theorem, traveling-
wave solutions, convergence analysis, exact numerical method, positivity and bounded-
ness, nonlinear finite-difference scheme
MSC 2000: 65Q10, 65M06, 65Q30, 35K55

1 Continuous model
1.1 Burgers-Fisher equation
Throughout this work, we let α and T be real numbers with T > 0. We will use the
nomenclature R+ to represent the set of nonnegative real numbers. Throughout, we will
consider an open and connected domain Ω ⊆ R2 , and suppose that u is a real function

c
CMMSE Page 584 of 1268 ISBN: 978-84-617-2230-3
 A convergent and exact numerical method

defined on Ω × [0, T ], which satisfies the nonlinear (2 + 1)-dimensional partial differential

equation  
∂u ∂u ∂u
+ αu + − ∆u − f (u) = 0, (1)
∂t ∂x ∂y
for every (x, t) ∈ Ω × (0, T ). The parameter α represents the advection/convection coeffi-
cient, and we consider a reaction term of the form

f (u) = u(1 − u). (2)

This model is a generalized form of both the Burgers’ and the Fisher’s equations, and we
will refer to it simply as the Burgers-Fisher equation. The first step in our investigation is
to establish the existence and uniqueness of positive and bounded solutions of this equation.
Let r be a positive integer, and suppose that ' · 'ℓ is any vector norm in R3 . The real
function F = F (x, t, z1 , . . . , zr ) defined for every (x, t, z1 , . . . , zr ) ∈ Ω × [0, T ] × Rr is locally
Hölder continuous with respect to the vector (x, t) if for every compact set B ⊆ Ω × [0, T ],
there exist constants C ∈ R+ and 0 < κ < 1 such that the inequality

|F (x, t, z1 , . . . , zr ) − F (y, s, z1 , . . . , zr )| ≤ C '(x, t) − (y, s)'κℓ (3)

holds for every (x, t), (y, s) ∈ B. The following proposition is found in [1], and it will lead to
establishing conditions under which positive and bounded solutions of the Burgers-Fisher
equation exist and are unique.
Proposition 1. Let A be an elliptic operator of order 2, and consider real functions F and
) defined on Ω × [0, T ] × R × R2 and (Ω × {0}) ∪ (∂Ω × [0, T ]), respectively. Assume that
u
the following conditions are satisfied:
1. u
) is smooth,

2. F (x, t, u, ∇u) is locally Hölder with respect to (x, t),

3. F (x, t, u, ∇u) is uniformly Lipschitz continuous in u, with respect to bounded subsets

of Ω × [0, T ] × R × R2 , and

)t = A)
4. the identity u u + F (x, 0, u
), ∇)
u) is satisfied for (x, t) ∈ ∂Ω × {0}.
Then there exists a positive real number T0 ≤ T such that the parabolic initial-boundary-
value problem
0 ∂u
= Au + F (x, t, u, ∇u), for (x, t) ∈ Ω × (0, T ),
∂t (4)
u(x, t) = u
)(x, t), for (x, t) ∈ (Ω × {0}) ∪ (∂Ω × [0, T ]),

has a unique solution in Ω × [0, T0 ].

c
CMMSE Page 585 of 1268 ISBN: 978-84-617-2230-3
A. E. González, J. E. Macı́as-Dı́az

1.2 Existence and uniqueness

The next result provides conditions under which initial-boundary-value problems on the
Burgers-Fisher equation have a unique solution. Its proof is a direct consequence of the
previous proposition, and it makes use of the assumption that the initial-boundary condition
u
) satisfies properties 1 and 2 above.
Theorem 2 (Existence and uniqueness). Consider the partial differential equation (1) on
the set Ω × (0, T ), and suppose that the initial-boundary conditions satisfy the conditions of
Proposition 1. Then there exists 0 < T0 ≤ T such that the corresponding initial-boundary-
value problem has a unique solution in Ω × [0, T0 ].
Proof. Mimicking the approach in [2], let us define F : Ω × [0, T ] × R × R2 → R by
 
1
F (x, t, u, v) = −αuv · + uf (u), (5)
1
and let B ⊆ Ω × [0, T ] be a compact set. Notice that for each (x, t), (y, s) ∈ B the identity
|F (x, t, u, ∇u) − F (y, s, u, ∇u)| = 0 is true, whence it follows that F (x, t, u, ∇u) is locally
Hölder continuous on (x, t). On the other hand, for any bounded set B ⊆ Ω × [0, T ] × R × R2
and for every (x, t, u1 , ∇u), (x, t, u2 , ∇u) ∈ B, the inequality
|F (x, t, u1 , ∇u) − F (x, t, u2 , ∇u)| = |P (u1 ) − P (u2 )| ≤ K|u1 − u2 | (6)
is satisfied. Here, K = sup(x,t,u,∇u)∈B |P ′ (u)|, and P is the quadratic polynomial in u with
variable coefficients that depend on ∇u, defined by the expression
 
2 1
P (u) = u + u − αu∇u · . (7)
1
If follows that F is uniformly Lipschitz continuous in u with respect to bounded subsets
of Ω × [0, T ] × R × R2 . Proposition 1 establishes then that there exists a positive number
T0 ≤ T such that a unique solution of (1) exists in Ω × [0, T0 ].

The properties of positivity and boundedness of solutions of (1) with reaction factor
(2) are established next.
Theorem 3 (Positivity and boundedness). Let u be a classical solution of (1) which satisfies
0 ≤ u(x, t) < 1, for each (x, t) ∈ (Ω × {0}) ∪ (∂Ω × [0, T ]). Then 0 ≤ u(x, t) < 1 holds for
each (x, t) ∈ Ω × [0, T ].
In addition to the results on the existence and the uniqueness of positive and bounded
solutions of the Burgers-Fisher equation, there are analytic solutions in exact form available
in the specialized literature. The function u : R × R → R defined by
    
1 1 α α 2
u(x, t) = + tanh x− + t (8)
2 2 4 2 α

c
CMMSE Page 586 of 1268 ISBN: 978-84-617-2230-3
 A convergent and exact numerical method

is a traveling-wave solution of (1) which is positive and bounded from above by 1. This
expression represents a monotone function in both variables x and t.

2 Discrete model
This section is devoted to describe the numerical method to approximate the solutions of
the partial differential equation (1), following a finite-difference approach. To that end, we
let a, b, c and d be real numbers satisfying a < b and c < d, and we fix a square spatial
domain Ω = [a, b] × [c, d] ⊆ R2 . Suppose that K, M and N are fixed positive integers, and
consider uniform partitions

a = x0 < x1 < . . . < xm < . . . < xM = b, m ∈ {0, 1, . . . , M }, (9)

c = y0 < y1 < . . . < yn < . . . < yN = d, n ∈ {0, 1, . . . , N }, (10)

of the intervals [a, b] and [c, d], respectively, each of them with norms denoted by ∆x and
∆y. Let T be any positive real number, and consider a (not necessarily uniform) partition

0 = t0 < t1 < . . . < tk < . . . < tK = T, k ∈ {0, 1, . . . , K}, (11)

of the temporal interval [0, T ]. For each k = 0, 1, . . . , K − 1, we define ∆tk = tk+1 − tk .

For each k ∈ {0, 1, . . . , K}, each m ∈ {0, 1, . . . , M } and each n ∈ {0, 1, . . . , N }, we use
ukm,n to represent an approximation to the exact value of u at the node xm,n = (xm , yn )
and time tk . For the sake of convenience, we let δz+ and δz− be respectively the tradi-
tional forward- and the backward-difference operators in the z direction, where z = x, y, t.
Consider the discrete operators

δt+ ukm,n
δt ukm,n = , (12)
∆tk
(δz+ + δz− )ukm,n
δz(1) ukm,n = , (13)
2∆z
(δz+ − δz− )ukm,n
δz(2) ukm,n = , (14)
(∆z)2

for each k ∈ {0, 1, . . . , K −1}, each m ∈ {1, . . . , M −1}, each n ∈ {1, . . . , N −1} and z = x, y.
Notice that (12) provides an approximation of order O(∆t) to the exact value of ut at the
point (xm,n , tk ). In turn, Equation (13) provides an approximation of order O((∆z)2 ) to
the exact value of uz at (xm,n , tk ). Finally, the operator (14) yields an estimation of order
O((∆z)2 ) to the exact values of uzz at the same grid point.

c
CMMSE Page 587 of 1268 ISBN: 978-84-617-2230-3
A. E. González, J. E. Macı́as-Dı́az

Let z = x, y. For each k ∈ {0, 1, . . . , K − 1}, each m ∈ {1, . . . , M − 1} and each

n ∈ {1, . . . , N − 1}, let us define the computational constants
∆tk
rzk = , (15)
2∆z
∆tk
Rzk = , (16)
(∆z)2
 
k
Cm,n = 1 − ∆tk + α∆tk δx(1) + δy(1) ukm,n , (17)
-  .
k
Dm,n = I + ∆tk δx(2) + δy(2) ukm,n , (18)

where I is the identity operator. With this notation at hand, the finite-difference scheme
to approximate the solution of (1) is given by
   
(1) k (1) k (2) k (2) k
δt ukm,n + αuk+1
m,n δ x u m,n + δ y u m,n − δ x u m,n + δ y u k+1
m,n − f (um,n ) = 0, (19)

for each k ∈ {0, 1, . . . , K − 1}, each m ∈ {1, . . . , M − 1} and each n ∈ {1, . . . , N − 1}. It is
easy to see from (19) that uk+1 m,n is a root of the quadratic polynomial

k
Fm,n (u) = ∆tk u2 + Cm,n
k k
u − Dm,n , (20)

for each k ∈ {0, 1, . . . , K − 1}, m ∈ {1, . . . , M − 1} and n ∈ {1, . . . , N − 1}.

Initial and boundary conditions are required for the method to be defined completely.
In simulations, we employ an exact solution u of (1) in order to supply the discrete initial
and Dirichlet boundary data. More concretely, we let

u0m,n = u(xm , yn , 0), (21)

hold for each m ∈ {0, 1, . . . , M } and each n ∈ {0, 1, . . . , N }; additionally, the following
conditions will be satisfied for each k ∈ {0, 1, . . . , K}:
! k
um,0 = u(xm , c, tk ), ukm,N = u(xm , d, tk ),
(22)
uk0,n = u(a, yn , tk ), ukM,n = u(b, yn , tk ).

3 Numerical properties
In the present section, we will establish the most important properties of our method. More
concretely, we will show that the method is a convergent technique, and provide conditions
under which our technique preserves the positivity, the boundedness and the monotonicity
of solutions. Throughout, we observe the conventions of the previous sections. In addition,
we will use the notation uk to represent the vector of numerical approximations at the time
tk ordered lexicographically.

c
CMMSE Page 588 of 1268 ISBN: 978-84-617-2230-3
 A convergent and exact numerical method

3.1 Positivity and boundedness

Let u be a real vector. In the following, we will use the nomenclature u > 0 to represent that
all the components of u are positive. Meanwhile, u > 0 will denote that all the components
of u are less than 1. If both conditions hold, we will simply write 0 ≤ u < 1.
The main result of this section establishes the existence and uniqueness of positive and
bounded solutions of (19). The proof follows after some preliminary lemmas. For the sake
of convenience, we will obviate the dependence of most of the constants on the temporal
step.
Lemma 4. Suppose that 2(Rx + Ry ) < 1. If uk > 0 for some k ∈ {0, 1, . . . , K}, then
k (0) < 0 is satisfied for each m ∈ {1, . . . , M − 1} and n ∈ {1, . . . , N − 1}.
Fm,n
k (0) = −D k .
Proof. Let m ∈ {1, . . . , M − 1} and n ∈ {1, . . . , N − 1}, and notice that Fm,n m,n
The conclusion follows from this fact together with the identity
   
k
Dm,n = (1 − 2Rx − 2Ry )ukm,n + Rx ukm+1,n + ukm−1,n + Ry ukm,n+1 + ukm,n−1 , (23)

which is positive under the assumptions of the lemma.

Lemma 5. Suppose that 2(Rx + Ry ) < 1, and that |α|rz < Rz for z = x, y. If uk < 1, then
k (1) > 0 holds for each m ∈ {1, . . . , M − 1} and n ∈ {1, . . . , N − 1}.
Fm,n
Proof. By the hypotheses, the inequalities Rz ± αrz > 0 must be satisfied for z = x, y.
Using the fact that uk < 1, we obtain that
k (1) = 1 − uk (1 − 2R − 2R ) − (R − αr )uk
Fm,n k
m,n x y x x m+1,n − (Rx + αrx )um−1,n
−(Ry − αry )ukm,n+1 − (Ry + αry )ukm,n−1 (24)
> 1 − 1 + 2Rx + 2Ry − Rx + αrx − Rx − αrx − Ry + αry − Ry − αry .
After simplification, the right-hand side of this inequality is reduced to zero.

The next proposition is the discrete analogue of Theorem 3.

Theorem 6. Let 2(Rx + Ry ) < 1, and |α|rz < Rz for z = x, y. If the discrete conditions
(21)-(22) are in (0, 1), then there exists a unique sequence (uk )∞
k=0 satisfying (19), such that
0 < uk < 1 for each k ∈ Z+ ∪ {0}.
Proof. The condition 0 < u0 < 1 is true by hypothesis. Suppose that 0 < uk < 1, and let
m ∈ {1, . . . , M − 1} and n ∈ {1, . . . , N − 1}. Using Lemmas 4 and 5, the function Fm,nk is
continuous and takes on a negative and a positive values at the left and the right endpoints
k
of [0, 1], respectively. So Fm,n has at least a root in (0, 1), and the uniqueness follows from
the fact that this function is a quadratic. We denote this unique root by ukm,n , whence
0 < uk+1 < 1 holds. The theorem follows by induction.

c
CMMSE Page 589 of 1268 ISBN: 978-84-617-2230-3
A. E. González, J. E. Macı́as-Dı́az

3.2 Monotonicity and stability

We say that two real vectors u and v of the same dimension satisfy u < v if each of the
components of the first is less than the respective component of the second. Let u0 and
v0 be initial approximations of an iterative method that satisfy u0 < v0 . We say that
the numerical technique preserves monotonicity if the successive approximations satisfy
uk < vk , for each k ∈ Z+ ∪ {0}.
The next result shows that (19) is a monotonicity-preserving method under the same
conditions of Theorem 6.

Theorem 7. Let 2(Rx + Ry ) < 1 and |α|rz < Rz for z = x, y. Let 0 < u0 < v0 < 1,
k
and suppose that 0 < ukm,0 < vm,0 < 1, 0 < ukm,N < vm,Nk < 1, 0 < uk0,n < v0,n k < 1
k k
and 0 < uM,n < vM,n < 1 are satisfied for each m ∈ {0, 1, . . . , M }, n ∈ {0, 1, . . . , N } and
k ∈ Z+ ∪ {0}. Then 0 < uk < vk < 1 holds for each and k ∈ Z+ ∪ {0}.

Proof. The proof goes by induction. Since the result is true for k = 0, we assume that
0 < uk < vk < 1 holds for an arbitrary k ∈ Z+ ∪ {0}. Theorem 6 guarantees that
0 < uk+1 < 1 and 0 < vk+1 < 1, so we only need to show that uk+1 < vk+1 . Let
m ∈ {1, . . . , M − 1} and n ∈ {1, . . . , N − 1}, and recall that uk+1 k
m,n is the unique root of Fm,n
k+1 is the unique zero in (0, 1) of the function
in (0, 1). Likewise, vm,n

Gkm,n (v) = ∆tv 2 + Em,n

k k
v − Jm,n , (25)

where
 
k
Em,n = 1 − ∆tk + α∆t δx(1) + δy(1) vm,n
k
, (26)
-  .
k
Jm,n = I + ∆tk δx(2) + δy(2) vm,n
k
. (27)

k (w) = F k (w) − Gk (w) = C k

 k
  k k

Let Hm,n m,n m,n m,n − Em,n w − Dm,n − Jm,n , for each w ∈ [0, 1].
k
It is clear that Hm,n represents a straight line. Moreover, after algebraic simplifications we
obtain
k (0) = [1 − 2R − 2R ] v k
 k
  k k

Hm,n x k y m,n − u m,n + Rx vm+1,n − um+1,n

+Rx vm−1,n k k − ukm,n+1
 k − um−1,n k
+ Ry vm,n+1 (28)
+Ry vm,n−1 − um,n−1 .

Under the induction hypotheses, the right-hand side for this identity is positive. Similarly,
k (1) = [1 − 2R − 2R ] v k
 k
  k k

Hm,n x y m,n − um,n +[Rx − αrx ] vm+1,n  k − u m+1,n 
k
+[Rx + αrx ] vm−1,n k − ukm,n+1
 k − um−1,n + [Ry − αry ] vm,n+1 (29)
+[Ry + αry ] vm,n−1 − ukm,n−1 ,

c
CMMSE Page 590 of 1268 ISBN: 978-84-617-2230-3
 A convergent and exact numerical method

which is a positive number. It follows that Hm,n k is positive on [0, 1] or, equivalently, that
k
Fm,n > Gm,n on that interval. As a consequence, the roots of these quadratic functions
satisfy uk+1 k+1
m,n < vm,n , for each m ∈ {1, . . . , M − 1} and each n ∈ {1, . . . , N − 1}. Considering
the conditions on the boundary, it follows that uk+1 < vk+1 , as desired.

The capability of preserving the monotonicity of solutions assures that the method
investigated in this work is stable under the parametric conditions of the last theorem, for
initial approximations and discrete Dirichlet data bounded in (0, 1).

3.3 Exactness and convergence

Recall from Section 2 that the approximation uk+1
m,n is one of the solution of the quadratic
k , which are given by the general formula. It is easy to check that out of
polynomial Fm,n
the two possible solutions, only one may be a number in (0, 1), namely,
1
k
−Cm,n + k )2 + 4∆t D k
(Cm,n k m,n
uk+1
m,n = . (30)
2∆tk

In view of the theorem on the existence and uniqueness of positive and bounded solutions
of (19), the expression (30) gives exactly the unique discrete solution at the node (xm , yn )
and the time tk .

Lemma 8 (Discrete Gronwall’s Inequality [3]). Let K > 1, and suppose that A, B and Ck
are nonnegative numbers for each k ∈ {0, 1, . . . , K}. If (A + B)∆t ≤ K−1 k K
2K , and if {w }k=0
k
satisfies w − w k−1 k
≤ A∆tw + B∆tw k−1 + Ck ∆t for each k = 1, . . . , K, then

   K

 
max wk  ≤ w0 + ∆t Cl e2(A+B)T . (31)
1≤k≤K
l=1

We will need the previous lemma to prove now the convergence properties of (19).
For convenience, we consider a regular partition of a temporal interval [0, T ] consisting of
K ∈ Z+ subintervals, with partition norm equal to ∆t = T /K.

Theorem 9. Suppose that 2(Rx + Ry ) < 1, |α|rz < Rz con z = x, y, and assume that
4,4,2
u ∈ Cx,y,t ([a, b] × [0, T ]) is a positive solution of (1) with 'u'∞ < 1. If the inequality
(∆x) + (∆y)2 < 6 is satisfied, then there exists a unique solution w of (19) converging to
2

u, and a constant C , ∈ R+ satisfying

 
, ∆t + (∆x)2 + (∆y)2 .
max |uk − wk | ≤ C (32)
0≤k≤K

c
CMMSE Page 591 of 1268 ISBN: 978-84-617-2230-3
A. E. González, J. E. Macı́as-Dı́az

Proof. By the hypotheses, there exists a unique solution of (19) satisfying (21)–(22). Also,
the solution u of the continuous initial-boundary-value problem satisfies the difference equa-
k . More precisely,
tion (19) with truncation error Rm,n
   
(1) (1)
δt ukm,n + αuk+1
m,n δx + δy ukm,n − δx(2) + δy(2) ukm,n
(33)
−uk+1 k+1 k
m,n (1 − um,n ) + Rm,n = 0

holds for every m ∈ {1, . . . , M − 1}, n ∈ {1, . . . , N − 1} and k ∈ {0, 1, . . . , K − 1}. We want
to show that there exists C ∈ R+ such that |Rm,n k | ≤ C(∆t + (∆x)2 + (∆y)2 ). Let C ∈ R+
1
be any common bound for the functions ∂tt u, ∂x u, ∂xxx u, ∂xxxx u, ∂y u, ∂yyy u and ∂yyyy u in
the interior of Ω, and such that the following inequalities hold therein:
    
  k   k+1 

 δt um,n − ∂t ukm,n  ≤ 12 C1 ∆t, um,n − ukm,n  ≤ C1 ∆t
    
 (1) k   (1) k 
δ u − ∂x ukm,n  ≤ 16 C1 (∆x)2 , δ u − ∂y ukm,n  ≤ 16 C1 (∆y)2 (34)

  x m,n   y m,n 
  
 δx(2) ukm,n − ∂xx ukm,n  ≤ 1 C1 (∆x)2 ,  
12 δy(2) ukm,n − ∂yy ukm,n  ≤ 12 1
C1 (∆y)2 .

Using these inequalities, we reach

      
 k+1 (1) k   (1) k k   k+1 k  k 
um,n δz um,n − ukm,n ∂z ukm,n  ≤ uk+1
m,n δz um,n − ∂z um,n  + um,n − um,n  ∂z um,n 
2
≤ 1
6 C1 (∆z) + C12 ∆t,

for z = x, y.
It follows that
        
 k   k (1) (1) (2) (2)
Rm,n  = δt um,n + αuk+1
m,n δ x + δ y u k
m,n − δ x + δ y u k
m,n − u k+1
m,n 1 − u k+1
m,n
 

−∂t ukm,n − αukm,n (∂x + ∂y ) ukm,n + (∂xx + ∂yy ) ukm,n + ukm,n 1 − ukm,n 
! <
C1 C1 8 9 C1 8 9
≤ ∆t + |α| (∆x)2 + (∆y)2 + 2C12 ∆t + (∆x)2 + (∆y)2 + 3C1 ∆t
2 6   12
7 1 |α| 8 9
= C1 + 2|α|C1 ∆t + C1 + (∆x) + (∆y)2 .
2
 2  12 6
≤ C ∆t + (∆x)2 + (∆y)2 ,

where C = max{D1 , D2 }, for the constants D1 = C1 ( 72 + 2|α|C1 ) and D2 = C1 ( 12 1

+ 16 |α|).
k k k
It only remains to establish the estimate (32). So, let us define εm,n = wm,n − um,n for each
m ∈ {1, . . . , M − 1}, each n ∈ {1, . . . , N − 1} and each k ∈ {0, 1, . . . , K − 1}. Notice then
that    
δt εkm,n + αwm,n
k+1
δx(1) + δy(1) εkm,n + αεk+1 (1)
m,n δx + δy
(1)
uk
    m,n
− δx(2) + δy(2) εkm,n − εk+1 k+1 k+1 k
m,n 1 − wm,n − um,n − Rm,n = 0.

c
CMMSE Page 592 of 1268 ISBN: 978-84-617-2230-3
 A convergent and exact numerical method

Substituting the expressions of the discrete difference operators, we obtain the following
identities for each m ∈ {1, . . . , M − 1}, n ∈ {1, . . . , N − 1} and k ∈ {0, 1, . . . , K − 1}:
 
k k k k
εk+1 k
m,n − εm,n k+1
ε m+1,n − ε m−1,n ε m,n+1 − ε m,n−1
+ αwm,n + +
∆t 2∆x 2∆y
  εkm+1,n − 2εkm,n + εkm−1,n
(1) (1)
αεk+1
m,n δ x + δ y u k
m,n − −
(∆x)2
εkm,n+1 − 2εkm,n + εkm,n−1  
k+1 k+1 k+1 k
− ε m,n 1 − w m,n − u m,n − Rm,n = 0.
(∆y)2
Multiplying both ends of each of these identities by ∆t and regrouping conveniently, we
obtain that
- . - .
k k k+1 k k+1
εk+1
m,n = εm,n [1 − 2Rx − 2Ry ] + εm+1,n Rx − αwm,n rx + εm−1,n Rx + αwm,n rx
- . - .
+εkm,n+1 Ry − αwm,n k+1
ry + εkm,n−1 Ry + αwm,n k+1
ry + ∆tRm,nk
-   .
(1) (1)
+∆tεk+1 k+1 k+1
m,n 1 − wm,n − um,n − α δx + δy ukm,n .

Let ξ k = max{|εkm,n | : m = 0, 1, . . . , M ; n = 0, 1, . . . , N } for each k ∈ {0, 1, . . . , K}. In

the following, we will use the fact that both the continuous and the discrete solutions
are bounded in (0, 1). We will take absolute value in our last identity, we will apply the
triangle inequality, and with use the fact that each term of the form |εkm,n | ≤ ξ k , for each
m ∈ {0, 1, . . . , M − 1}, each n ∈ {0, 1, . . . , N − 1} and each k ∈ {0, 1, . . . , K}. Observe also
that all the remaining factors are positivity in order to simplify most of the terms. It is
easy to see then that
   
 (1) (1) 
ξ k+1 − ξ k ≤ ∆tξ k+1 1 − wm,n k+1
− uk+1
m,n − α δx + δy ukm,n 
  (35)
+∆tC ∆t + (∆x)2 + (∆y)2 .
is satisfied for each k ∈ {0, 1, . . . , K − 1}. Observe then that the term Tm,nk in absolute value
that multiplies ξ k+1 , satisfies the inequality
  
k | ≤ |1 − w k+1 | + |uk+1 | + |α|  δ (1) + δ (1) uk 
|Tm,n m,n m,n  x y m,n 
 (36)
(∆x)2 + (∆y)2
≤ 2 + |α|C1 2 + ≤ 2 + 3|α|C1 .
6
 
From this and (35), we get ξ k+1 k
− ξ ≤ ∆tξ k+1
[2 + 3|α|C1 ] + ∆tC ∆t + (∆x)2 + (∆y)2 .
K−1
Suppose that ∆t ≤ 2K(2+3|α|C 1)
. An application of Lemma 8 yields now the estimate
  8  9
 
max ξ k  ≤ e2(2+3|α|C1 )T ξ 0 + K∆tC ∆t + (∆x)2 + (∆y)2 . (37)
1≤k≤K
Finally, observe that the exactness of the form of the initial conditions guarantees that
, =
ξ 0 = max{|ε0n | : n = 0, 1, . . . , N } = 0. The conclusion of this theorem follows with C
e 2(2+3|α|C 1 )T T C.

c
CMMSE Page 593 of 1268 ISBN: 978-84-617-2230-3
A. E. González, J. E. Macı́as-Dı́az

4 Conclusions
In this manuscript, we proposed a finite-difference scheme to approximate the solutions of
an advection-diffusion-reaction equation that generalizes the classical Burgers’ and Fisher’s
equations, The numerical methodology is a nonlinear discretization of the Burgers-Fisher
equation of interest, but it may be expressed in an exact form in order to approximate
positive solutions of the model of interest. We established in this work that the numerical
method is a positivity-, boundedness- and monotonicity-preserving technique under suitable
choices of the computational parameters. In addition to the fact that it is a convergent and
stable finite-difference scheme, these properties make our numerical technique an ideal tool
in the investigation of traveling-wave solutions of the advection-diffusion-reaction equation
of interest.
Finally, we would like to add that the properties on the preservation of the positivity
and the boundedness of numerical approximations are in perfect agreement with the ana-
lytical results derived in this work. Indeed, we have established analytically the existence
and the uniqueness of positive and bounded solutions of the model under investigation for
smooth initial-boundary data. In addition to this result, the literature reports on exact
traveling-wave solutions of the one-dimensional Burgers-Fisher equation which are posi-
tive and bounded. one of these exact solutions is employed to illustrate numerically the
properties of convergence of the method reported in this manuscript.

References
[1] A. Friedman. Partial Differential Equations of Parabolic Type. Prentice Hall, Inc., New
Jersey, first edition, 1964.

[2] V. J. Ervin, J. E. Macı́as-Dı́az, and J. Ruiz-Ramı́rez. A positive and bounded finite

element approximation of the generalized Burgers–Huxley equation. J. Math. Anal.
Appl., 424(2):1143–1160, 2015.

[3] T. Wang and B. Guo. Analysis of some finite difference schemes for two-dimensional
Ginzburg-Landau equation. Numer. Meth. Part. Diff. Eq., 27(5):1340–1363, 2011.

c
CMMSE Page 594 of 1268 ISBN: 978-84-617-2230-3